| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* predicate.c |
| 4 |
* POSTGRES predicate locking |
| 5 |
* to support full serializable transaction isolation |
| 6 |
* |
| 7 |
* |
| 8 |
* The approach taken is to implement Serializable Snapshot Isolation (SSI) |
| 9 |
* as initially described in this paper: |
| 10 |
* |
| 11 |
* Michael J. Cahill, Uwe Röhm, and Alan D. Fekete. 2008. |
| 12 |
* Serializable isolation for snapshot databases. |
| 13 |
* In SIGMOD '08: Proceedings of the 2008 ACM SIGMOD |
| 14 |
* international conference on Management of data, |
| 15 |
* pages 729-738, New York, NY, USA. ACM. |
| 16 |
* http://doi.acm.org/10.1145/1376616.1376690 |
| 17 |
* |
| 18 |
* and further elaborated in Cahill's doctoral thesis: |
| 19 |
* |
| 20 |
* Michael James Cahill. 2009. |
| 21 |
* Serializable Isolation for Snapshot Databases. |
| 22 |
* Sydney Digital Theses. |
| 23 |
* University of Sydney, School of Information Technologies. |
| 24 |
* http://hdl.handle.net/2123/5353 |
| 25 |
* |
| 26 |
* |
| 27 |
* Predicate locks for Serializable Snapshot Isolation (SSI) are SIREAD |
| 28 |
* locks, which are so different from normal locks that a distinct set of |
| 29 |
* structures is required to handle them. They are needed to detect |
| 30 |
* rw-conflicts when the read happens before the write. (When the write |
| 31 |
* occurs first, the reading transaction can check for a conflict by |
| 32 |
* examining the MVCC data.) |
| 33 |
* |
| 34 |
* (1) Besides tuples actually read, they must cover ranges of tuples |
| 35 |
* which would have been read based on the predicate. This will |
| 36 |
* require modelling the predicates through locks against database |
| 37 |
* objects such as pages, index ranges, or entire tables. |
| 38 |
* |
| 39 |
* (2) They must be kept in RAM for quick access. Because of this, it |
| 40 |
* isn't possible to always maintain tuple-level granularity -- when |
| 41 |
* the space allocated to store these approaches exhaustion, a |
| 42 |
* request for a lock may need to scan for situations where a single |
| 43 |
* transaction holds many fine-grained locks which can be coalesced |
| 44 |
* into a single coarser-grained lock. |
| 45 |
* |
| 46 |
* (3) They never block anything; they are more like flags than locks |
| 47 |
* in that regard; although they refer to database objects and are |
| 48 |
* used to identify rw-conflicts with normal write locks. |
| 49 |
* |
| 50 |
* (4) While they are associated with a transaction, they must survive |
| 51 |
* a successful COMMIT of that transaction, and remain until all |
| 52 |
* overlapping transactions complete. This even means that they |
| 53 |
* must survive termination of the transaction's process. If a |
| 54 |
* top level transaction is rolled back, however, it is immediately |
| 55 |
* flagged so that it can be ignored, and its SIREAD locks can be |
| 56 |
* released any time after that. |
| 57 |
* |
| 58 |
* (5) The only transactions which create SIREAD locks or check for |
| 59 |
* conflicts with them are serializable transactions. |
| 60 |
* |
| 61 |
* (6) When a write lock for a top level transaction is found to cover |
| 62 |
* an existing SIREAD lock for the same transaction, the SIREAD lock |
| 63 |
* can be deleted. |
| 64 |
* |
| 65 |
* (7) A write from a serializable transaction must ensure that an xact |
| 66 |
* record exists for the transaction, with the same lifespan (until |
| 67 |
* all concurrent transaction complete or the transaction is rolled |
| 68 |
* back) so that rw-dependencies to that transaction can be |
| 69 |
* detected. |
| 70 |
* |
| 71 |
* We use an optimization for read-only transactions. Under certain |
| 72 |
* circumstances, a read-only transaction's snapshot can be shown to |
| 73 |
* never have conflicts with other transactions. This is referred to |
| 74 |
* as a "safe" snapshot (and one known not to be is "unsafe"). |
| 75 |
* However, it can't be determined whether a snapshot is safe until |
| 76 |
* all concurrent read/write transactions complete. |
| 77 |
* |
| 78 |
* Once a read-only transaction is known to have a safe snapshot, it |
| 79 |
* can release its predicate locks and exempt itself from further |
| 80 |
* predicate lock tracking. READ ONLY DEFERRABLE transactions run only |
| 81 |
* on safe snapshots, waiting as necessary for one to be available. |
| 82 |
* |
| 83 |
* |
| 84 |
* Lightweight locks to manage access to the predicate locking shared |
| 85 |
* memory objects must be taken in this order, and should be released in |
| 86 |
* reverse order: |
| 87 |
* |
| 88 |
* SerializableFinishedListLock |
| 89 |
* - Protects the list of transactions which have completed but which |
| 90 |
* may yet matter because they overlap still-active transactions. |
| 91 |
* |
| 92 |
* SerializablePredicateListLock |
| 93 |
* - Protects the linked list of locks held by a transaction. Note |
| 94 |
* that the locks themselves are also covered by the partition |
| 95 |
* locks of their respective lock targets; this lock only affects |
| 96 |
* the linked list connecting the locks related to a transaction. |
| 97 |
* - All transactions share this single lock (with no partitioning). |
| 98 |
* - There is never a need for a process other than the one running |
| 99 |
* an active transaction to walk the list of locks held by that |
| 100 |
* transaction, except parallel query workers sharing the leader's |
| 101 |
* transaction. In the parallel case, an extra per-sxact lock is |
| 102 |
* taken; see below. |
| 103 |
* - It is relatively infrequent that another process needs to |
| 104 |
* modify the list for a transaction, but it does happen for such |
| 105 |
* things as index page splits for pages with predicate locks and |
| 106 |
* freeing of predicate locked pages by a vacuum process. When |
| 107 |
* removing a lock in such cases, the lock itself contains the |
| 108 |
* pointers needed to remove it from the list. When adding a |
| 109 |
* lock in such cases, the lock can be added using the anchor in |
| 110 |
* the transaction structure. Neither requires walking the list. |
| 111 |
* - Cleaning up the list for a terminated transaction is sometimes |
| 112 |
* not done on a retail basis, in which case no lock is required. |
| 113 |
* - Due to the above, a process accessing its active transaction's |
| 114 |
* list always uses a shared lock, regardless of whether it is |
| 115 |
* walking or maintaining the list. This improves concurrency |
| 116 |
* for the common access patterns. |
| 117 |
* - A process which needs to alter the list of a transaction other |
| 118 |
* than its own active transaction must acquire an exclusive |
| 119 |
* lock. |
| 120 |
* |
| 121 |
* SERIALIZABLEXACT's member 'perXactPredicateListLock' |
| 122 |
* - Protects the linked list of predicate locks held by a transaction. |
| 123 |
* Only needed for parallel mode, where multiple backends share the |
| 124 |
* same SERIALIZABLEXACT object. Not needed if |
| 125 |
* SerializablePredicateListLock is held exclusively. |
| 126 |
* |
| 127 |
* PredicateLockHashPartitionLock(hashcode) |
| 128 |
* - The same lock protects a target, all locks on that target, and |
| 129 |
* the linked list of locks on the target. |
| 130 |
* - When more than one is needed, acquire in ascending address order. |
| 131 |
* - When all are needed (rare), acquire in ascending index order with |
| 132 |
* PredicateLockHashPartitionLockByIndex(index). |
| 133 |
* |
| 134 |
* SerializableXactHashLock |
| 135 |
* - Protects both PredXact and SerializableXidHash. |
| 136 |
* |
| 137 |
* SerialControlLock |
| 138 |
* - Protects SerialControlData members |
| 139 |
* |
| 140 |
* SLRU per-bank locks |
| 141 |
* - Protects SerialSlruCtl |
| 142 |
* |
| 143 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 144 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 145 |
* |
| 146 |
* |
| 147 |
* IDENTIFICATION |
| 148 |
* src/backend/storage/lmgr/predicate.c |
| 149 |
* |
| 150 |
*------------------------------------------------------------------------- |
| 151 |
*/ |
| 152 |
/* |
| 153 |
* INTERFACE ROUTINES |
| 154 |
* |
| 155 |
* housekeeping for setting up shared memory predicate lock structures |
| 156 |
* PredicateLockShmemInit(void) |
| 157 |
* PredicateLockShmemSize(void) |
| 158 |
* |
| 159 |
* predicate lock reporting |
| 160 |
* GetPredicateLockStatusData(void) |
| 161 |
* PageIsPredicateLocked(Relation relation, BlockNumber blkno) |
| 162 |
* |
| 163 |
* predicate lock maintenance |
| 164 |
* GetSerializableTransactionSnapshot(Snapshot snapshot) |
| 165 |
* SetSerializableTransactionSnapshot(Snapshot snapshot, |
| 166 |
* VirtualTransactionId *sourcevxid) |
| 167 |
* RegisterPredicateLockingXid(void) |
| 168 |
* PredicateLockRelation(Relation relation, Snapshot snapshot) |
| 169 |
* PredicateLockPage(Relation relation, BlockNumber blkno, |
| 170 |
* Snapshot snapshot) |
| 171 |
* PredicateLockTID(Relation relation, ItemPointer tid, Snapshot snapshot, |
| 172 |
* TransactionId tuple_xid) |
| 173 |
* PredicateLockPageSplit(Relation relation, BlockNumber oldblkno, |
| 174 |
* BlockNumber newblkno) |
| 175 |
* PredicateLockPageCombine(Relation relation, BlockNumber oldblkno, |
| 176 |
* BlockNumber newblkno) |
| 177 |
* TransferPredicateLocksToHeapRelation(Relation relation) |
| 178 |
* ReleasePredicateLocks(bool isCommit, bool isReadOnlySafe) |
| 179 |
* |
| 180 |
* conflict detection (may also trigger rollback) |
| 181 |
* CheckForSerializableConflictOut(Relation relation, TransactionId xid, |
| 182 |
* Snapshot snapshot) |
| 183 |
* CheckForSerializableConflictIn(Relation relation, ItemPointer tid, |
| 184 |
* BlockNumber blkno) |
| 185 |
* CheckTableForSerializableConflictIn(Relation relation) |
| 186 |
* |
| 187 |
* final rollback checking |
| 188 |
* PreCommit_CheckForSerializationFailure(void) |
| 189 |
* |
| 190 |
* two-phase commit support |
| 191 |
* AtPrepare_PredicateLocks(void); |
| 192 |
* PostPrepare_PredicateLocks(TransactionId xid); |
| 193 |
* PredicateLockTwoPhaseFinish(TransactionId xid, bool isCommit); |
| 194 |
* predicatelock_twophase_recover(TransactionId xid, uint16 info, |
| 195 |
* void *recdata, uint32 len); |
| 196 |
*/ |
| 197 |
|
| 198 |
#include "postgres.h" |
| 199 |
|
| 200 |
#include "access/parallel.h" |
| 201 |
#include "access/slru.h" |
| 202 |
#include "access/transam.h" |
| 203 |
#include "access/twophase.h" |
| 204 |
#include "access/twophase_rmgr.h" |
| 205 |
#include "access/xact.h" |
| 206 |
#include "access/xlog.h" |
| 207 |
#include "miscadmin.h" |
| 208 |
#include "pgstat.h" |
| 209 |
#include "port/pg_lfind.h" |
| 210 |
#include "storage/predicate.h" |
| 211 |
#include "storage/predicate_internals.h" |
| 212 |
#include "storage/proc.h" |
| 213 |
#include "storage/procarray.h" |
| 214 |
#include "utils/guc_hooks.h" |
| 215 |
#include "utils/rel.h" |
| 216 |
#include "utils/snapmgr.h" |
| 217 |
|
| 218 |
/* Uncomment the next line to test the graceful degradation code. */ |
| 219 |
/* #define TEST_SUMMARIZE_SERIAL */ |
| 220 |
|
| 221 |
/* |
| 222 |
* Test the most selective fields first, for performance. |
| 223 |
* |
| 224 |
* a is covered by b if all of the following hold: |
| 225 |
* 1) a.database = b.database |
| 226 |
* 2) a.relation = b.relation |
| 227 |
* 3) b.offset is invalid (b is page-granularity or higher) |
| 228 |
* 4) either of the following: |
| 229 |
* 4a) a.offset is valid (a is tuple-granularity) and a.page = b.page |
| 230 |
* or 4b) a.offset is invalid and b.page is invalid (a is |
| 231 |
* page-granularity and b is relation-granularity |
| 232 |
*/ |
| 233 |
#define TargetTagIsCoveredBy(covered_target, covering_target) \ |
| 234 |
((GET_PREDICATELOCKTARGETTAG_RELATION(covered_target) == /* (2) */ \ |
| 235 |
GET_PREDICATELOCKTARGETTAG_RELATION(covering_target)) \ |
| 236 |
&& (GET_PREDICATELOCKTARGETTAG_OFFSET(covering_target) == \ |
| 237 |
InvalidOffsetNumber) /* (3) */ \ |
| 238 |
&& (((GET_PREDICATELOCKTARGETTAG_OFFSET(covered_target) != \ |
| 239 |
InvalidOffsetNumber) /* (4a) */ \ |
| 240 |
&& (GET_PREDICATELOCKTARGETTAG_PAGE(covering_target) == \ |
| 241 |
GET_PREDICATELOCKTARGETTAG_PAGE(covered_target))) \ |
| 242 |
|| ((GET_PREDICATELOCKTARGETTAG_PAGE(covering_target) == \ |
| 243 |
InvalidBlockNumber) /* (4b) */ \ |
| 244 |
&& (GET_PREDICATELOCKTARGETTAG_PAGE(covered_target) \ |
| 245 |
!= InvalidBlockNumber))) \ |
| 246 |
&& (GET_PREDICATELOCKTARGETTAG_DB(covered_target) == /* (1) */ \ |
| 247 |
GET_PREDICATELOCKTARGETTAG_DB(covering_target))) |
| 248 |
|
| 249 |
/* |
| 250 |
* The predicate locking target and lock shared hash tables are partitioned to |
| 251 |
* reduce contention. To determine which partition a given target belongs to, |
| 252 |
* compute the tag's hash code with PredicateLockTargetTagHashCode(), then |
| 253 |
* apply one of these macros. |
| 254 |
* NB: NUM_PREDICATELOCK_PARTITIONS must be a power of 2! |
| 255 |
*/ |
| 256 |
#define PredicateLockHashPartition(hashcode) \ |
| 257 |
((hashcode) % NUM_PREDICATELOCK_PARTITIONS) |
| 258 |
#define PredicateLockHashPartitionLock(hashcode) \ |
| 259 |
(&MainLWLockArray[PREDICATELOCK_MANAGER_LWLOCK_OFFSET + \ |
| 260 |
PredicateLockHashPartition(hashcode)].lock) |
| 261 |
#define PredicateLockHashPartitionLockByIndex(i) \ |
| 262 |
(&MainLWLockArray[PREDICATELOCK_MANAGER_LWLOCK_OFFSET + (i)].lock) |
| 263 |
|
| 264 |
#define NPREDICATELOCKTARGETENTS() \ |
| 265 |
mul_size(max_predicate_locks_per_xact, add_size(MaxBackends, max_prepared_xacts)) |
| 266 |
|
| 267 |
#define SxactIsOnFinishedList(sxact) (!dlist_node_is_detached(&(sxact)->finishedLink)) |
| 268 |
|
| 269 |
/* |
| 270 |
* Note that a sxact is marked "prepared" once it has passed |
| 271 |
* PreCommit_CheckForSerializationFailure, even if it isn't using |
| 272 |
* 2PC. This is the point at which it can no longer be aborted. |
| 273 |
* |
| 274 |
* The PREPARED flag remains set after commit, so SxactIsCommitted |
| 275 |
* implies SxactIsPrepared. |
| 276 |
*/ |
| 277 |
#define SxactIsCommitted(sxact) (((sxact)->flags & SXACT_FLAG_COMMITTED) != 0) |
| 278 |
#define SxactIsPrepared(sxact) (((sxact)->flags & SXACT_FLAG_PREPARED) != 0) |
| 279 |
#define SxactIsRolledBack(sxact) (((sxact)->flags & SXACT_FLAG_ROLLED_BACK) != 0) |
| 280 |
#define SxactIsDoomed(sxact) (((sxact)->flags & SXACT_FLAG_DOOMED) != 0) |
| 281 |
#define SxactIsReadOnly(sxact) (((sxact)->flags & SXACT_FLAG_READ_ONLY) != 0) |
| 282 |
#define SxactHasSummaryConflictIn(sxact) (((sxact)->flags & SXACT_FLAG_SUMMARY_CONFLICT_IN) != 0) |
| 283 |
#define SxactHasSummaryConflictOut(sxact) (((sxact)->flags & SXACT_FLAG_SUMMARY_CONFLICT_OUT) != 0) |
| 284 |
/* |
| 285 |
* The following macro actually means that the specified transaction has a |
| 286 |
* conflict out *to a transaction which committed ahead of it*. It's hard |
| 287 |
* to get that into a name of a reasonable length. |
| 288 |
*/ |
| 289 |
#define SxactHasConflictOut(sxact) (((sxact)->flags & SXACT_FLAG_CONFLICT_OUT) != 0) |
| 290 |
#define SxactIsDeferrableWaiting(sxact) (((sxact)->flags & SXACT_FLAG_DEFERRABLE_WAITING) != 0) |
| 291 |
#define SxactIsROSafe(sxact) (((sxact)->flags & SXACT_FLAG_RO_SAFE) != 0) |
| 292 |
#define SxactIsROUnsafe(sxact) (((sxact)->flags & SXACT_FLAG_RO_UNSAFE) != 0) |
| 293 |
#define SxactIsPartiallyReleased(sxact) (((sxact)->flags & SXACT_FLAG_PARTIALLY_RELEASED) != 0) |
| 294 |
|
| 295 |
/* |
| 296 |
* Compute the hash code associated with a PREDICATELOCKTARGETTAG. |
| 297 |
* |
| 298 |
* To avoid unnecessary recomputations of the hash code, we try to do this |
| 299 |
* just once per function, and then pass it around as needed. Aside from |
| 300 |
* passing the hashcode to hash_search_with_hash_value(), we can extract |
| 301 |
* the lock partition number from the hashcode. |
| 302 |
*/ |
| 303 |
#define PredicateLockTargetTagHashCode(predicatelocktargettag) \ |
| 304 |
get_hash_value(PredicateLockTargetHash, predicatelocktargettag) |
| 305 |
|
| 306 |
/* |
| 307 |
* Given a predicate lock tag, and the hash for its target, |
| 308 |
* compute the lock hash. |
| 309 |
* |
| 310 |
* To make the hash code also depend on the transaction, we xor the sxid |
| 311 |
* struct's address into the hash code, left-shifted so that the |
| 312 |
* partition-number bits don't change. Since this is only a hash, we |
| 313 |
* don't care if we lose high-order bits of the address; use an |
| 314 |
* intermediate variable to suppress cast-pointer-to-int warnings. |
| 315 |
*/ |
| 316 |
#define PredicateLockHashCodeFromTargetHashCode(predicatelocktag, targethash) \ |
| 317 |
((targethash) ^ ((uint32) PointerGetDatum((predicatelocktag)->myXact)) \ |
| 318 |
<< LOG2_NUM_PREDICATELOCK_PARTITIONS) |
| 319 |
|
| 320 |
|
| 321 |
/* |
| 322 |
* The SLRU buffer area through which we access the old xids. |
| 323 |
*/ |
| 324 |
static SlruCtlData SerialSlruCtlData; |
| 325 |
|
| 326 |
#define SerialSlruCtl (&SerialSlruCtlData) |
| 327 |
|
| 328 |
#define SERIAL_PAGESIZE BLCKSZ |
| 329 |
#define SERIAL_ENTRYSIZE sizeof(SerCommitSeqNo) |
| 330 |
#define SERIAL_ENTRIESPERPAGE (SERIAL_PAGESIZE / SERIAL_ENTRYSIZE) |
| 331 |
|
| 332 |
/* |
| 333 |
* Set maximum pages based on the number needed to track all transactions. |
| 334 |
*/ |
| 335 |
#define SERIAL_MAX_PAGE (MaxTransactionId / SERIAL_ENTRIESPERPAGE) |
| 336 |
|
| 337 |
#define SerialNextPage(page) (((page) >= SERIAL_MAX_PAGE) ? 0 : (page) + 1) |
| 338 |
|
| 339 |
#define SerialValue(slotno, xid) (*((SerCommitSeqNo *) \ |
| 340 |
(SerialSlruCtl->shared->page_buffer[slotno] + \ |
| 341 |
((((uint32) (xid)) % SERIAL_ENTRIESPERPAGE) * SERIAL_ENTRYSIZE)))) |
| 342 |
|
| 343 |
#define SerialPage(xid) (((uint32) (xid)) / SERIAL_ENTRIESPERPAGE) |
| 344 |
|
| 345 |
typedef struct SerialControlData |
| 346 |
{ |
| 347 |
int64 headPage; /* newest initialized page */ |
| 348 |
TransactionId headXid; /* newest valid Xid in the SLRU */ |
| 349 |
TransactionId tailXid; /* oldest xmin we might be interested in */ |
| 350 |
} SerialControlData; |
| 351 |
|
| 352 |
typedef struct SerialControlData *SerialControl; |
| 353 |
|
| 354 |
static SerialControl serialControl; |
| 355 |
|
| 356 |
/* |
| 357 |
* When the oldest committed transaction on the "finished" list is moved to |
| 358 |
* SLRU, its predicate locks will be moved to this "dummy" transaction, |
| 359 |
* collapsing duplicate targets. When a duplicate is found, the later |
| 360 |
* commitSeqNo is used. |
| 361 |
*/ |
| 362 |
static SERIALIZABLEXACT *OldCommittedSxact; |
| 363 |
|
| 364 |
|
| 365 |
/* |
| 366 |
* These configuration variables are used to set the predicate lock table size |
| 367 |
* and to control promotion of predicate locks to coarser granularity in an |
| 368 |
* attempt to degrade performance (mostly as false positive serialization |
| 369 |
* failure) gracefully in the face of memory pressure. |
| 370 |
*/ |
| 371 |
int max_predicate_locks_per_xact; /* in guc_tables.c */ |
| 372 |
int max_predicate_locks_per_relation; /* in guc_tables.c */ |
| 373 |
int max_predicate_locks_per_page; /* in guc_tables.c */ |
| 374 |
|
| 375 |
/* |
| 376 |
* This provides a list of objects in order to track transactions |
| 377 |
* participating in predicate locking. Entries in the list are fixed size, |
| 378 |
* and reside in shared memory. The memory address of an entry must remain |
| 379 |
* fixed during its lifetime. The list will be protected from concurrent |
| 380 |
* update externally; no provision is made in this code to manage that. The |
| 381 |
* number of entries in the list, and the size allowed for each entry is |
| 382 |
* fixed upon creation. |
| 383 |
*/ |
| 384 |
static PredXactList PredXact; |
| 385 |
|
| 386 |
/* |
| 387 |
* This provides a pool of RWConflict data elements to use in conflict lists |
| 388 |
* between transactions. |
| 389 |
*/ |
| 390 |
static RWConflictPoolHeader RWConflictPool; |
| 391 |
|
| 392 |
/* |
| 393 |
* The predicate locking hash tables are in shared memory. |
| 394 |
* Each backend keeps pointers to them. |
| 395 |
*/ |
| 396 |
static HTAB *SerializableXidHash; |
| 397 |
static HTAB *PredicateLockTargetHash; |
| 398 |
static HTAB *PredicateLockHash; |
| 399 |
static dlist_head *FinishedSerializableTransactions; |
| 400 |
|
| 401 |
/* |
| 402 |
* Tag for a dummy entry in PredicateLockTargetHash. By temporarily removing |
| 403 |
* this entry, you can ensure that there's enough scratch space available for |
| 404 |
* inserting one entry in the hash table. This is an otherwise-invalid tag. |
| 405 |
*/ |
| 406 |
static const PREDICATELOCKTARGETTAG ScratchTargetTag = {0, 0, 0, 0}; |
| 407 |
static uint32 ScratchTargetTagHash; |
| 408 |
static LWLock *ScratchPartitionLock; |
| 409 |
|
| 410 |
/* |
| 411 |
* The local hash table used to determine when to combine multiple fine- |
| 412 |
* grained locks into a single courser-grained lock. |
| 413 |
*/ |
| 414 |
static HTAB *LocalPredicateLockHash = NULL; |
| 415 |
|
| 416 |
/* |
| 417 |
* Keep a pointer to the currently-running serializable transaction (if any) |
| 418 |
* for quick reference. Also, remember if we have written anything that could |
| 419 |
* cause a rw-conflict. |
| 420 |
*/ |
| 421 |
static SERIALIZABLEXACT *MySerializableXact = InvalidSerializableXact; |
| 422 |
static bool MyXactDidWrite = false; |
| 423 |
|
| 424 |
/* |
| 425 |
* The SXACT_FLAG_RO_UNSAFE optimization might lead us to release |
| 426 |
* MySerializableXact early. If that happens in a parallel query, the leader |
| 427 |
* needs to defer the destruction of the SERIALIZABLEXACT until end of |
| 428 |
* transaction, because the workers still have a reference to it. In that |
| 429 |
* case, the leader stores it here. |
| 430 |
*/ |
| 431 |
static SERIALIZABLEXACT *SavedSerializableXact = InvalidSerializableXact; |
| 432 |
|
| 433 |
/* local functions */ |
| 434 |
|
| 435 |
static SERIALIZABLEXACT *CreatePredXact(void); |
| 436 |
static void ReleasePredXact(SERIALIZABLEXACT *sxact); |
| 437 |
|
| 438 |
static bool RWConflictExists(const SERIALIZABLEXACT *reader, const SERIALIZABLEXACT *writer); |
| 439 |
static void SetRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer); |
| 440 |
static void SetPossibleUnsafeConflict(SERIALIZABLEXACT *roXact, SERIALIZABLEXACT *activeXact); |
| 441 |
static void ReleaseRWConflict(RWConflict conflict); |
| 442 |
static void FlagSxactUnsafe(SERIALIZABLEXACT *sxact); |
| 443 |
|
| 444 |
static bool SerialPagePrecedesLogically(int64 page1, int64 page2); |
| 445 |
static void SerialInit(void); |
| 446 |
static void SerialAdd(TransactionId xid, SerCommitSeqNo minConflictCommitSeqNo); |
| 447 |
static SerCommitSeqNo SerialGetMinConflictCommitSeqNo(TransactionId xid); |
| 448 |
static void SerialSetActiveSerXmin(TransactionId xid); |
| 449 |
|
| 450 |
static uint32 predicatelock_hash(const void *key, Size keysize); |
| 451 |
static void SummarizeOldestCommittedSxact(void); |
| 452 |
static Snapshot GetSafeSnapshot(Snapshot origSnapshot); |
| 453 |
static Snapshot GetSerializableTransactionSnapshotInt(Snapshot snapshot, |
| 454 |
VirtualTransactionId *sourcevxid, |
| 455 |
int sourcepid); |
| 456 |
static bool PredicateLockExists(const PREDICATELOCKTARGETTAG *targettag); |
| 457 |
static bool GetParentPredicateLockTag(const PREDICATELOCKTARGETTAG *tag, |
| 458 |
PREDICATELOCKTARGETTAG *parent); |
| 459 |
static bool CoarserLockCovers(const PREDICATELOCKTARGETTAG *newtargettag); |
| 460 |
static void RemoveScratchTarget(bool lockheld); |
| 461 |
static void RestoreScratchTarget(bool lockheld); |
| 462 |
static void RemoveTargetIfNoLongerUsed(PREDICATELOCKTARGET *target, |
| 463 |
uint32 targettaghash); |
| 464 |
static void DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag); |
| 465 |
static int MaxPredicateChildLocks(const PREDICATELOCKTARGETTAG *tag); |
| 466 |
static bool CheckAndPromotePredicateLockRequest(const PREDICATELOCKTARGETTAG *reqtag); |
| 467 |
static void DecrementParentLocks(const PREDICATELOCKTARGETTAG *targettag); |
| 468 |
static void CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag, |
| 469 |
uint32 targettaghash, |
| 470 |
SERIALIZABLEXACT *sxact); |
| 471 |
static void DeleteLockTarget(PREDICATELOCKTARGET *target, uint32 targettaghash); |
| 472 |
static bool TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag, |
| 473 |
PREDICATELOCKTARGETTAG newtargettag, |
| 474 |
bool removeOld); |
| 475 |
static void PredicateLockAcquire(const PREDICATELOCKTARGETTAG *targettag); |
| 476 |
static void DropAllPredicateLocksFromTable(Relation relation, |
| 477 |
bool transfer); |
| 478 |
static void SetNewSxactGlobalXmin(void); |
| 479 |
static void ClearOldPredicateLocks(void); |
| 480 |
static void ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial, |
| 481 |
bool summarize); |
| 482 |
static bool XidIsConcurrent(TransactionId xid); |
| 483 |
static void CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag); |
| 484 |
static void FlagRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer); |
| 485 |
static void OnConflict_CheckForSerializationFailure(const SERIALIZABLEXACT *reader, |
| 486 |
SERIALIZABLEXACT *writer); |
| 487 |
static void CreateLocalPredicateLockHash(void); |
| 488 |
static void ReleasePredicateLocksLocal(void); |
| 489 |
|
| 490 |
|
| 491 |
/*------------------------------------------------------------------------*/ |
| 492 |
|
| 493 |
/* |
| 494 |
* Does this relation participate in predicate locking? Temporary and system |
| 495 |
* relations are exempt. |
| 496 |
*/ |
| 497 |
static inline bool |
| 498 |
PredicateLockingNeededForRelation(Relation relation) |
| 499 |
{ |
| 500 |
return !(relation->rd_id < FirstUnpinnedObjectId || |
| 501 |
RelationUsesLocalBuffers(relation)); |
| 502 |
} |
| 503 |
|
| 504 |
/* |
| 505 |
* When a public interface method is called for a read, this is the test to |
| 506 |
* see if we should do a quick return. |
| 507 |
* |
| 508 |
* Note: this function has side-effects! If this transaction has been flagged |
| 509 |
* as RO-safe since the last call, we release all predicate locks and reset |
| 510 |
* MySerializableXact. That makes subsequent calls to return quickly. |
| 511 |
* |
| 512 |
* This is marked as 'inline' to eliminate the function call overhead in the |
| 513 |
* common case that serialization is not needed. |
| 514 |
*/ |
| 515 |
static inline bool |
| 516 |
SerializationNeededForRead(Relation relation, Snapshot snapshot) |
| 517 |
{ |
| 518 |
/* Nothing to do if this is not a serializable transaction */ |
| 519 |
if (MySerializableXact == InvalidSerializableXact) |
| 520 |
return false; |
| 521 |
|
| 522 |
/* |
| 523 |
* Don't acquire locks or conflict when scanning with a special snapshot. |
| 524 |
* This excludes things like CLUSTER and REINDEX. They use the wholesale |
| 525 |
* functions TransferPredicateLocksToHeapRelation() and |
| 526 |
* CheckTableForSerializableConflictIn() to participate in serialization, |
| 527 |
* but the scans involved don't need serialization. |
| 528 |
*/ |
| 529 |
if (!IsMVCCSnapshot(snapshot)) |
| 530 |
return false; |
| 531 |
|
| 532 |
/* |
| 533 |
* Check if we have just become "RO-safe". If we have, immediately release |
| 534 |
* all locks as they're not needed anymore. This also resets |
| 535 |
* MySerializableXact, so that subsequent calls to this function can exit |
| 536 |
* quickly. |
| 537 |
* |
| 538 |
* A transaction is flagged as RO_SAFE if all concurrent R/W transactions |
| 539 |
* commit without having conflicts out to an earlier snapshot, thus |
| 540 |
* ensuring that no conflicts are possible for this transaction. |
| 541 |
*/ |
| 542 |
if (SxactIsROSafe(MySerializableXact)) |
| 543 |
{ |
| 544 |
ReleasePredicateLocks(false, true); |
| 545 |
return false; |
| 546 |
} |
| 547 |
|
| 548 |
/* Check if the relation doesn't participate in predicate locking */ |
| 549 |
if (!PredicateLockingNeededForRelation(relation)) |
| 550 |
return false; |
| 551 |
|
| 552 |
return true; /* no excuse to skip predicate locking */ |
| 553 |
} |
| 554 |
|
| 555 |
/* |
| 556 |
* Like SerializationNeededForRead(), but called on writes. |
| 557 |
* The logic is the same, but there is no snapshot and we can't be RO-safe. |
| 558 |
*/ |
| 559 |
static inline bool |
| 560 |
SerializationNeededForWrite(Relation relation) |
| 561 |
{ |
| 562 |
/* Nothing to do if this is not a serializable transaction */ |
| 563 |
if (MySerializableXact == InvalidSerializableXact) |
| 564 |
return false; |
| 565 |
|
| 566 |
/* Check if the relation doesn't participate in predicate locking */ |
| 567 |
if (!PredicateLockingNeededForRelation(relation)) |
| 568 |
return false; |
| 569 |
|
| 570 |
return true; /* no excuse to skip predicate locking */ |
| 571 |
} |
| 572 |
|
| 573 |
|
| 574 |
/*------------------------------------------------------------------------*/ |
| 575 |
|
| 576 |
/* |
| 577 |
* These functions are a simple implementation of a list for this specific |
| 578 |
* type of struct. If there is ever a generalized shared memory list, we |
| 579 |
* should probably switch to that. |
| 580 |
*/ |
| 581 |
static SERIALIZABLEXACT * |
| 582 |
CreatePredXact(void) |
| 583 |
{ |
| 584 |
SERIALIZABLEXACT *sxact; |
| 585 |
|
| 586 |
if (dlist_is_empty(&PredXact->availableList)) |
| 587 |
return NULL; |
| 588 |
|
| 589 |
sxact = dlist_container(SERIALIZABLEXACT, xactLink, |
| 590 |
dlist_pop_head_node(&PredXact->availableList)); |
| 591 |
dlist_push_tail(&PredXact->activeList, &sxact->xactLink); |
| 592 |
return sxact; |
| 593 |
} |
| 594 |
|
| 595 |
static void |
| 596 |
ReleasePredXact(SERIALIZABLEXACT *sxact) |
| 597 |
{ |
| 598 |
Assert(ShmemAddrIsValid(sxact)); |
| 599 |
|
| 600 |
dlist_delete(&sxact->xactLink); |
| 601 |
dlist_push_tail(&PredXact->availableList, &sxact->xactLink); |
| 602 |
} |
| 603 |
|
| 604 |
/*------------------------------------------------------------------------*/ |
| 605 |
|
| 606 |
/* |
| 607 |
* These functions manage primitive access to the RWConflict pool and lists. |
| 608 |
*/ |
| 609 |
static bool |
| 610 |
RWConflictExists(const SERIALIZABLEXACT *reader, const SERIALIZABLEXACT *writer) |
| 611 |
{ |
| 612 |
dlist_iter iter; |
| 613 |
|
| 614 |
Assert(reader != writer); |
| 615 |
|
| 616 |
/* Check the ends of the purported conflict first. */ |
| 617 |
if (SxactIsDoomed(reader) |
| 618 |
|| SxactIsDoomed(writer) |
| 619 |
|| dlist_is_empty(&reader->outConflicts) |
| 620 |
|| dlist_is_empty(&writer->inConflicts)) |
| 621 |
return false; |
| 622 |
|
| 623 |
/* |
| 624 |
* A conflict is possible; walk the list to find out. |
| 625 |
* |
| 626 |
* The unconstify is needed as we have no const version of |
| 627 |
* dlist_foreach(). |
| 628 |
*/ |
| 629 |
dlist_foreach(iter, &unconstify(SERIALIZABLEXACT *, reader)->outConflicts) |
| 630 |
{ |
| 631 |
RWConflict conflict = |
| 632 |
dlist_container(RWConflictData, outLink, iter.cur); |
| 633 |
|
| 634 |
if (conflict->sxactIn == writer) |
| 635 |
return true; |
| 636 |
} |
| 637 |
|
| 638 |
/* No conflict found. */ |
| 639 |
return false; |
| 640 |
} |
| 641 |
|
| 642 |
static void |
| 643 |
SetRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer) |
| 644 |
{ |
| 645 |
RWConflict conflict; |
| 646 |
|
| 647 |
Assert(reader != writer); |
| 648 |
Assert(!RWConflictExists(reader, writer)); |
| 649 |
|
| 650 |
if (dlist_is_empty(&RWConflictPool->availableList)) |
| 651 |
ereport(ERROR, |
| 652 |
(errcode(ERRCODE_OUT_OF_MEMORY), |
| 653 |
errmsg("not enough elements in RWConflictPool to record a read/write conflict"), |
| 654 |
errhint("You might need to run fewer transactions at a time or increase \"max_connections\"."))); |
| 655 |
|
| 656 |
conflict = dlist_head_element(RWConflictData, outLink, &RWConflictPool->availableList); |
| 657 |
dlist_delete(&conflict->outLink); |
| 658 |
|
| 659 |
conflict->sxactOut = reader; |
| 660 |
conflict->sxactIn = writer; |
| 661 |
dlist_push_tail(&reader->outConflicts, &conflict->outLink); |
| 662 |
dlist_push_tail(&writer->inConflicts, &conflict->inLink); |
| 663 |
} |
| 664 |
|
| 665 |
static void |
| 666 |
SetPossibleUnsafeConflict(SERIALIZABLEXACT *roXact, |
| 667 |
SERIALIZABLEXACT *activeXact) |
| 668 |
{ |
| 669 |
RWConflict conflict; |
| 670 |
|
| 671 |
Assert(roXact != activeXact); |
| 672 |
Assert(SxactIsReadOnly(roXact)); |
| 673 |
Assert(!SxactIsReadOnly(activeXact)); |
| 674 |
|
| 675 |
if (dlist_is_empty(&RWConflictPool->availableList)) |
| 676 |
ereport(ERROR, |
| 677 |
(errcode(ERRCODE_OUT_OF_MEMORY), |
| 678 |
errmsg("not enough elements in RWConflictPool to record a potential read/write conflict"), |
| 679 |
errhint("You might need to run fewer transactions at a time or increase \"max_connections\"."))); |
| 680 |
|
| 681 |
conflict = dlist_head_element(RWConflictData, outLink, &RWConflictPool->availableList); |
| 682 |
dlist_delete(&conflict->outLink); |
| 683 |
|
| 684 |
conflict->sxactOut = activeXact; |
| 685 |
conflict->sxactIn = roXact; |
| 686 |
dlist_push_tail(&activeXact->possibleUnsafeConflicts, &conflict->outLink); |
| 687 |
dlist_push_tail(&roXact->possibleUnsafeConflicts, &conflict->inLink); |
| 688 |
} |
| 689 |
|
| 690 |
static void |
| 691 |
ReleaseRWConflict(RWConflict conflict) |
| 692 |
{ |
| 693 |
dlist_delete(&conflict->inLink); |
| 694 |
dlist_delete(&conflict->outLink); |
| 695 |
dlist_push_tail(&RWConflictPool->availableList, &conflict->outLink); |
| 696 |
} |
| 697 |
|
| 698 |
static void |
| 699 |
FlagSxactUnsafe(SERIALIZABLEXACT *sxact) |
| 700 |
{ |
| 701 |
dlist_mutable_iter iter; |
| 702 |
|
| 703 |
Assert(SxactIsReadOnly(sxact)); |
| 704 |
Assert(!SxactIsROSafe(sxact)); |
| 705 |
|
| 706 |
sxact->flags |= SXACT_FLAG_RO_UNSAFE; |
| 707 |
|
| 708 |
/* |
| 709 |
* We know this isn't a safe snapshot, so we can stop looking for other |
| 710 |
* potential conflicts. |
| 711 |
*/ |
| 712 |
dlist_foreach_modify(iter, &sxact->possibleUnsafeConflicts) |
| 713 |
{ |
| 714 |
RWConflict conflict = |
| 715 |
dlist_container(RWConflictData, inLink, iter.cur); |
| 716 |
|
| 717 |
Assert(!SxactIsReadOnly(conflict->sxactOut)); |
| 718 |
Assert(sxact == conflict->sxactIn); |
| 719 |
|
| 720 |
ReleaseRWConflict(conflict); |
| 721 |
} |
| 722 |
} |
| 723 |
|
| 724 |
/*------------------------------------------------------------------------*/ |
| 725 |
|
| 726 |
/* |
| 727 |
* Decide whether a Serial page number is "older" for truncation purposes. |
| 728 |
* Analogous to CLOGPagePrecedes(). |
| 729 |
*/ |
| 730 |
static bool |
| 731 |
SerialPagePrecedesLogically(int64 page1, int64 page2) |
| 732 |
{ |
| 733 |
TransactionId xid1; |
| 734 |
TransactionId xid2; |
| 735 |
|
| 736 |
xid1 = ((TransactionId) page1) * SERIAL_ENTRIESPERPAGE; |
| 737 |
xid1 += FirstNormalTransactionId + 1; |
| 738 |
xid2 = ((TransactionId) page2) * SERIAL_ENTRIESPERPAGE; |
| 739 |
xid2 += FirstNormalTransactionId + 1; |
| 740 |
|
| 741 |
return (TransactionIdPrecedes(xid1, xid2) && |
| 742 |
TransactionIdPrecedes(xid1, xid2 + SERIAL_ENTRIESPERPAGE - 1)); |
| 743 |
} |
| 744 |
|
| 745 |
#ifdef USE_ASSERT_CHECKING |
| 746 |
static void |
| 747 |
SerialPagePrecedesLogicallyUnitTests(void) |
| 748 |
{ |
| 749 |
int per_page = SERIAL_ENTRIESPERPAGE, |
| 750 |
offset = per_page / 2; |
| 751 |
int64 newestPage, |
| 752 |
oldestPage, |
| 753 |
headPage, |
| 754 |
targetPage; |
| 755 |
TransactionId newestXact, |
| 756 |
oldestXact; |
| 757 |
|
| 758 |
/* GetNewTransactionId() has assigned the last XID it can safely use. */ |
| 759 |
newestPage = 2 * SLRU_PAGES_PER_SEGMENT - 1; /* nothing special */ |
| 760 |
newestXact = newestPage * per_page + offset; |
| 761 |
Assert(newestXact / per_page == newestPage); |
| 762 |
oldestXact = newestXact + 1; |
| 763 |
oldestXact -= 1U << 31; |
| 764 |
oldestPage = oldestXact / per_page; |
| 765 |
|
| 766 |
/* |
| 767 |
* In this scenario, the SLRU headPage pertains to the last ~1000 XIDs |
| 768 |
* assigned. oldestXact finishes, ~2B XIDs having elapsed since it |
| 769 |
* started. Further transactions cause us to summarize oldestXact to |
| 770 |
* tailPage. Function must return false so SerialAdd() doesn't zero |
| 771 |
* tailPage (which may contain entries for other old, recently-finished |
| 772 |
* XIDs) and half the SLRU. Reaching this requires burning ~2B XIDs in |
| 773 |
* single-user mode, a negligible possibility. |
| 774 |
*/ |
| 775 |
headPage = newestPage; |
| 776 |
targetPage = oldestPage; |
| 777 |
Assert(!SerialPagePrecedesLogically(headPage, targetPage)); |
| 778 |
|
| 779 |
/* |
| 780 |
* In this scenario, the SLRU headPage pertains to oldestXact. We're |
| 781 |
* summarizing an XID near newestXact. (Assume few other XIDs used |
| 782 |
* SERIALIZABLE, hence the minimal headPage advancement. Assume |
| 783 |
* oldestXact was long-running and only recently reached the SLRU.) |
| 784 |
* Function must return true to make SerialAdd() create targetPage. |
| 785 |
* |
| 786 |
* Today's implementation mishandles this case, but it doesn't matter |
| 787 |
* enough to fix. Verify that the defect affects just one page by |
| 788 |
* asserting correct treatment of its prior page. Reaching this case |
| 789 |
* requires burning ~2B XIDs in single-user mode, a negligible |
| 790 |
* possibility. Moreover, if it does happen, the consequence would be |
| 791 |
* mild, namely a new transaction failing in SimpleLruReadPage(). |
| 792 |
*/ |
| 793 |
headPage = oldestPage; |
| 794 |
targetPage = newestPage; |
| 795 |
Assert(SerialPagePrecedesLogically(headPage, targetPage - 1)); |
| 796 |
#if 0 |
| 797 |
Assert(SerialPagePrecedesLogically(headPage, targetPage)); |
| 798 |
#endif |
| 799 |
} |
| 800 |
#endif |
| 801 |
|
| 802 |
/* |
| 803 |
* Initialize for the tracking of old serializable committed xids. |
| 804 |
*/ |
| 805 |
static void |
| 806 |
SerialInit(void) |
| 807 |
{ |
| 808 |
bool found; |
| 809 |
|
| 810 |
/* |
| 811 |
* Set up SLRU management of the pg_serial data. |
| 812 |
*/ |
| 813 |
SerialSlruCtl->PagePrecedes = SerialPagePrecedesLogically; |
| 814 |
SimpleLruInit(SerialSlruCtl, "serializable", |
| 815 |
serializable_buffers, 0, "pg_serial", |
| 816 |
LWTRANCHE_SERIAL_BUFFER, LWTRANCHE_SERIAL_SLRU, |
| 817 |
SYNC_HANDLER_NONE, false); |
| 818 |
#ifdef USE_ASSERT_CHECKING |
| 819 |
SerialPagePrecedesLogicallyUnitTests(); |
| 820 |
#endif |
| 821 |
SlruPagePrecedesUnitTests(SerialSlruCtl, SERIAL_ENTRIESPERPAGE); |
| 822 |
|
| 823 |
/* |
| 824 |
* Create or attach to the SerialControl structure. |
| 825 |
*/ |
| 826 |
serialControl = (SerialControl) |
| 827 |
ShmemInitStruct("SerialControlData", sizeof(SerialControlData), &found); |
| 828 |
|
| 829 |
Assert(found == IsUnderPostmaster); |
| 830 |
if (!found) |
| 831 |
{ |
| 832 |
/* |
| 833 |
* Set control information to reflect empty SLRU. |
| 834 |
*/ |
| 835 |
LWLockAcquire(SerialControlLock, LW_EXCLUSIVE); |
| 836 |
serialControl->headPage = -1; |
| 837 |
serialControl->headXid = InvalidTransactionId; |
| 838 |
serialControl->tailXid = InvalidTransactionId; |
| 839 |
LWLockRelease(SerialControlLock); |
| 840 |
} |
| 841 |
} |
| 842 |
|
| 843 |
/* |
| 844 |
* GUC check_hook for serializable_buffers |
| 845 |
*/ |
| 846 |
bool |
| 847 |
check_serial_buffers(int *newval, void **extra, GucSource source) |
| 848 |
{ |
| 849 |
return check_slru_buffers("serializable_buffers", newval); |
| 850 |
} |
| 851 |
|
| 852 |
/* |
| 853 |
* Record a committed read write serializable xid and the minimum |
| 854 |
* commitSeqNo of any transactions to which this xid had a rw-conflict out. |
| 855 |
* An invalid commitSeqNo means that there were no conflicts out from xid. |
| 856 |
*/ |
| 857 |
static void |
| 858 |
SerialAdd(TransactionId xid, SerCommitSeqNo minConflictCommitSeqNo) |
| 859 |
{ |
| 860 |
TransactionId tailXid; |
| 861 |
int64 targetPage; |
| 862 |
int slotno; |
| 863 |
int64 firstZeroPage; |
| 864 |
bool isNewPage; |
| 865 |
LWLock *lock; |
| 866 |
|
| 867 |
Assert(TransactionIdIsValid(xid)); |
| 868 |
|
| 869 |
targetPage = SerialPage(xid); |
| 870 |
lock = SimpleLruGetBankLock(SerialSlruCtl, targetPage); |
| 871 |
|
| 872 |
/* |
| 873 |
* In this routine, we must hold both SerialControlLock and the SLRU bank |
| 874 |
* lock simultaneously while making the SLRU data catch up with the new |
| 875 |
* state that we determine. |
| 876 |
*/ |
| 877 |
LWLockAcquire(SerialControlLock, LW_EXCLUSIVE); |
| 878 |
|
| 879 |
/* |
| 880 |
* If 'xid' is older than the global xmin (== tailXid), there's no need to |
| 881 |
* store it, after all. This can happen if the oldest transaction holding |
| 882 |
* back the global xmin just finished, making 'xid' uninteresting, but |
| 883 |
* ClearOldPredicateLocks() has not yet run. |
| 884 |
*/ |
| 885 |
tailXid = serialControl->tailXid; |
| 886 |
if (!TransactionIdIsValid(tailXid) || TransactionIdPrecedes(xid, tailXid)) |
| 887 |
{ |
| 888 |
LWLockRelease(SerialControlLock); |
| 889 |
return; |
| 890 |
} |
| 891 |
|
| 892 |
/* |
| 893 |
* If the SLRU is currently unused, zero out the whole active region from |
| 894 |
* tailXid to headXid before taking it into use. Otherwise zero out only |
| 895 |
* any new pages that enter the tailXid-headXid range as we advance |
| 896 |
* headXid. |
| 897 |
*/ |
| 898 |
if (serialControl->headPage < 0) |
| 899 |
{ |
| 900 |
firstZeroPage = SerialPage(tailXid); |
| 901 |
isNewPage = true; |
| 902 |
} |
| 903 |
else |
| 904 |
{ |
| 905 |
firstZeroPage = SerialNextPage(serialControl->headPage); |
| 906 |
isNewPage = SerialPagePrecedesLogically(serialControl->headPage, |
| 907 |
targetPage); |
| 908 |
} |
| 909 |
|
| 910 |
if (!TransactionIdIsValid(serialControl->headXid) |
| 911 |
|| TransactionIdFollows(xid, serialControl->headXid)) |
| 912 |
serialControl->headXid = xid; |
| 913 |
if (isNewPage) |
| 914 |
serialControl->headPage = targetPage; |
| 915 |
|
| 916 |
if (isNewPage) |
| 917 |
{ |
| 918 |
/* Initialize intervening pages; might involve trading locks */ |
| 919 |
for (;;) |
| 920 |
{ |
| 921 |
lock = SimpleLruGetBankLock(SerialSlruCtl, firstZeroPage); |
| 922 |
LWLockAcquire(lock, LW_EXCLUSIVE); |
| 923 |
slotno = SimpleLruZeroPage(SerialSlruCtl, firstZeroPage); |
| 924 |
if (firstZeroPage == targetPage) |
| 925 |
break; |
| 926 |
firstZeroPage = SerialNextPage(firstZeroPage); |
| 927 |
LWLockRelease(lock); |
| 928 |
} |
| 929 |
} |
| 930 |
else |
| 931 |
{ |
| 932 |
LWLockAcquire(lock, LW_EXCLUSIVE); |
| 933 |
slotno = SimpleLruReadPage(SerialSlruCtl, targetPage, true, xid); |
| 934 |
} |
| 935 |
|
| 936 |
SerialValue(slotno, xid) = minConflictCommitSeqNo; |
| 937 |
SerialSlruCtl->shared->page_dirty[slotno] = true; |
| 938 |
|
| 939 |
LWLockRelease(lock); |
| 940 |
LWLockRelease(SerialControlLock); |
| 941 |
} |
| 942 |
|
| 943 |
/* |
| 944 |
* Get the minimum commitSeqNo for any conflict out for the given xid. For |
| 945 |
* a transaction which exists but has no conflict out, InvalidSerCommitSeqNo |
| 946 |
* will be returned. |
| 947 |
*/ |
| 948 |
static SerCommitSeqNo |
| 949 |
SerialGetMinConflictCommitSeqNo(TransactionId xid) |
| 950 |
{ |
| 951 |
TransactionId headXid; |
| 952 |
TransactionId tailXid; |
| 953 |
SerCommitSeqNo val; |
| 954 |
int slotno; |
| 955 |
|
| 956 |
Assert(TransactionIdIsValid(xid)); |
| 957 |
|
| 958 |
LWLockAcquire(SerialControlLock, LW_SHARED); |
| 959 |
headXid = serialControl->headXid; |
| 960 |
tailXid = serialControl->tailXid; |
| 961 |
LWLockRelease(SerialControlLock); |
| 962 |
|
| 963 |
if (!TransactionIdIsValid(headXid)) |
| 964 |
return 0; |
| 965 |
|
| 966 |
Assert(TransactionIdIsValid(tailXid)); |
| 967 |
|
| 968 |
if (TransactionIdPrecedes(xid, tailXid) |
| 969 |
|| TransactionIdFollows(xid, headXid)) |
| 970 |
return 0; |
| 971 |
|
| 972 |
/* |
| 973 |
* The following function must be called without holding SLRU bank lock, |
| 974 |
* but will return with that lock held, which must then be released. |
| 975 |
*/ |
| 976 |
slotno = SimpleLruReadPage_ReadOnly(SerialSlruCtl, |
| 977 |
SerialPage(xid), xid); |
| 978 |
val = SerialValue(slotno, xid); |
| 979 |
LWLockRelease(SimpleLruGetBankLock(SerialSlruCtl, SerialPage(xid))); |
| 980 |
return val; |
| 981 |
} |
| 982 |
|
| 983 |
/* |
| 984 |
* Call this whenever there is a new xmin for active serializable |
| 985 |
* transactions. We don't need to keep information on transactions which |
| 986 |
* precede that. InvalidTransactionId means none active, so everything in |
| 987 |
* the SLRU can be discarded. |
| 988 |
*/ |
| 989 |
static void |
| 990 |
SerialSetActiveSerXmin(TransactionId xid) |
| 991 |
{ |
| 992 |
LWLockAcquire(SerialControlLock, LW_EXCLUSIVE); |
| 993 |
|
| 994 |
/* |
| 995 |
* When no sxacts are active, nothing overlaps, set the xid values to |
| 996 |
* invalid to show that there are no valid entries. Don't clear headPage, |
| 997 |
* though. A new xmin might still land on that page, and we don't want to |
| 998 |
* repeatedly zero out the same page. |
| 999 |
*/ |
| 1000 |
if (!TransactionIdIsValid(xid)) |
| 1001 |
{ |
| 1002 |
serialControl->tailXid = InvalidTransactionId; |
| 1003 |
serialControl->headXid = InvalidTransactionId; |
| 1004 |
LWLockRelease(SerialControlLock); |
| 1005 |
return; |
| 1006 |
} |
| 1007 |
|
| 1008 |
/* |
| 1009 |
* When we're recovering prepared transactions, the global xmin might move |
| 1010 |
* backwards depending on the order they're recovered. Normally that's not |
| 1011 |
* OK, but during recovery no serializable transactions will commit, so |
| 1012 |
* the SLRU is empty and we can get away with it. |
| 1013 |
*/ |
| 1014 |
if (RecoveryInProgress()) |
| 1015 |
{ |
| 1016 |
Assert(serialControl->headPage < 0); |
| 1017 |
if (!TransactionIdIsValid(serialControl->tailXid) |
| 1018 |
|| TransactionIdPrecedes(xid, serialControl->tailXid)) |
| 1019 |
{ |
| 1020 |
serialControl->tailXid = xid; |
| 1021 |
} |
| 1022 |
LWLockRelease(SerialControlLock); |
| 1023 |
return; |
| 1024 |
} |
| 1025 |
|
| 1026 |
Assert(!TransactionIdIsValid(serialControl->tailXid) |
| 1027 |
|| TransactionIdFollows(xid, serialControl->tailXid)); |
| 1028 |
|
| 1029 |
serialControl->tailXid = xid; |
| 1030 |
|
| 1031 |
LWLockRelease(SerialControlLock); |
| 1032 |
} |
| 1033 |
|
| 1034 |
/* |
| 1035 |
* Perform a checkpoint --- either during shutdown, or on-the-fly |
| 1036 |
* |
| 1037 |
* We don't have any data that needs to survive a restart, but this is a |
| 1038 |
* convenient place to truncate the SLRU. |
| 1039 |
*/ |
| 1040 |
void |
| 1041 |
CheckPointPredicate(void) |
| 1042 |
{ |
| 1043 |
int64 truncateCutoffPage; |
| 1044 |
|
| 1045 |
LWLockAcquire(SerialControlLock, LW_EXCLUSIVE); |
| 1046 |
|
| 1047 |
/* Exit quickly if the SLRU is currently not in use. */ |
| 1048 |
if (serialControl->headPage < 0) |
| 1049 |
{ |
| 1050 |
LWLockRelease(SerialControlLock); |
| 1051 |
return; |
| 1052 |
} |
| 1053 |
|
| 1054 |
if (TransactionIdIsValid(serialControl->tailXid)) |
| 1055 |
{ |
| 1056 |
int64 tailPage; |
| 1057 |
|
| 1058 |
tailPage = SerialPage(serialControl->tailXid); |
| 1059 |
|
| 1060 |
/* |
| 1061 |
* It is possible for the tailXid to be ahead of the headXid. This |
| 1062 |
* occurs if we checkpoint while there are in-progress serializable |
| 1063 |
* transaction(s) advancing the tail but we are yet to summarize the |
| 1064 |
* transactions. In this case, we cutoff up to the headPage and the |
| 1065 |
* next summary will advance the headXid. |
| 1066 |
*/ |
| 1067 |
if (SerialPagePrecedesLogically(tailPage, serialControl->headPage)) |
| 1068 |
{ |
| 1069 |
/* We can truncate the SLRU up to the page containing tailXid */ |
| 1070 |
truncateCutoffPage = tailPage; |
| 1071 |
} |
| 1072 |
else |
| 1073 |
truncateCutoffPage = serialControl->headPage; |
| 1074 |
} |
| 1075 |
else |
| 1076 |
{ |
| 1077 |
/*---------- |
| 1078 |
* The SLRU is no longer needed. Truncate to head before we set head |
| 1079 |
* invalid. |
| 1080 |
* |
| 1081 |
* XXX: It's possible that the SLRU is not needed again until XID |
| 1082 |
* wrap-around has happened, so that the segment containing headPage |
| 1083 |
* that we leave behind will appear to be new again. In that case it |
| 1084 |
* won't be removed until XID horizon advances enough to make it |
| 1085 |
* current again. |
| 1086 |
* |
| 1087 |
* XXX: This should happen in vac_truncate_clog(), not in checkpoints. |
| 1088 |
* Consider this scenario, starting from a system with no in-progress |
| 1089 |
* transactions and VACUUM FREEZE having maximized oldestXact: |
| 1090 |
* - Start a SERIALIZABLE transaction. |
| 1091 |
* - Start, finish, and summarize a SERIALIZABLE transaction, creating |
| 1092 |
* one SLRU page. |
| 1093 |
* - Consume XIDs to reach xidStopLimit. |
| 1094 |
* - Finish all transactions. Due to the long-running SERIALIZABLE |
| 1095 |
* transaction, earlier checkpoints did not touch headPage. The |
| 1096 |
* next checkpoint will change it, but that checkpoint happens after |
| 1097 |
* the end of the scenario. |
| 1098 |
* - VACUUM to advance XID limits. |
| 1099 |
* - Consume ~2M XIDs, crossing the former xidWrapLimit. |
| 1100 |
* - Start, finish, and summarize a SERIALIZABLE transaction. |
| 1101 |
* SerialAdd() declines to create the targetPage, because headPage |
| 1102 |
* is not regarded as in the past relative to that targetPage. The |
| 1103 |
* transaction instigating the summarize fails in |
| 1104 |
* SimpleLruReadPage(). |
| 1105 |
*/ |
| 1106 |
truncateCutoffPage = serialControl->headPage; |
| 1107 |
serialControl->headPage = -1; |
| 1108 |
} |
| 1109 |
|
| 1110 |
LWLockRelease(SerialControlLock); |
| 1111 |
|
| 1112 |
/* |
| 1113 |
* Truncate away pages that are no longer required. Note that no |
| 1114 |
* additional locking is required, because this is only called as part of |
| 1115 |
* a checkpoint, and the validity limits have already been determined. |
| 1116 |
*/ |
| 1117 |
SimpleLruTruncate(SerialSlruCtl, truncateCutoffPage); |
| 1118 |
|
| 1119 |
/* |
| 1120 |
* Write dirty SLRU pages to disk |
| 1121 |
* |
| 1122 |
* This is not actually necessary from a correctness point of view. We do |
| 1123 |
* it merely as a debugging aid. |
| 1124 |
* |
| 1125 |
* We're doing this after the truncation to avoid writing pages right |
| 1126 |
* before deleting the file in which they sit, which would be completely |
| 1127 |
* pointless. |
| 1128 |
*/ |
| 1129 |
SimpleLruWriteAll(SerialSlruCtl, true); |
| 1130 |
} |
| 1131 |
|
| 1132 |
/*------------------------------------------------------------------------*/ |
| 1133 |
|
| 1134 |
/* |
| 1135 |
* PredicateLockShmemInit -- Initialize the predicate locking data structures. |
| 1136 |
* |
| 1137 |
* This is called from CreateSharedMemoryAndSemaphores(), which see for |
| 1138 |
* more comments. In the normal postmaster case, the shared hash tables |
| 1139 |
* are created here. Backends inherit the pointers |
| 1140 |
* to the shared tables via fork(). In the EXEC_BACKEND case, each |
| 1141 |
* backend re-executes this code to obtain pointers to the already existing |
| 1142 |
* shared hash tables. |
| 1143 |
*/ |
| 1144 |
void |
| 1145 |
PredicateLockShmemInit(void) |
| 1146 |
{ |
| 1147 |
HASHCTL info; |
| 1148 |
long max_table_size; |
| 1149 |
Size requestSize; |
| 1150 |
bool found; |
| 1151 |
|
| 1152 |
#ifndef EXEC_BACKEND |
| 1153 |
Assert(!IsUnderPostmaster); |
| 1154 |
#endif |
| 1155 |
|
| 1156 |
/* |
| 1157 |
* Compute size of predicate lock target hashtable. Note these |
| 1158 |
* calculations must agree with PredicateLockShmemSize! |
| 1159 |
*/ |
| 1160 |
max_table_size = NPREDICATELOCKTARGETENTS(); |
| 1161 |
|
| 1162 |
/* |
| 1163 |
* Allocate hash table for PREDICATELOCKTARGET structs. This stores |
| 1164 |
* per-predicate-lock-target information. |
| 1165 |
*/ |
| 1166 |
info.keysize = sizeof(PREDICATELOCKTARGETTAG); |
| 1167 |
info.entrysize = sizeof(PREDICATELOCKTARGET); |
| 1168 |
info.num_partitions = NUM_PREDICATELOCK_PARTITIONS; |
| 1169 |
|
| 1170 |
PredicateLockTargetHash = ShmemInitHash("PREDICATELOCKTARGET hash", |
| 1171 |
max_table_size, |
| 1172 |
max_table_size, |
| 1173 |
&info, |
| 1174 |
HASH_ELEM | HASH_BLOBS | |
| 1175 |
HASH_PARTITION | HASH_FIXED_SIZE); |
| 1176 |
|
| 1177 |
/* |
| 1178 |
* Reserve a dummy entry in the hash table; we use it to make sure there's |
| 1179 |
* always one entry available when we need to split or combine a page, |
| 1180 |
* because running out of space there could mean aborting a |
| 1181 |
* non-serializable transaction. |
| 1182 |
*/ |
| 1183 |
if (!IsUnderPostmaster) |
| 1184 |
{ |
| 1185 |
(void) hash_search(PredicateLockTargetHash, &ScratchTargetTag, |
| 1186 |
HASH_ENTER, &found); |
| 1187 |
Assert(!found); |
| 1188 |
} |
| 1189 |
|
| 1190 |
/* Pre-calculate the hash and partition lock of the scratch entry */ |
| 1191 |
ScratchTargetTagHash = PredicateLockTargetTagHashCode(&ScratchTargetTag); |
| 1192 |
ScratchPartitionLock = PredicateLockHashPartitionLock(ScratchTargetTagHash); |
| 1193 |
|
| 1194 |
/* |
| 1195 |
* Allocate hash table for PREDICATELOCK structs. This stores per |
| 1196 |
* xact-lock-of-a-target information. |
| 1197 |
*/ |
| 1198 |
info.keysize = sizeof(PREDICATELOCKTAG); |
| 1199 |
info.entrysize = sizeof(PREDICATELOCK); |
| 1200 |
info.hash = predicatelock_hash; |
| 1201 |
info.num_partitions = NUM_PREDICATELOCK_PARTITIONS; |
| 1202 |
|
| 1203 |
/* Assume an average of 2 xacts per target */ |
| 1204 |
max_table_size *= 2; |
| 1205 |
|
| 1206 |
PredicateLockHash = ShmemInitHash("PREDICATELOCK hash", |
| 1207 |
max_table_size, |
| 1208 |
max_table_size, |
| 1209 |
&info, |
| 1210 |
HASH_ELEM | HASH_FUNCTION | |
| 1211 |
HASH_PARTITION | HASH_FIXED_SIZE); |
| 1212 |
|
| 1213 |
/* |
| 1214 |
* Compute size for serializable transaction hashtable. Note these |
| 1215 |
* calculations must agree with PredicateLockShmemSize! |
| 1216 |
*/ |
| 1217 |
max_table_size = (MaxBackends + max_prepared_xacts); |
| 1218 |
|
| 1219 |
/* |
| 1220 |
* Allocate a list to hold information on transactions participating in |
| 1221 |
* predicate locking. |
| 1222 |
* |
| 1223 |
* Assume an average of 10 predicate locking transactions per backend. |
| 1224 |
* This allows aggressive cleanup while detail is present before data must |
| 1225 |
* be summarized for storage in SLRU and the "dummy" transaction. |
| 1226 |
*/ |
| 1227 |
max_table_size *= 10; |
| 1228 |
|
| 1229 |
requestSize = add_size(PredXactListDataSize, |
| 1230 |
(mul_size((Size) max_table_size, |
| 1231 |
sizeof(SERIALIZABLEXACT)))); |
| 1232 |
|
| 1233 |
PredXact = ShmemInitStruct("PredXactList", |
| 1234 |
requestSize, |
| 1235 |
&found); |
| 1236 |
Assert(found == IsUnderPostmaster); |
| 1237 |
if (!found) |
| 1238 |
{ |
| 1239 |
int i; |
| 1240 |
|
| 1241 |
/* clean everything, both the header and the element */ |
| 1242 |
memset(PredXact, 0, requestSize); |
| 1243 |
|
| 1244 |
dlist_init(&PredXact->availableList); |
| 1245 |
dlist_init(&PredXact->activeList); |
| 1246 |
PredXact->SxactGlobalXmin = InvalidTransactionId; |
| 1247 |
PredXact->SxactGlobalXminCount = 0; |
| 1248 |
PredXact->WritableSxactCount = 0; |
| 1249 |
PredXact->LastSxactCommitSeqNo = FirstNormalSerCommitSeqNo - 1; |
| 1250 |
PredXact->CanPartialClearThrough = 0; |
| 1251 |
PredXact->HavePartialClearedThrough = 0; |
| 1252 |
PredXact->element |
| 1253 |
= (SERIALIZABLEXACT *) ((char *) PredXact + PredXactListDataSize); |
| 1254 |
/* Add all elements to available list, clean. */ |
| 1255 |
for (i = 0; i < max_table_size; i++) |
| 1256 |
{ |
| 1257 |
LWLockInitialize(&PredXact->element[i].perXactPredicateListLock, |
| 1258 |
LWTRANCHE_PER_XACT_PREDICATE_LIST); |
| 1259 |
dlist_push_tail(&PredXact->availableList, &PredXact->element[i].xactLink); |
| 1260 |
} |
| 1261 |
PredXact->OldCommittedSxact = CreatePredXact(); |
| 1262 |
SetInvalidVirtualTransactionId(PredXact->OldCommittedSxact->vxid); |
| 1263 |
PredXact->OldCommittedSxact->prepareSeqNo = 0; |
| 1264 |
PredXact->OldCommittedSxact->commitSeqNo = 0; |
| 1265 |
PredXact->OldCommittedSxact->SeqNo.lastCommitBeforeSnapshot = 0; |
| 1266 |
dlist_init(&PredXact->OldCommittedSxact->outConflicts); |
| 1267 |
dlist_init(&PredXact->OldCommittedSxact->inConflicts); |
| 1268 |
dlist_init(&PredXact->OldCommittedSxact->predicateLocks); |
| 1269 |
dlist_node_init(&PredXact->OldCommittedSxact->finishedLink); |
| 1270 |
dlist_init(&PredXact->OldCommittedSxact->possibleUnsafeConflicts); |
| 1271 |
PredXact->OldCommittedSxact->topXid = InvalidTransactionId; |
| 1272 |
PredXact->OldCommittedSxact->finishedBefore = InvalidTransactionId; |
| 1273 |
PredXact->OldCommittedSxact->xmin = InvalidTransactionId; |
| 1274 |
PredXact->OldCommittedSxact->flags = SXACT_FLAG_COMMITTED; |
| 1275 |
PredXact->OldCommittedSxact->pid = 0; |
| 1276 |
PredXact->OldCommittedSxact->pgprocno = INVALID_PROC_NUMBER; |
| 1277 |
} |
| 1278 |
/* This never changes, so let's keep a local copy. */ |
| 1279 |
OldCommittedSxact = PredXact->OldCommittedSxact; |
| 1280 |
|
| 1281 |
/* |
| 1282 |
* Allocate hash table for SERIALIZABLEXID structs. This stores per-xid |
| 1283 |
* information for serializable transactions which have accessed data. |
| 1284 |
*/ |
| 1285 |
info.keysize = sizeof(SERIALIZABLEXIDTAG); |
| 1286 |
info.entrysize = sizeof(SERIALIZABLEXID); |
| 1287 |
|
| 1288 |
SerializableXidHash = ShmemInitHash("SERIALIZABLEXID hash", |
| 1289 |
max_table_size, |
| 1290 |
max_table_size, |
| 1291 |
&info, |
| 1292 |
HASH_ELEM | HASH_BLOBS | |
| 1293 |
HASH_FIXED_SIZE); |
| 1294 |
|
| 1295 |
/* |
| 1296 |
* Allocate space for tracking rw-conflicts in lists attached to the |
| 1297 |
* transactions. |
| 1298 |
* |
| 1299 |
* Assume an average of 5 conflicts per transaction. Calculations suggest |
| 1300 |
* that this will prevent resource exhaustion in even the most pessimal |
| 1301 |
* loads up to max_connections = 200 with all 200 connections pounding the |
| 1302 |
* database with serializable transactions. Beyond that, there may be |
| 1303 |
* occasional transactions canceled when trying to flag conflicts. That's |
| 1304 |
* probably OK. |
| 1305 |
*/ |
| 1306 |
max_table_size *= 5; |
| 1307 |
|
| 1308 |
requestSize = RWConflictPoolHeaderDataSize + |
| 1309 |
mul_size((Size) max_table_size, |
| 1310 |
RWConflictDataSize); |
| 1311 |
|
| 1312 |
RWConflictPool = ShmemInitStruct("RWConflictPool", |
| 1313 |
requestSize, |
| 1314 |
&found); |
| 1315 |
Assert(found == IsUnderPostmaster); |
| 1316 |
if (!found) |
| 1317 |
{ |
| 1318 |
int i; |
| 1319 |
|
| 1320 |
/* clean everything, including the elements */ |
| 1321 |
memset(RWConflictPool, 0, requestSize); |
| 1322 |
|
| 1323 |
dlist_init(&RWConflictPool->availableList); |
| 1324 |
RWConflictPool->element = (RWConflict) ((char *) RWConflictPool + |
| 1325 |
RWConflictPoolHeaderDataSize); |
| 1326 |
/* Add all elements to available list, clean. */ |
| 1327 |
for (i = 0; i < max_table_size; i++) |
| 1328 |
{ |
| 1329 |
dlist_push_tail(&RWConflictPool->availableList, |
| 1330 |
&RWConflictPool->element[i].outLink); |
| 1331 |
} |
| 1332 |
} |
| 1333 |
|
| 1334 |
/* |
| 1335 |
* Create or attach to the header for the list of finished serializable |
| 1336 |
* transactions. |
| 1337 |
*/ |
| 1338 |
FinishedSerializableTransactions = (dlist_head *) |
| 1339 |
ShmemInitStruct("FinishedSerializableTransactions", |
| 1340 |
sizeof(dlist_head), |
| 1341 |
&found); |
| 1342 |
Assert(found == IsUnderPostmaster); |
| 1343 |
if (!found) |
| 1344 |
dlist_init(FinishedSerializableTransactions); |
| 1345 |
|
| 1346 |
/* |
| 1347 |
* Initialize the SLRU storage for old committed serializable |
| 1348 |
* transactions. |
| 1349 |
*/ |
| 1350 |
SerialInit(); |
| 1351 |
} |
| 1352 |
|
| 1353 |
/* |
| 1354 |
* Estimate shared-memory space used for predicate lock table |
| 1355 |
*/ |
| 1356 |
Size |
| 1357 |
PredicateLockShmemSize(void) |
| 1358 |
{ |
| 1359 |
Size size = 0; |
| 1360 |
long max_table_size; |
| 1361 |
|
| 1362 |
/* predicate lock target hash table */ |
| 1363 |
max_table_size = NPREDICATELOCKTARGETENTS(); |
| 1364 |
size = add_size(size, hash_estimate_size(max_table_size, |
| 1365 |
sizeof(PREDICATELOCKTARGET))); |
| 1366 |
|
| 1367 |
/* predicate lock hash table */ |
| 1368 |
max_table_size *= 2; |
| 1369 |
size = add_size(size, hash_estimate_size(max_table_size, |
| 1370 |
sizeof(PREDICATELOCK))); |
| 1371 |
|
| 1372 |
/* |
| 1373 |
* Since NPREDICATELOCKTARGETENTS is only an estimate, add 10% safety |
| 1374 |
* margin. |
| 1375 |
*/ |
| 1376 |
size = add_size(size, size / 10); |
| 1377 |
|
| 1378 |
/* transaction list */ |
| 1379 |
max_table_size = MaxBackends + max_prepared_xacts; |
| 1380 |
max_table_size *= 10; |
| 1381 |
size = add_size(size, PredXactListDataSize); |
| 1382 |
size = add_size(size, mul_size((Size) max_table_size, |
| 1383 |
sizeof(SERIALIZABLEXACT))); |
| 1384 |
|
| 1385 |
/* transaction xid table */ |
| 1386 |
size = add_size(size, hash_estimate_size(max_table_size, |
| 1387 |
sizeof(SERIALIZABLEXID))); |
| 1388 |
|
| 1389 |
/* rw-conflict pool */ |
| 1390 |
max_table_size *= 5; |
| 1391 |
size = add_size(size, RWConflictPoolHeaderDataSize); |
| 1392 |
size = add_size(size, mul_size((Size) max_table_size, |
| 1393 |
RWConflictDataSize)); |
| 1394 |
|
| 1395 |
/* Head for list of finished serializable transactions. */ |
| 1396 |
size = add_size(size, sizeof(dlist_head)); |
| 1397 |
|
| 1398 |
/* Shared memory structures for SLRU tracking of old committed xids. */ |
| 1399 |
size = add_size(size, sizeof(SerialControlData)); |
| 1400 |
size = add_size(size, SimpleLruShmemSize(serializable_buffers, 0)); |
| 1401 |
|
| 1402 |
return size; |
| 1403 |
} |
| 1404 |
|
| 1405 |
|
| 1406 |
/* |
| 1407 |
* Compute the hash code associated with a PREDICATELOCKTAG. |
| 1408 |
* |
| 1409 |
* Because we want to use just one set of partition locks for both the |
| 1410 |
* PREDICATELOCKTARGET and PREDICATELOCK hash tables, we have to make sure |
| 1411 |
* that PREDICATELOCKs fall into the same partition number as their |
| 1412 |
* associated PREDICATELOCKTARGETs. dynahash.c expects the partition number |
| 1413 |
* to be the low-order bits of the hash code, and therefore a |
| 1414 |
* PREDICATELOCKTAG's hash code must have the same low-order bits as the |
| 1415 |
* associated PREDICATELOCKTARGETTAG's hash code. We achieve this with this |
| 1416 |
* specialized hash function. |
| 1417 |
*/ |
| 1418 |
static uint32 |
| 1419 |
predicatelock_hash(const void *key, Size keysize) |
| 1420 |
{ |
| 1421 |
const PREDICATELOCKTAG *predicatelocktag = (const PREDICATELOCKTAG *) key; |
| 1422 |
uint32 targethash; |
| 1423 |
|
| 1424 |
Assert(keysize == sizeof(PREDICATELOCKTAG)); |
| 1425 |
|
| 1426 |
/* Look into the associated target object, and compute its hash code */ |
| 1427 |
targethash = PredicateLockTargetTagHashCode(&predicatelocktag->myTarget->tag); |
| 1428 |
|
| 1429 |
return PredicateLockHashCodeFromTargetHashCode(predicatelocktag, targethash); |
| 1430 |
} |
| 1431 |
|
| 1432 |
|
| 1433 |
/* |
| 1434 |
* GetPredicateLockStatusData |
| 1435 |
* Return a table containing the internal state of the predicate |
| 1436 |
* lock manager for use in pg_lock_status. |
| 1437 |
* |
| 1438 |
* Like GetLockStatusData, this function tries to hold the partition LWLocks |
| 1439 |
* for as short a time as possible by returning two arrays that simply |
| 1440 |
* contain the PREDICATELOCKTARGETTAG and SERIALIZABLEXACT for each lock |
| 1441 |
* table entry. Multiple copies of the same PREDICATELOCKTARGETTAG and |
| 1442 |
* SERIALIZABLEXACT will likely appear. |
| 1443 |
*/ |
| 1444 |
PredicateLockData * |
| 1445 |
GetPredicateLockStatusData(void) |
| 1446 |
{ |
| 1447 |
PredicateLockData *data; |
| 1448 |
int i; |
| 1449 |
int els, |
| 1450 |
el; |
| 1451 |
HASH_SEQ_STATUS seqstat; |
| 1452 |
PREDICATELOCK *predlock; |
| 1453 |
|
| 1454 |
data = (PredicateLockData *) palloc(sizeof(PredicateLockData)); |
| 1455 |
|
| 1456 |
/* |
| 1457 |
* To ensure consistency, take simultaneous locks on all partition locks |
| 1458 |
* in ascending order, then SerializableXactHashLock. |
| 1459 |
*/ |
| 1460 |
for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++) |
| 1461 |
LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_SHARED); |
| 1462 |
LWLockAcquire(SerializableXactHashLock, LW_SHARED); |
| 1463 |
|
| 1464 |
/* Get number of locks and allocate appropriately-sized arrays. */ |
| 1465 |
els = hash_get_num_entries(PredicateLockHash); |
| 1466 |
data->nelements = els; |
| 1467 |
data->locktags = (PREDICATELOCKTARGETTAG *) |
| 1468 |
palloc(sizeof(PREDICATELOCKTARGETTAG) * els); |
| 1469 |
data->xacts = (SERIALIZABLEXACT *) |
| 1470 |
palloc(sizeof(SERIALIZABLEXACT) * els); |
| 1471 |
|
| 1472 |
|
| 1473 |
/* Scan through PredicateLockHash and copy contents */ |
| 1474 |
hash_seq_init(&seqstat, PredicateLockHash); |
| 1475 |
|
| 1476 |
el = 0; |
| 1477 |
|
| 1478 |
while ((predlock = (PREDICATELOCK *) hash_seq_search(&seqstat))) |
| 1479 |
{ |
| 1480 |
data->locktags[el] = predlock->tag.myTarget->tag; |
| 1481 |
data->xacts[el] = *predlock->tag.myXact; |
| 1482 |
el++; |
| 1483 |
} |
| 1484 |
|
| 1485 |
Assert(el == els); |
| 1486 |
|
| 1487 |
/* Release locks in reverse order */ |
| 1488 |
LWLockRelease(SerializableXactHashLock); |
| 1489 |
for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--) |
| 1490 |
LWLockRelease(PredicateLockHashPartitionLockByIndex(i)); |
| 1491 |
|
| 1492 |
return data; |
| 1493 |
} |
| 1494 |
|
| 1495 |
/* |
| 1496 |
* Free up shared memory structures by pushing the oldest sxact (the one at |
| 1497 |
* the front of the SummarizeOldestCommittedSxact queue) into summary form. |
| 1498 |
* Each call will free exactly one SERIALIZABLEXACT structure and may also |
| 1499 |
* free one or more of these structures: SERIALIZABLEXID, PREDICATELOCK, |
| 1500 |
* PREDICATELOCKTARGET, RWConflictData. |
| 1501 |
*/ |
| 1502 |
static void |
| 1503 |
SummarizeOldestCommittedSxact(void) |
| 1504 |
{ |
| 1505 |
SERIALIZABLEXACT *sxact; |
| 1506 |
|
| 1507 |
LWLockAcquire(SerializableFinishedListLock, LW_EXCLUSIVE); |
| 1508 |
|
| 1509 |
/* |
| 1510 |
* This function is only called if there are no sxact slots available. |
| 1511 |
* Some of them must belong to old, already-finished transactions, so |
| 1512 |
* there should be something in FinishedSerializableTransactions list that |
| 1513 |
* we can summarize. However, there's a race condition: while we were not |
| 1514 |
* holding any locks, a transaction might have ended and cleaned up all |
| 1515 |
* the finished sxact entries already, freeing up their sxact slots. In |
| 1516 |
* that case, we have nothing to do here. The caller will find one of the |
| 1517 |
* slots released by the other backend when it retries. |
| 1518 |
*/ |
| 1519 |
if (dlist_is_empty(FinishedSerializableTransactions)) |
| 1520 |
{ |
| 1521 |
LWLockRelease(SerializableFinishedListLock); |
| 1522 |
return; |
| 1523 |
} |
| 1524 |
|
| 1525 |
/* |
| 1526 |
* Grab the first sxact off the finished list -- this will be the earliest |
| 1527 |
* commit. Remove it from the list. |
| 1528 |
*/ |
| 1529 |
sxact = dlist_head_element(SERIALIZABLEXACT, finishedLink, |
| 1530 |
FinishedSerializableTransactions); |
| 1531 |
dlist_delete_thoroughly(&sxact->finishedLink); |
| 1532 |
|
| 1533 |
/* Add to SLRU summary information. */ |
| 1534 |
if (TransactionIdIsValid(sxact->topXid) && !SxactIsReadOnly(sxact)) |
| 1535 |
SerialAdd(sxact->topXid, SxactHasConflictOut(sxact) |
| 1536 |
? sxact->SeqNo.earliestOutConflictCommit : InvalidSerCommitSeqNo); |
| 1537 |
|
| 1538 |
/* Summarize and release the detail. */ |
| 1539 |
ReleaseOneSerializableXact(sxact, false, true); |
| 1540 |
|
| 1541 |
LWLockRelease(SerializableFinishedListLock); |
| 1542 |
} |
| 1543 |
|
| 1544 |
/* |
| 1545 |
* GetSafeSnapshot |
| 1546 |
* Obtain and register a snapshot for a READ ONLY DEFERRABLE |
| 1547 |
* transaction. Ensures that the snapshot is "safe", i.e. a |
| 1548 |
* read-only transaction running on it can execute serializably |
| 1549 |
* without further checks. This requires waiting for concurrent |
| 1550 |
* transactions to complete, and retrying with a new snapshot if |
| 1551 |
* one of them could possibly create a conflict. |
| 1552 |
* |
| 1553 |
* As with GetSerializableTransactionSnapshot (which this is a subroutine |
| 1554 |
* for), the passed-in Snapshot pointer should reference a static data |
| 1555 |
* area that can safely be passed to GetSnapshotData. |
| 1556 |
*/ |
| 1557 |
static Snapshot |
| 1558 |
GetSafeSnapshot(Snapshot origSnapshot) |
| 1559 |
{ |
| 1560 |
Snapshot snapshot; |
| 1561 |
|
| 1562 |
Assert(XactReadOnly && XactDeferrable); |
| 1563 |
|
| 1564 |
while (true) |
| 1565 |
{ |
| 1566 |
/* |
| 1567 |
* GetSerializableTransactionSnapshotInt is going to call |
| 1568 |
* GetSnapshotData, so we need to provide it the static snapshot area |
| 1569 |
* our caller passed to us. The pointer returned is actually the same |
| 1570 |
* one passed to it, but we avoid assuming that here. |
| 1571 |
*/ |
| 1572 |
snapshot = GetSerializableTransactionSnapshotInt(origSnapshot, |
| 1573 |
NULL, InvalidPid); |
| 1574 |
|
| 1575 |
if (MySerializableXact == InvalidSerializableXact) |
| 1576 |
return snapshot; /* no concurrent r/w xacts; it's safe */ |
| 1577 |
|
| 1578 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 1579 |
|
| 1580 |
/* |
| 1581 |
* Wait for concurrent transactions to finish. Stop early if one of |
| 1582 |
* them marked us as conflicted. |
| 1583 |
*/ |
| 1584 |
MySerializableXact->flags |= SXACT_FLAG_DEFERRABLE_WAITING; |
| 1585 |
while (!(dlist_is_empty(&MySerializableXact->possibleUnsafeConflicts) || |
| 1586 |
SxactIsROUnsafe(MySerializableXact))) |
| 1587 |
{ |
| 1588 |
LWLockRelease(SerializableXactHashLock); |
| 1589 |
ProcWaitForSignal(WAIT_EVENT_SAFE_SNAPSHOT); |
| 1590 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 1591 |
} |
| 1592 |
MySerializableXact->flags &= ~SXACT_FLAG_DEFERRABLE_WAITING; |
| 1593 |
|
| 1594 |
if (!SxactIsROUnsafe(MySerializableXact)) |
| 1595 |
{ |
| 1596 |
LWLockRelease(SerializableXactHashLock); |
| 1597 |
break; /* success */ |
| 1598 |
} |
| 1599 |
|
| 1600 |
LWLockRelease(SerializableXactHashLock); |
| 1601 |
|
| 1602 |
/* else, need to retry... */ |
| 1603 |
ereport(DEBUG2, |
| 1604 |
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE), |
| 1605 |
errmsg_internal("deferrable snapshot was unsafe; trying a new one"))); |
| 1606 |
ReleasePredicateLocks(false, false); |
| 1607 |
} |
| 1608 |
|
| 1609 |
/* |
| 1610 |
* Now we have a safe snapshot, so we don't need to do any further checks. |
| 1611 |
*/ |
| 1612 |
Assert(SxactIsROSafe(MySerializableXact)); |
| 1613 |
ReleasePredicateLocks(false, true); |
| 1614 |
|
| 1615 |
return snapshot; |
| 1616 |
} |
| 1617 |
|
| 1618 |
/* |
| 1619 |
* GetSafeSnapshotBlockingPids |
| 1620 |
* If the specified process is currently blocked in GetSafeSnapshot, |
| 1621 |
* write the process IDs of all processes that it is blocked by |
| 1622 |
* into the caller-supplied buffer output[]. The list is truncated at |
| 1623 |
* output_size, and the number of PIDs written into the buffer is |
| 1624 |
* returned. Returns zero if the given PID is not currently blocked |
| 1625 |
* in GetSafeSnapshot. |
| 1626 |
*/ |
| 1627 |
int |
| 1628 |
GetSafeSnapshotBlockingPids(int blocked_pid, int *output, int output_size) |
| 1629 |
{ |
| 1630 |
int num_written = 0; |
| 1631 |
dlist_iter iter; |
| 1632 |
SERIALIZABLEXACT *blocking_sxact = NULL; |
| 1633 |
|
| 1634 |
LWLockAcquire(SerializableXactHashLock, LW_SHARED); |
| 1635 |
|
| 1636 |
/* Find blocked_pid's SERIALIZABLEXACT by linear search. */ |
| 1637 |
dlist_foreach(iter, &PredXact->activeList) |
| 1638 |
{ |
| 1639 |
SERIALIZABLEXACT *sxact = |
| 1640 |
dlist_container(SERIALIZABLEXACT, xactLink, iter.cur); |
| 1641 |
|
| 1642 |
if (sxact->pid == blocked_pid) |
| 1643 |
{ |
| 1644 |
blocking_sxact = sxact; |
| 1645 |
break; |
| 1646 |
} |
| 1647 |
} |
| 1648 |
|
| 1649 |
/* Did we find it, and is it currently waiting in GetSafeSnapshot? */ |
| 1650 |
if (blocking_sxact != NULL && SxactIsDeferrableWaiting(blocking_sxact)) |
| 1651 |
{ |
| 1652 |
/* Traverse the list of possible unsafe conflicts collecting PIDs. */ |
| 1653 |
dlist_foreach(iter, &blocking_sxact->possibleUnsafeConflicts) |
| 1654 |
{ |
| 1655 |
RWConflict possibleUnsafeConflict = |
| 1656 |
dlist_container(RWConflictData, inLink, iter.cur); |
| 1657 |
|
| 1658 |
output[num_written++] = possibleUnsafeConflict->sxactOut->pid; |
| 1659 |
|
| 1660 |
if (num_written >= output_size) |
| 1661 |
break; |
| 1662 |
} |
| 1663 |
} |
| 1664 |
|
| 1665 |
LWLockRelease(SerializableXactHashLock); |
| 1666 |
|
| 1667 |
return num_written; |
| 1668 |
} |
| 1669 |
|
| 1670 |
/* |
| 1671 |
* Acquire a snapshot that can be used for the current transaction. |
| 1672 |
* |
| 1673 |
* Make sure we have a SERIALIZABLEXACT reference in MySerializableXact. |
| 1674 |
* It should be current for this process and be contained in PredXact. |
| 1675 |
* |
| 1676 |
* The passed-in Snapshot pointer should reference a static data area that |
| 1677 |
* can safely be passed to GetSnapshotData. The return value is actually |
| 1678 |
* always this same pointer; no new snapshot data structure is allocated |
| 1679 |
* within this function. |
| 1680 |
*/ |
| 1681 |
Snapshot |
| 1682 |
GetSerializableTransactionSnapshot(Snapshot snapshot) |
| 1683 |
{ |
| 1684 |
Assert(IsolationIsSerializable()); |
| 1685 |
|
| 1686 |
/* |
| 1687 |
* Can't use serializable mode while recovery is still active, as it is, |
| 1688 |
* for example, on a hot standby. We could get here despite the check in |
| 1689 |
* check_transaction_isolation() if default_transaction_isolation is set |
| 1690 |
* to serializable, so phrase the hint accordingly. |
| 1691 |
*/ |
| 1692 |
if (RecoveryInProgress()) |
| 1693 |
ereport(ERROR, |
| 1694 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 1695 |
errmsg("cannot use serializable mode in a hot standby"), |
| 1696 |
errdetail("\"default_transaction_isolation\" is set to \"serializable\"."), |
| 1697 |
errhint("You can use \"SET default_transaction_isolation = 'repeatable read'\" to change the default."))); |
| 1698 |
|
| 1699 |
/* |
| 1700 |
* A special optimization is available for SERIALIZABLE READ ONLY |
| 1701 |
* DEFERRABLE transactions -- we can wait for a suitable snapshot and |
| 1702 |
* thereby avoid all SSI overhead once it's running. |
| 1703 |
*/ |
| 1704 |
if (XactReadOnly && XactDeferrable) |
| 1705 |
return GetSafeSnapshot(snapshot); |
| 1706 |
|
| 1707 |
return GetSerializableTransactionSnapshotInt(snapshot, |
| 1708 |
NULL, InvalidPid); |
| 1709 |
} |
| 1710 |
|
| 1711 |
/* |
| 1712 |
* Import a snapshot to be used for the current transaction. |
| 1713 |
* |
| 1714 |
* This is nearly the same as GetSerializableTransactionSnapshot, except that |
| 1715 |
* we don't take a new snapshot, but rather use the data we're handed. |
| 1716 |
* |
| 1717 |
* The caller must have verified that the snapshot came from a serializable |
| 1718 |
* transaction; and if we're read-write, the source transaction must not be |
| 1719 |
* read-only. |
| 1720 |
*/ |
| 1721 |
void |
| 1722 |
SetSerializableTransactionSnapshot(Snapshot snapshot, |
| 1723 |
VirtualTransactionId *sourcevxid, |
| 1724 |
int sourcepid) |
| 1725 |
{ |
| 1726 |
Assert(IsolationIsSerializable()); |
| 1727 |
|
| 1728 |
/* |
| 1729 |
* If this is called by parallel.c in a parallel worker, we don't want to |
| 1730 |
* create a SERIALIZABLEXACT just yet because the leader's |
| 1731 |
* SERIALIZABLEXACT will be installed with AttachSerializableXact(). We |
| 1732 |
* also don't want to reject SERIALIZABLE READ ONLY DEFERRABLE in this |
| 1733 |
* case, because the leader has already determined that the snapshot it |
| 1734 |
* has passed us is safe. So there is nothing for us to do. |
| 1735 |
*/ |
| 1736 |
if (IsParallelWorker()) |
| 1737 |
return; |
| 1738 |
|
| 1739 |
/* |
| 1740 |
* We do not allow SERIALIZABLE READ ONLY DEFERRABLE transactions to |
| 1741 |
* import snapshots, since there's no way to wait for a safe snapshot when |
| 1742 |
* we're using the snap we're told to. (XXX instead of throwing an error, |
| 1743 |
* we could just ignore the XactDeferrable flag?) |
| 1744 |
*/ |
| 1745 |
if (XactReadOnly && XactDeferrable) |
| 1746 |
ereport(ERROR, |
| 1747 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 1748 |
errmsg("a snapshot-importing transaction must not be READ ONLY DEFERRABLE"))); |
| 1749 |
|
| 1750 |
(void) GetSerializableTransactionSnapshotInt(snapshot, sourcevxid, |
| 1751 |
sourcepid); |
| 1752 |
} |
| 1753 |
|
| 1754 |
/* |
| 1755 |
* Guts of GetSerializableTransactionSnapshot |
| 1756 |
* |
| 1757 |
* If sourcevxid is valid, this is actually an import operation and we should |
| 1758 |
* skip calling GetSnapshotData, because the snapshot contents are already |
| 1759 |
* loaded up. HOWEVER: to avoid race conditions, we must check that the |
| 1760 |
* source xact is still running after we acquire SerializableXactHashLock. |
| 1761 |
* We do that by calling ProcArrayInstallImportedXmin. |
| 1762 |
*/ |
| 1763 |
static Snapshot |
| 1764 |
GetSerializableTransactionSnapshotInt(Snapshot snapshot, |
| 1765 |
VirtualTransactionId *sourcevxid, |
| 1766 |
int sourcepid) |
| 1767 |
{ |
| 1768 |
PGPROC *proc; |
| 1769 |
VirtualTransactionId vxid; |
| 1770 |
SERIALIZABLEXACT *sxact, |
| 1771 |
*othersxact; |
| 1772 |
|
| 1773 |
/* We only do this for serializable transactions. Once. */ |
| 1774 |
Assert(MySerializableXact == InvalidSerializableXact); |
| 1775 |
|
| 1776 |
Assert(!RecoveryInProgress()); |
| 1777 |
|
| 1778 |
/* |
| 1779 |
* Since all parts of a serializable transaction must use the same |
| 1780 |
* snapshot, it is too late to establish one after a parallel operation |
| 1781 |
* has begun. |
| 1782 |
*/ |
| 1783 |
if (IsInParallelMode()) |
| 1784 |
elog(ERROR, "cannot establish serializable snapshot during a parallel operation"); |
| 1785 |
|
| 1786 |
proc = MyProc; |
| 1787 |
Assert(proc != NULL); |
| 1788 |
GET_VXID_FROM_PGPROC(vxid, *proc); |
| 1789 |
|
| 1790 |
/* |
| 1791 |
* First we get the sxact structure, which may involve looping and access |
| 1792 |
* to the "finished" list to free a structure for use. |
| 1793 |
* |
| 1794 |
* We must hold SerializableXactHashLock when taking/checking the snapshot |
| 1795 |
* to avoid race conditions, for much the same reasons that |
| 1796 |
* GetSnapshotData takes the ProcArrayLock. Since we might have to |
| 1797 |
* release SerializableXactHashLock to call SummarizeOldestCommittedSxact, |
| 1798 |
* this means we have to create the sxact first, which is a bit annoying |
| 1799 |
* (in particular, an elog(ERROR) in procarray.c would cause us to leak |
| 1800 |
* the sxact). Consider refactoring to avoid this. |
| 1801 |
*/ |
| 1802 |
#ifdef TEST_SUMMARIZE_SERIAL |
| 1803 |
SummarizeOldestCommittedSxact(); |
| 1804 |
#endif |
| 1805 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 1806 |
do |
| 1807 |
{ |
| 1808 |
sxact = CreatePredXact(); |
| 1809 |
/* If null, push out committed sxact to SLRU summary & retry. */ |
| 1810 |
if (!sxact) |
| 1811 |
{ |
| 1812 |
LWLockRelease(SerializableXactHashLock); |
| 1813 |
SummarizeOldestCommittedSxact(); |
| 1814 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 1815 |
} |
| 1816 |
} while (!sxact); |
| 1817 |
|
| 1818 |
/* Get the snapshot, or check that it's safe to use */ |
| 1819 |
if (!sourcevxid) |
| 1820 |
snapshot = GetSnapshotData(snapshot); |
| 1821 |
else if (!ProcArrayInstallImportedXmin(snapshot->xmin, sourcevxid)) |
| 1822 |
{ |
| 1823 |
ReleasePredXact(sxact); |
| 1824 |
LWLockRelease(SerializableXactHashLock); |
| 1825 |
ereport(ERROR, |
| 1826 |
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), |
| 1827 |
errmsg("could not import the requested snapshot"), |
| 1828 |
errdetail("The source process with PID %d is not running anymore.", |
| 1829 |
sourcepid))); |
| 1830 |
} |
| 1831 |
|
| 1832 |
/* |
| 1833 |
* If there are no serializable transactions which are not read-only, we |
| 1834 |
* can "opt out" of predicate locking and conflict checking for a |
| 1835 |
* read-only transaction. |
| 1836 |
* |
| 1837 |
* The reason this is safe is that a read-only transaction can only become |
| 1838 |
* part of a dangerous structure if it overlaps a writable transaction |
| 1839 |
* which in turn overlaps a writable transaction which committed before |
| 1840 |
* the read-only transaction started. A new writable transaction can |
| 1841 |
* overlap this one, but it can't meet the other condition of overlapping |
| 1842 |
* a transaction which committed before this one started. |
| 1843 |
*/ |
| 1844 |
if (XactReadOnly && PredXact->WritableSxactCount == 0) |
| 1845 |
{ |
| 1846 |
ReleasePredXact(sxact); |
| 1847 |
LWLockRelease(SerializableXactHashLock); |
| 1848 |
return snapshot; |
| 1849 |
} |
| 1850 |
|
| 1851 |
/* Initialize the structure. */ |
| 1852 |
sxact->vxid = vxid; |
| 1853 |
sxact->SeqNo.lastCommitBeforeSnapshot = PredXact->LastSxactCommitSeqNo; |
| 1854 |
sxact->prepareSeqNo = InvalidSerCommitSeqNo; |
| 1855 |
sxact->commitSeqNo = InvalidSerCommitSeqNo; |
| 1856 |
dlist_init(&(sxact->outConflicts)); |
| 1857 |
dlist_init(&(sxact->inConflicts)); |
| 1858 |
dlist_init(&(sxact->possibleUnsafeConflicts)); |
| 1859 |
sxact->topXid = GetTopTransactionIdIfAny(); |
| 1860 |
sxact->finishedBefore = InvalidTransactionId; |
| 1861 |
sxact->xmin = snapshot->xmin; |
| 1862 |
sxact->pid = MyProcPid; |
| 1863 |
sxact->pgprocno = MyProcNumber; |
| 1864 |
dlist_init(&sxact->predicateLocks); |
| 1865 |
dlist_node_init(&sxact->finishedLink); |
| 1866 |
sxact->flags = 0; |
| 1867 |
if (XactReadOnly) |
| 1868 |
{ |
| 1869 |
dlist_iter iter; |
| 1870 |
|
| 1871 |
sxact->flags |= SXACT_FLAG_READ_ONLY; |
| 1872 |
|
| 1873 |
/* |
| 1874 |
* Register all concurrent r/w transactions as possible conflicts; if |
| 1875 |
* all of them commit without any outgoing conflicts to earlier |
| 1876 |
* transactions then this snapshot can be deemed safe (and we can run |
| 1877 |
* without tracking predicate locks). |
| 1878 |
*/ |
| 1879 |
dlist_foreach(iter, &PredXact->activeList) |
| 1880 |
{ |
| 1881 |
othersxact = dlist_container(SERIALIZABLEXACT, xactLink, iter.cur); |
| 1882 |
|
| 1883 |
if (!SxactIsCommitted(othersxact) |
| 1884 |
&& !SxactIsDoomed(othersxact) |
| 1885 |
&& !SxactIsReadOnly(othersxact)) |
| 1886 |
{ |
| 1887 |
SetPossibleUnsafeConflict(sxact, othersxact); |
| 1888 |
} |
| 1889 |
} |
| 1890 |
|
| 1891 |
/* |
| 1892 |
* If we didn't find any possibly unsafe conflicts because every |
| 1893 |
* uncommitted writable transaction turned out to be doomed, then we |
| 1894 |
* can "opt out" immediately. See comments above the earlier check |
| 1895 |
* for PredXact->WritableSxactCount == 0. |
| 1896 |
*/ |
| 1897 |
if (dlist_is_empty(&sxact->possibleUnsafeConflicts)) |
| 1898 |
{ |
| 1899 |
ReleasePredXact(sxact); |
| 1900 |
LWLockRelease(SerializableXactHashLock); |
| 1901 |
return snapshot; |
| 1902 |
} |
| 1903 |
} |
| 1904 |
else |
| 1905 |
{ |
| 1906 |
++(PredXact->WritableSxactCount); |
| 1907 |
Assert(PredXact->WritableSxactCount <= |
| 1908 |
(MaxBackends + max_prepared_xacts)); |
| 1909 |
} |
| 1910 |
|
| 1911 |
/* Maintain serializable global xmin info. */ |
| 1912 |
if (!TransactionIdIsValid(PredXact->SxactGlobalXmin)) |
| 1913 |
{ |
| 1914 |
Assert(PredXact->SxactGlobalXminCount == 0); |
| 1915 |
PredXact->SxactGlobalXmin = snapshot->xmin; |
| 1916 |
PredXact->SxactGlobalXminCount = 1; |
| 1917 |
SerialSetActiveSerXmin(snapshot->xmin); |
| 1918 |
} |
| 1919 |
else if (TransactionIdEquals(snapshot->xmin, PredXact->SxactGlobalXmin)) |
| 1920 |
{ |
| 1921 |
Assert(PredXact->SxactGlobalXminCount > 0); |
| 1922 |
PredXact->SxactGlobalXminCount++; |
| 1923 |
} |
| 1924 |
else |
| 1925 |
{ |
| 1926 |
Assert(TransactionIdFollows(snapshot->xmin, PredXact->SxactGlobalXmin)); |
| 1927 |
} |
| 1928 |
|
| 1929 |
MySerializableXact = sxact; |
| 1930 |
MyXactDidWrite = false; /* haven't written anything yet */ |
| 1931 |
|
| 1932 |
LWLockRelease(SerializableXactHashLock); |
| 1933 |
|
| 1934 |
CreateLocalPredicateLockHash(); |
| 1935 |
|
| 1936 |
return snapshot; |
| 1937 |
} |
| 1938 |
|
| 1939 |
static void |
| 1940 |
CreateLocalPredicateLockHash(void) |
| 1941 |
{ |
| 1942 |
HASHCTL hash_ctl; |
| 1943 |
|
| 1944 |
/* Initialize the backend-local hash table of parent locks */ |
| 1945 |
Assert(LocalPredicateLockHash == NULL); |
| 1946 |
hash_ctl.keysize = sizeof(PREDICATELOCKTARGETTAG); |
| 1947 |
hash_ctl.entrysize = sizeof(LOCALPREDICATELOCK); |
| 1948 |
LocalPredicateLockHash = hash_create("Local predicate lock", |
| 1949 |
max_predicate_locks_per_xact, |
| 1950 |
&hash_ctl, |
| 1951 |
HASH_ELEM | HASH_BLOBS); |
| 1952 |
} |
| 1953 |
|
| 1954 |
/* |
| 1955 |
* Register the top level XID in SerializableXidHash. |
| 1956 |
* Also store it for easy reference in MySerializableXact. |
| 1957 |
*/ |
| 1958 |
void |
| 1959 |
RegisterPredicateLockingXid(TransactionId xid) |
| 1960 |
{ |
| 1961 |
SERIALIZABLEXIDTAG sxidtag; |
| 1962 |
SERIALIZABLEXID *sxid; |
| 1963 |
bool found; |
| 1964 |
|
| 1965 |
/* |
| 1966 |
* If we're not tracking predicate lock data for this transaction, we |
| 1967 |
* should ignore the request and return quickly. |
| 1968 |
*/ |
| 1969 |
if (MySerializableXact == InvalidSerializableXact) |
| 1970 |
return; |
| 1971 |
|
| 1972 |
/* We should have a valid XID and be at the top level. */ |
| 1973 |
Assert(TransactionIdIsValid(xid)); |
| 1974 |
|
| 1975 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 1976 |
|
| 1977 |
/* This should only be done once per transaction. */ |
| 1978 |
Assert(MySerializableXact->topXid == InvalidTransactionId); |
| 1979 |
|
| 1980 |
MySerializableXact->topXid = xid; |
| 1981 |
|
| 1982 |
sxidtag.xid = xid; |
| 1983 |
sxid = (SERIALIZABLEXID *) hash_search(SerializableXidHash, |
| 1984 |
&sxidtag, |
| 1985 |
HASH_ENTER, &found); |
| 1986 |
Assert(!found); |
| 1987 |
|
| 1988 |
/* Initialize the structure. */ |
| 1989 |
sxid->myXact = MySerializableXact; |
| 1990 |
LWLockRelease(SerializableXactHashLock); |
| 1991 |
} |
| 1992 |
|
| 1993 |
|
| 1994 |
/* |
| 1995 |
* Check whether there are any predicate locks held by any transaction |
| 1996 |
* for the page at the given block number. |
| 1997 |
* |
| 1998 |
* Note that the transaction may be completed but not yet subject to |
| 1999 |
* cleanup due to overlapping serializable transactions. This must |
| 2000 |
* return valid information regardless of transaction isolation level. |
| 2001 |
* |
| 2002 |
* Also note that this doesn't check for a conflicting relation lock, |
| 2003 |
* just a lock specifically on the given page. |
| 2004 |
* |
| 2005 |
* One use is to support proper behavior during GiST index vacuum. |
| 2006 |
*/ |
| 2007 |
bool |
| 2008 |
PageIsPredicateLocked(Relation relation, BlockNumber blkno) |
| 2009 |
{ |
| 2010 |
PREDICATELOCKTARGETTAG targettag; |
| 2011 |
uint32 targettaghash; |
| 2012 |
LWLock *partitionLock; |
| 2013 |
PREDICATELOCKTARGET *target; |
| 2014 |
|
| 2015 |
SET_PREDICATELOCKTARGETTAG_PAGE(targettag, |
| 2016 |
relation->rd_locator.dbOid, |
| 2017 |
relation->rd_id, |
| 2018 |
blkno); |
| 2019 |
|
| 2020 |
targettaghash = PredicateLockTargetTagHashCode(&targettag); |
| 2021 |
partitionLock = PredicateLockHashPartitionLock(targettaghash); |
| 2022 |
LWLockAcquire(partitionLock, LW_SHARED); |
| 2023 |
target = (PREDICATELOCKTARGET *) |
| 2024 |
hash_search_with_hash_value(PredicateLockTargetHash, |
| 2025 |
&targettag, targettaghash, |
| 2026 |
HASH_FIND, NULL); |
| 2027 |
LWLockRelease(partitionLock); |
| 2028 |
|
| 2029 |
return (target != NULL); |
| 2030 |
} |
| 2031 |
|
| 2032 |
|
| 2033 |
/* |
| 2034 |
* Check whether a particular lock is held by this transaction. |
| 2035 |
* |
| 2036 |
* Important note: this function may return false even if the lock is |
| 2037 |
* being held, because it uses the local lock table which is not |
| 2038 |
* updated if another transaction modifies our lock list (e.g. to |
| 2039 |
* split an index page). It can also return true when a coarser |
| 2040 |
* granularity lock that covers this target is being held. Be careful |
| 2041 |
* to only use this function in circumstances where such errors are |
| 2042 |
* acceptable! |
| 2043 |
*/ |
| 2044 |
static bool |
| 2045 |
PredicateLockExists(const PREDICATELOCKTARGETTAG *targettag) |
| 2046 |
{ |
| 2047 |
LOCALPREDICATELOCK *lock; |
| 2048 |
|
| 2049 |
/* check local hash table */ |
| 2050 |
lock = (LOCALPREDICATELOCK *) hash_search(LocalPredicateLockHash, |
| 2051 |
targettag, |
| 2052 |
HASH_FIND, NULL); |
| 2053 |
|
| 2054 |
if (!lock) |
| 2055 |
return false; |
| 2056 |
|
| 2057 |
/* |
| 2058 |
* Found entry in the table, but still need to check whether it's actually |
| 2059 |
* held -- it could just be a parent of some held lock. |
| 2060 |
*/ |
| 2061 |
return lock->held; |
| 2062 |
} |
| 2063 |
|
| 2064 |
/* |
| 2065 |
* Return the parent lock tag in the lock hierarchy: the next coarser |
| 2066 |
* lock that covers the provided tag. |
| 2067 |
* |
| 2068 |
* Returns true and sets *parent to the parent tag if one exists, |
| 2069 |
* returns false if none exists. |
| 2070 |
*/ |
| 2071 |
static bool |
| 2072 |
GetParentPredicateLockTag(const PREDICATELOCKTARGETTAG *tag, |
| 2073 |
PREDICATELOCKTARGETTAG *parent) |
| 2074 |
{ |
| 2075 |
switch (GET_PREDICATELOCKTARGETTAG_TYPE(*tag)) |
| 2076 |
{ |
| 2077 |
case PREDLOCKTAG_RELATION: |
| 2078 |
/* relation locks have no parent lock */ |
| 2079 |
return false; |
| 2080 |
|
| 2081 |
case PREDLOCKTAG_PAGE: |
| 2082 |
/* parent lock is relation lock */ |
| 2083 |
SET_PREDICATELOCKTARGETTAG_RELATION(*parent, |
| 2084 |
GET_PREDICATELOCKTARGETTAG_DB(*tag), |
| 2085 |
GET_PREDICATELOCKTARGETTAG_RELATION(*tag)); |
| 2086 |
|
| 2087 |
return true; |
| 2088 |
|
| 2089 |
case PREDLOCKTAG_TUPLE: |
| 2090 |
/* parent lock is page lock */ |
| 2091 |
SET_PREDICATELOCKTARGETTAG_PAGE(*parent, |
| 2092 |
GET_PREDICATELOCKTARGETTAG_DB(*tag), |
| 2093 |
GET_PREDICATELOCKTARGETTAG_RELATION(*tag), |
| 2094 |
GET_PREDICATELOCKTARGETTAG_PAGE(*tag)); |
| 2095 |
return true; |
| 2096 |
} |
| 2097 |
|
| 2098 |
/* not reachable */ |
| 2099 |
Assert(false); |
| 2100 |
return false; |
| 2101 |
} |
| 2102 |
|
| 2103 |
/* |
| 2104 |
* Check whether the lock we are considering is already covered by a |
| 2105 |
* coarser lock for our transaction. |
| 2106 |
* |
| 2107 |
* Like PredicateLockExists, this function might return a false |
| 2108 |
* negative, but it will never return a false positive. |
| 2109 |
*/ |
| 2110 |
static bool |
| 2111 |
CoarserLockCovers(const PREDICATELOCKTARGETTAG *newtargettag) |
| 2112 |
{ |
| 2113 |
PREDICATELOCKTARGETTAG targettag, |
| 2114 |
parenttag; |
| 2115 |
|
| 2116 |
targettag = *newtargettag; |
| 2117 |
|
| 2118 |
/* check parents iteratively until no more */ |
| 2119 |
while (GetParentPredicateLockTag(&targettag, &parenttag)) |
| 2120 |
{ |
| 2121 |
targettag = parenttag; |
| 2122 |
if (PredicateLockExists(&targettag)) |
| 2123 |
return true; |
| 2124 |
} |
| 2125 |
|
| 2126 |
/* no more parents to check; lock is not covered */ |
| 2127 |
return false; |
| 2128 |
} |
| 2129 |
|
| 2130 |
/* |
| 2131 |
* Remove the dummy entry from the predicate lock target hash, to free up some |
| 2132 |
* scratch space. The caller must be holding SerializablePredicateListLock, |
| 2133 |
* and must restore the entry with RestoreScratchTarget() before releasing the |
| 2134 |
* lock. |
| 2135 |
* |
| 2136 |
* If lockheld is true, the caller is already holding the partition lock |
| 2137 |
* of the partition containing the scratch entry. |
| 2138 |
*/ |
| 2139 |
static void |
| 2140 |
RemoveScratchTarget(bool lockheld) |
| 2141 |
{ |
| 2142 |
bool found; |
| 2143 |
|
| 2144 |
Assert(LWLockHeldByMe(SerializablePredicateListLock)); |
| 2145 |
|
| 2146 |
if (!lockheld) |
| 2147 |
LWLockAcquire(ScratchPartitionLock, LW_EXCLUSIVE); |
| 2148 |
hash_search_with_hash_value(PredicateLockTargetHash, |
| 2149 |
&ScratchTargetTag, |
| 2150 |
ScratchTargetTagHash, |
| 2151 |
HASH_REMOVE, &found); |
| 2152 |
Assert(found); |
| 2153 |
if (!lockheld) |
| 2154 |
LWLockRelease(ScratchPartitionLock); |
| 2155 |
} |
| 2156 |
|
| 2157 |
/* |
| 2158 |
* Re-insert the dummy entry in predicate lock target hash. |
| 2159 |
*/ |
| 2160 |
static void |
| 2161 |
RestoreScratchTarget(bool lockheld) |
| 2162 |
{ |
| 2163 |
bool found; |
| 2164 |
|
| 2165 |
Assert(LWLockHeldByMe(SerializablePredicateListLock)); |
| 2166 |
|
| 2167 |
if (!lockheld) |
| 2168 |
LWLockAcquire(ScratchPartitionLock, LW_EXCLUSIVE); |
| 2169 |
hash_search_with_hash_value(PredicateLockTargetHash, |
| 2170 |
&ScratchTargetTag, |
| 2171 |
ScratchTargetTagHash, |
| 2172 |
HASH_ENTER, &found); |
| 2173 |
Assert(!found); |
| 2174 |
if (!lockheld) |
| 2175 |
LWLockRelease(ScratchPartitionLock); |
| 2176 |
} |
| 2177 |
|
| 2178 |
/* |
| 2179 |
* Check whether the list of related predicate locks is empty for a |
| 2180 |
* predicate lock target, and remove the target if it is. |
| 2181 |
*/ |
| 2182 |
static void |
| 2183 |
RemoveTargetIfNoLongerUsed(PREDICATELOCKTARGET *target, uint32 targettaghash) |
| 2184 |
{ |
| 2185 |
PREDICATELOCKTARGET *rmtarget PG_USED_FOR_ASSERTS_ONLY; |
| 2186 |
|
| 2187 |
Assert(LWLockHeldByMe(SerializablePredicateListLock)); |
| 2188 |
|
| 2189 |
/* Can't remove it until no locks at this target. */ |
| 2190 |
if (!dlist_is_empty(&target->predicateLocks)) |
| 2191 |
return; |
| 2192 |
|
| 2193 |
/* Actually remove the target. */ |
| 2194 |
rmtarget = hash_search_with_hash_value(PredicateLockTargetHash, |
| 2195 |
&target->tag, |
| 2196 |
targettaghash, |
| 2197 |
HASH_REMOVE, NULL); |
| 2198 |
Assert(rmtarget == target); |
| 2199 |
} |
| 2200 |
|
| 2201 |
/* |
| 2202 |
* Delete child target locks owned by this process. |
| 2203 |
* This implementation is assuming that the usage of each target tag field |
| 2204 |
* is uniform. No need to make this hard if we don't have to. |
| 2205 |
* |
| 2206 |
* We acquire an LWLock in the case of parallel mode, because worker |
| 2207 |
* backends have access to the leader's SERIALIZABLEXACT. Otherwise, |
| 2208 |
* we aren't acquiring LWLocks for the predicate lock or lock |
| 2209 |
* target structures associated with this transaction unless we're going |
| 2210 |
* to modify them, because no other process is permitted to modify our |
| 2211 |
* locks. |
| 2212 |
*/ |
| 2213 |
static void |
| 2214 |
DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag) |
| 2215 |
{ |
| 2216 |
SERIALIZABLEXACT *sxact; |
| 2217 |
PREDICATELOCK *predlock; |
| 2218 |
dlist_mutable_iter iter; |
| 2219 |
|
| 2220 |
LWLockAcquire(SerializablePredicateListLock, LW_SHARED); |
| 2221 |
sxact = MySerializableXact; |
| 2222 |
if (IsInParallelMode()) |
| 2223 |
LWLockAcquire(&sxact->perXactPredicateListLock, LW_EXCLUSIVE); |
| 2224 |
|
| 2225 |
dlist_foreach_modify(iter, &sxact->predicateLocks) |
| 2226 |
{ |
| 2227 |
PREDICATELOCKTAG oldlocktag; |
| 2228 |
PREDICATELOCKTARGET *oldtarget; |
| 2229 |
PREDICATELOCKTARGETTAG oldtargettag; |
| 2230 |
|
| 2231 |
predlock = dlist_container(PREDICATELOCK, xactLink, iter.cur); |
| 2232 |
|
| 2233 |
oldlocktag = predlock->tag; |
| 2234 |
Assert(oldlocktag.myXact == sxact); |
| 2235 |
oldtarget = oldlocktag.myTarget; |
| 2236 |
oldtargettag = oldtarget->tag; |
| 2237 |
|
| 2238 |
if (TargetTagIsCoveredBy(oldtargettag, *newtargettag)) |
| 2239 |
{ |
| 2240 |
uint32 oldtargettaghash; |
| 2241 |
LWLock *partitionLock; |
| 2242 |
PREDICATELOCK *rmpredlock PG_USED_FOR_ASSERTS_ONLY; |
| 2243 |
|
| 2244 |
oldtargettaghash = PredicateLockTargetTagHashCode(&oldtargettag); |
| 2245 |
partitionLock = PredicateLockHashPartitionLock(oldtargettaghash); |
| 2246 |
|
| 2247 |
LWLockAcquire(partitionLock, LW_EXCLUSIVE); |
| 2248 |
|
| 2249 |
dlist_delete(&predlock->xactLink); |
| 2250 |
dlist_delete(&predlock->targetLink); |
| 2251 |
rmpredlock = hash_search_with_hash_value |
| 2252 |
(PredicateLockHash, |
| 2253 |
&oldlocktag, |
| 2254 |
PredicateLockHashCodeFromTargetHashCode(&oldlocktag, |
| 2255 |
oldtargettaghash), |
| 2256 |
HASH_REMOVE, NULL); |
| 2257 |
Assert(rmpredlock == predlock); |
| 2258 |
|
| 2259 |
RemoveTargetIfNoLongerUsed(oldtarget, oldtargettaghash); |
| 2260 |
|
| 2261 |
LWLockRelease(partitionLock); |
| 2262 |
|
| 2263 |
DecrementParentLocks(&oldtargettag); |
| 2264 |
} |
| 2265 |
} |
| 2266 |
if (IsInParallelMode()) |
| 2267 |
LWLockRelease(&sxact->perXactPredicateListLock); |
| 2268 |
LWLockRelease(SerializablePredicateListLock); |
| 2269 |
} |
| 2270 |
|
| 2271 |
/* |
| 2272 |
* Returns the promotion limit for a given predicate lock target. This is the |
| 2273 |
* max number of descendant locks allowed before promoting to the specified |
| 2274 |
* tag. Note that the limit includes non-direct descendants (e.g., both tuples |
| 2275 |
* and pages for a relation lock). |
| 2276 |
* |
| 2277 |
* Currently the default limit is 2 for a page lock, and half of the value of |
| 2278 |
* max_pred_locks_per_transaction - 1 for a relation lock, to match behavior |
| 2279 |
* of earlier releases when upgrading. |
| 2280 |
* |
| 2281 |
* TODO SSI: We should probably add additional GUCs to allow a maximum ratio |
| 2282 |
* of page and tuple locks based on the pages in a relation, and the maximum |
| 2283 |
* ratio of tuple locks to tuples in a page. This would provide more |
| 2284 |
* generally "balanced" allocation of locks to where they are most useful, |
| 2285 |
* while still allowing the absolute numbers to prevent one relation from |
| 2286 |
* tying up all predicate lock resources. |
| 2287 |
*/ |
| 2288 |
static int |
| 2289 |
MaxPredicateChildLocks(const PREDICATELOCKTARGETTAG *tag) |
| 2290 |
{ |
| 2291 |
switch (GET_PREDICATELOCKTARGETTAG_TYPE(*tag)) |
| 2292 |
{ |
| 2293 |
case PREDLOCKTAG_RELATION: |
| 2294 |
return max_predicate_locks_per_relation < 0 |
| 2295 |
? (max_predicate_locks_per_xact |
| 2296 |
/ (-max_predicate_locks_per_relation)) - 1 |
| 2297 |
: max_predicate_locks_per_relation; |
| 2298 |
|
| 2299 |
case PREDLOCKTAG_PAGE: |
| 2300 |
return max_predicate_locks_per_page; |
| 2301 |
|
| 2302 |
case PREDLOCKTAG_TUPLE: |
| 2303 |
|
| 2304 |
/* |
| 2305 |
* not reachable: nothing is finer-granularity than a tuple, so we |
| 2306 |
* should never try to promote to it. |
| 2307 |
*/ |
| 2308 |
Assert(false); |
| 2309 |
return 0; |
| 2310 |
} |
| 2311 |
|
| 2312 |
/* not reachable */ |
| 2313 |
Assert(false); |
| 2314 |
return 0; |
| 2315 |
} |
| 2316 |
|
| 2317 |
/* |
| 2318 |
* For all ancestors of a newly-acquired predicate lock, increment |
| 2319 |
* their child count in the parent hash table. If any of them have |
| 2320 |
* more descendants than their promotion threshold, acquire the |
| 2321 |
* coarsest such lock. |
| 2322 |
* |
| 2323 |
* Returns true if a parent lock was acquired and false otherwise. |
| 2324 |
*/ |
| 2325 |
static bool |
| 2326 |
CheckAndPromotePredicateLockRequest(const PREDICATELOCKTARGETTAG *reqtag) |
| 2327 |
{ |
| 2328 |
PREDICATELOCKTARGETTAG targettag, |
| 2329 |
nexttag, |
| 2330 |
promotiontag; |
| 2331 |
LOCALPREDICATELOCK *parentlock; |
| 2332 |
bool found, |
| 2333 |
promote; |
| 2334 |
|
| 2335 |
promote = false; |
| 2336 |
|
| 2337 |
targettag = *reqtag; |
| 2338 |
|
| 2339 |
/* check parents iteratively */ |
| 2340 |
while (GetParentPredicateLockTag(&targettag, &nexttag)) |
| 2341 |
{ |
| 2342 |
targettag = nexttag; |
| 2343 |
parentlock = (LOCALPREDICATELOCK *) hash_search(LocalPredicateLockHash, |
| 2344 |
&targettag, |
| 2345 |
HASH_ENTER, |
| 2346 |
&found); |
| 2347 |
if (!found) |
| 2348 |
{ |
| 2349 |
parentlock->held = false; |
| 2350 |
parentlock->childLocks = 1; |
| 2351 |
} |
| 2352 |
else |
| 2353 |
parentlock->childLocks++; |
| 2354 |
|
| 2355 |
if (parentlock->childLocks > |
| 2356 |
MaxPredicateChildLocks(&targettag)) |
| 2357 |
{ |
| 2358 |
/* |
| 2359 |
* We should promote to this parent lock. Continue to check its |
| 2360 |
* ancestors, however, both to get their child counts right and to |
| 2361 |
* check whether we should just go ahead and promote to one of |
| 2362 |
* them. |
| 2363 |
*/ |
| 2364 |
promotiontag = targettag; |
| 2365 |
promote = true; |
| 2366 |
} |
| 2367 |
} |
| 2368 |
|
| 2369 |
if (promote) |
| 2370 |
{ |
| 2371 |
/* acquire coarsest ancestor eligible for promotion */ |
| 2372 |
PredicateLockAcquire(&promotiontag); |
| 2373 |
return true; |
| 2374 |
} |
| 2375 |
else |
| 2376 |
return false; |
| 2377 |
} |
| 2378 |
|
| 2379 |
/* |
| 2380 |
* When releasing a lock, decrement the child count on all ancestor |
| 2381 |
* locks. |
| 2382 |
* |
| 2383 |
* This is called only when releasing a lock via |
| 2384 |
* DeleteChildTargetLocks (i.e. when a lock becomes redundant because |
| 2385 |
* we've acquired its parent, possibly due to promotion) or when a new |
| 2386 |
* MVCC write lock makes the predicate lock unnecessary. There's no |
| 2387 |
* point in calling it when locks are released at transaction end, as |
| 2388 |
* this information is no longer needed. |
| 2389 |
*/ |
| 2390 |
static void |
| 2391 |
DecrementParentLocks(const PREDICATELOCKTARGETTAG *targettag) |
| 2392 |
{ |
| 2393 |
PREDICATELOCKTARGETTAG parenttag, |
| 2394 |
nexttag; |
| 2395 |
|
| 2396 |
parenttag = *targettag; |
| 2397 |
|
| 2398 |
while (GetParentPredicateLockTag(&parenttag, &nexttag)) |
| 2399 |
{ |
| 2400 |
uint32 targettaghash; |
| 2401 |
LOCALPREDICATELOCK *parentlock, |
| 2402 |
*rmlock PG_USED_FOR_ASSERTS_ONLY; |
| 2403 |
|
| 2404 |
parenttag = nexttag; |
| 2405 |
targettaghash = PredicateLockTargetTagHashCode(&parenttag); |
| 2406 |
parentlock = (LOCALPREDICATELOCK *) |
| 2407 |
hash_search_with_hash_value(LocalPredicateLockHash, |
| 2408 |
&parenttag, targettaghash, |
| 2409 |
HASH_FIND, NULL); |
| 2410 |
|
| 2411 |
/* |
| 2412 |
* There's a small chance the parent lock doesn't exist in the lock |
| 2413 |
* table. This can happen if we prematurely removed it because an |
| 2414 |
* index split caused the child refcount to be off. |
| 2415 |
*/ |
| 2416 |
if (parentlock == NULL) |
| 2417 |
continue; |
| 2418 |
|
| 2419 |
parentlock->childLocks--; |
| 2420 |
|
| 2421 |
/* |
| 2422 |
* Under similar circumstances the parent lock's refcount might be |
| 2423 |
* zero. This only happens if we're holding that lock (otherwise we |
| 2424 |
* would have removed the entry). |
| 2425 |
*/ |
| 2426 |
if (parentlock->childLocks < 0) |
| 2427 |
{ |
| 2428 |
Assert(parentlock->held); |
| 2429 |
parentlock->childLocks = 0; |
| 2430 |
} |
| 2431 |
|
| 2432 |
if ((parentlock->childLocks == 0) && (!parentlock->held)) |
| 2433 |
{ |
| 2434 |
rmlock = (LOCALPREDICATELOCK *) |
| 2435 |
hash_search_with_hash_value(LocalPredicateLockHash, |
| 2436 |
&parenttag, targettaghash, |
| 2437 |
HASH_REMOVE, NULL); |
| 2438 |
Assert(rmlock == parentlock); |
| 2439 |
} |
| 2440 |
} |
| 2441 |
} |
| 2442 |
|
| 2443 |
/* |
| 2444 |
* Indicate that a predicate lock on the given target is held by the |
| 2445 |
* specified transaction. Has no effect if the lock is already held. |
| 2446 |
* |
| 2447 |
* This updates the lock table and the sxact's lock list, and creates |
| 2448 |
* the lock target if necessary, but does *not* do anything related to |
| 2449 |
* granularity promotion or the local lock table. See |
| 2450 |
* PredicateLockAcquire for that. |
| 2451 |
*/ |
| 2452 |
static void |
| 2453 |
CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag, |
| 2454 |
uint32 targettaghash, |
| 2455 |
SERIALIZABLEXACT *sxact) |
| 2456 |
{ |
| 2457 |
PREDICATELOCKTARGET *target; |
| 2458 |
PREDICATELOCKTAG locktag; |
| 2459 |
PREDICATELOCK *lock; |
| 2460 |
LWLock *partitionLock; |
| 2461 |
bool found; |
| 2462 |
|
| 2463 |
partitionLock = PredicateLockHashPartitionLock(targettaghash); |
| 2464 |
|
| 2465 |
LWLockAcquire(SerializablePredicateListLock, LW_SHARED); |
| 2466 |
if (IsInParallelMode()) |
| 2467 |
LWLockAcquire(&sxact->perXactPredicateListLock, LW_EXCLUSIVE); |
| 2468 |
LWLockAcquire(partitionLock, LW_EXCLUSIVE); |
| 2469 |
|
| 2470 |
/* Make sure that the target is represented. */ |
| 2471 |
target = (PREDICATELOCKTARGET *) |
| 2472 |
hash_search_with_hash_value(PredicateLockTargetHash, |
| 2473 |
targettag, targettaghash, |
| 2474 |
HASH_ENTER_NULL, &found); |
| 2475 |
if (!target) |
| 2476 |
ereport(ERROR, |
| 2477 |
(errcode(ERRCODE_OUT_OF_MEMORY), |
| 2478 |
errmsg("out of shared memory"), |
| 2479 |
errhint("You might need to increase \"%s\".", "max_pred_locks_per_transaction"))); |
| 2480 |
if (!found) |
| 2481 |
dlist_init(&target->predicateLocks); |
| 2482 |
|
| 2483 |
/* We've got the sxact and target, make sure they're joined. */ |
| 2484 |
locktag.myTarget = target; |
| 2485 |
locktag.myXact = sxact; |
| 2486 |
lock = (PREDICATELOCK *) |
| 2487 |
hash_search_with_hash_value(PredicateLockHash, &locktag, |
| 2488 |
PredicateLockHashCodeFromTargetHashCode(&locktag, targettaghash), |
| 2489 |
HASH_ENTER_NULL, &found); |
| 2490 |
if (!lock) |
| 2491 |
ereport(ERROR, |
| 2492 |
(errcode(ERRCODE_OUT_OF_MEMORY), |
| 2493 |
errmsg("out of shared memory"), |
| 2494 |
errhint("You might need to increase \"%s\".", "max_pred_locks_per_transaction"))); |
| 2495 |
|
| 2496 |
if (!found) |
| 2497 |
{ |
| 2498 |
dlist_push_tail(&target->predicateLocks, &lock->targetLink); |
| 2499 |
dlist_push_tail(&sxact->predicateLocks, &lock->xactLink); |
| 2500 |
lock->commitSeqNo = InvalidSerCommitSeqNo; |
| 2501 |
} |
| 2502 |
|
| 2503 |
LWLockRelease(partitionLock); |
| 2504 |
if (IsInParallelMode()) |
| 2505 |
LWLockRelease(&sxact->perXactPredicateListLock); |
| 2506 |
LWLockRelease(SerializablePredicateListLock); |
| 2507 |
} |
| 2508 |
|
| 2509 |
/* |
| 2510 |
* Acquire a predicate lock on the specified target for the current |
| 2511 |
* connection if not already held. This updates the local lock table |
| 2512 |
* and uses it to implement granularity promotion. It will consolidate |
| 2513 |
* multiple locks into a coarser lock if warranted, and will release |
| 2514 |
* any finer-grained locks covered by the new one. |
| 2515 |
*/ |
| 2516 |
static void |
| 2517 |
PredicateLockAcquire(const PREDICATELOCKTARGETTAG *targettag) |
| 2518 |
{ |
| 2519 |
uint32 targettaghash; |
| 2520 |
bool found; |
| 2521 |
LOCALPREDICATELOCK *locallock; |
| 2522 |
|
| 2523 |
/* Do we have the lock already, or a covering lock? */ |
| 2524 |
if (PredicateLockExists(targettag)) |
| 2525 |
return; |
| 2526 |
|
| 2527 |
if (CoarserLockCovers(targettag)) |
| 2528 |
return; |
| 2529 |
|
| 2530 |
/* the same hash and LW lock apply to the lock target and the local lock. */ |
| 2531 |
targettaghash = PredicateLockTargetTagHashCode(targettag); |
| 2532 |
|
| 2533 |
/* Acquire lock in local table */ |
| 2534 |
locallock = (LOCALPREDICATELOCK *) |
| 2535 |
hash_search_with_hash_value(LocalPredicateLockHash, |
| 2536 |
targettag, targettaghash, |
| 2537 |
HASH_ENTER, &found); |
| 2538 |
locallock->held = true; |
| 2539 |
if (!found) |
| 2540 |
locallock->childLocks = 0; |
| 2541 |
|
| 2542 |
/* Actually create the lock */ |
| 2543 |
CreatePredicateLock(targettag, targettaghash, MySerializableXact); |
| 2544 |
|
| 2545 |
/* |
| 2546 |
* Lock has been acquired. Check whether it should be promoted to a |
| 2547 |
* coarser granularity, or whether there are finer-granularity locks to |
| 2548 |
* clean up. |
| 2549 |
*/ |
| 2550 |
if (CheckAndPromotePredicateLockRequest(targettag)) |
| 2551 |
{ |
| 2552 |
/* |
| 2553 |
* Lock request was promoted to a coarser-granularity lock, and that |
| 2554 |
* lock was acquired. It will delete this lock and any of its |
| 2555 |
* children, so we're done. |
| 2556 |
*/ |
| 2557 |
} |
| 2558 |
else |
| 2559 |
{ |
| 2560 |
/* Clean up any finer-granularity locks */ |
| 2561 |
if (GET_PREDICATELOCKTARGETTAG_TYPE(*targettag) != PREDLOCKTAG_TUPLE) |
| 2562 |
DeleteChildTargetLocks(targettag); |
| 2563 |
} |
| 2564 |
} |
| 2565 |
|
| 2566 |
|
| 2567 |
/* |
| 2568 |
* PredicateLockRelation |
| 2569 |
* |
| 2570 |
* Gets a predicate lock at the relation level. |
| 2571 |
* Skip if not in full serializable transaction isolation level. |
| 2572 |
* Skip if this is a temporary table. |
| 2573 |
* Clear any finer-grained predicate locks this session has on the relation. |
| 2574 |
*/ |
| 2575 |
void |
| 2576 |
PredicateLockRelation(Relation relation, Snapshot snapshot) |
| 2577 |
{ |
| 2578 |
PREDICATELOCKTARGETTAG tag; |
| 2579 |
|
| 2580 |
if (!SerializationNeededForRead(relation, snapshot)) |
| 2581 |
return; |
| 2582 |
|
| 2583 |
SET_PREDICATELOCKTARGETTAG_RELATION(tag, |
| 2584 |
relation->rd_locator.dbOid, |
| 2585 |
relation->rd_id); |
| 2586 |
PredicateLockAcquire(&tag); |
| 2587 |
} |
| 2588 |
|
| 2589 |
/* |
| 2590 |
* PredicateLockPage |
| 2591 |
* |
| 2592 |
* Gets a predicate lock at the page level. |
| 2593 |
* Skip if not in full serializable transaction isolation level. |
| 2594 |
* Skip if this is a temporary table. |
| 2595 |
* Skip if a coarser predicate lock already covers this page. |
| 2596 |
* Clear any finer-grained predicate locks this session has on the relation. |
| 2597 |
*/ |
| 2598 |
void |
| 2599 |
PredicateLockPage(Relation relation, BlockNumber blkno, Snapshot snapshot) |
| 2600 |
{ |
| 2601 |
PREDICATELOCKTARGETTAG tag; |
| 2602 |
|
| 2603 |
if (!SerializationNeededForRead(relation, snapshot)) |
| 2604 |
return; |
| 2605 |
|
| 2606 |
SET_PREDICATELOCKTARGETTAG_PAGE(tag, |
| 2607 |
relation->rd_locator.dbOid, |
| 2608 |
relation->rd_id, |
| 2609 |
blkno); |
| 2610 |
PredicateLockAcquire(&tag); |
| 2611 |
} |
| 2612 |
|
| 2613 |
/* |
| 2614 |
* PredicateLockTID |
| 2615 |
* |
| 2616 |
* Gets a predicate lock at the tuple level. |
| 2617 |
* Skip if not in full serializable transaction isolation level. |
| 2618 |
* Skip if this is a temporary table. |
| 2619 |
*/ |
| 2620 |
void |
| 2621 |
PredicateLockTID(Relation relation, ItemPointer tid, Snapshot snapshot, |
| 2622 |
TransactionId tuple_xid) |
| 2623 |
{ |
| 2624 |
PREDICATELOCKTARGETTAG tag; |
| 2625 |
|
| 2626 |
if (!SerializationNeededForRead(relation, snapshot)) |
| 2627 |
return; |
| 2628 |
|
| 2629 |
/* |
| 2630 |
* Return if this xact wrote it. |
| 2631 |
*/ |
| 2632 |
if (relation->rd_index == NULL) |
| 2633 |
{ |
| 2634 |
/* If we wrote it; we already have a write lock. */ |
| 2635 |
if (TransactionIdIsCurrentTransactionId(tuple_xid)) |
| 2636 |
return; |
| 2637 |
} |
| 2638 |
|
| 2639 |
/* |
| 2640 |
* Do quick-but-not-definitive test for a relation lock first. This will |
| 2641 |
* never cause a return when the relation is *not* locked, but will |
| 2642 |
* occasionally let the check continue when there really *is* a relation |
| 2643 |
* level lock. |
| 2644 |
*/ |
| 2645 |
SET_PREDICATELOCKTARGETTAG_RELATION(tag, |
| 2646 |
relation->rd_locator.dbOid, |
| 2647 |
relation->rd_id); |
| 2648 |
if (PredicateLockExists(&tag)) |
| 2649 |
return; |
| 2650 |
|
| 2651 |
SET_PREDICATELOCKTARGETTAG_TUPLE(tag, |
| 2652 |
relation->rd_locator.dbOid, |
| 2653 |
relation->rd_id, |
| 2654 |
ItemPointerGetBlockNumber(tid), |
| 2655 |
ItemPointerGetOffsetNumber(tid)); |
| 2656 |
PredicateLockAcquire(&tag); |
| 2657 |
} |
| 2658 |
|
| 2659 |
|
| 2660 |
/* |
| 2661 |
* DeleteLockTarget |
| 2662 |
* |
| 2663 |
* Remove a predicate lock target along with any locks held for it. |
| 2664 |
* |
| 2665 |
* Caller must hold SerializablePredicateListLock and the |
| 2666 |
* appropriate hash partition lock for the target. |
| 2667 |
*/ |
| 2668 |
static void |
| 2669 |
DeleteLockTarget(PREDICATELOCKTARGET *target, uint32 targettaghash) |
| 2670 |
{ |
| 2671 |
dlist_mutable_iter iter; |
| 2672 |
|
| 2673 |
Assert(LWLockHeldByMeInMode(SerializablePredicateListLock, |
| 2674 |
LW_EXCLUSIVE)); |
| 2675 |
Assert(LWLockHeldByMe(PredicateLockHashPartitionLock(targettaghash))); |
| 2676 |
|
| 2677 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 2678 |
|
| 2679 |
dlist_foreach_modify(iter, &target->predicateLocks) |
| 2680 |
{ |
| 2681 |
PREDICATELOCK *predlock = |
| 2682 |
dlist_container(PREDICATELOCK, targetLink, iter.cur); |
| 2683 |
bool found; |
| 2684 |
|
| 2685 |
dlist_delete(&(predlock->xactLink)); |
| 2686 |
dlist_delete(&(predlock->targetLink)); |
| 2687 |
|
| 2688 |
hash_search_with_hash_value |
| 2689 |
(PredicateLockHash, |
| 2690 |
&predlock->tag, |
| 2691 |
PredicateLockHashCodeFromTargetHashCode(&predlock->tag, |
| 2692 |
targettaghash), |
| 2693 |
HASH_REMOVE, &found); |
| 2694 |
Assert(found); |
| 2695 |
} |
| 2696 |
LWLockRelease(SerializableXactHashLock); |
| 2697 |
|
| 2698 |
/* Remove the target itself, if possible. */ |
| 2699 |
RemoveTargetIfNoLongerUsed(target, targettaghash); |
| 2700 |
} |
| 2701 |
|
| 2702 |
|
| 2703 |
/* |
| 2704 |
* TransferPredicateLocksToNewTarget |
| 2705 |
* |
| 2706 |
* Move or copy all the predicate locks for a lock target, for use by |
| 2707 |
* index page splits/combines and other things that create or replace |
| 2708 |
* lock targets. If 'removeOld' is true, the old locks and the target |
| 2709 |
* will be removed. |
| 2710 |
* |
| 2711 |
* Returns true on success, or false if we ran out of shared memory to |
| 2712 |
* allocate the new target or locks. Guaranteed to always succeed if |
| 2713 |
* removeOld is set (by using the scratch entry in PredicateLockTargetHash |
| 2714 |
* for scratch space). |
| 2715 |
* |
| 2716 |
* Warning: the "removeOld" option should be used only with care, |
| 2717 |
* because this function does not (indeed, can not) update other |
| 2718 |
* backends' LocalPredicateLockHash. If we are only adding new |
| 2719 |
* entries, this is not a problem: the local lock table is used only |
| 2720 |
* as a hint, so missing entries for locks that are held are |
| 2721 |
* OK. Having entries for locks that are no longer held, as can happen |
| 2722 |
* when using "removeOld", is not in general OK. We can only use it |
| 2723 |
* safely when replacing a lock with a coarser-granularity lock that |
| 2724 |
* covers it, or if we are absolutely certain that no one will need to |
| 2725 |
* refer to that lock in the future. |
| 2726 |
* |
| 2727 |
* Caller must hold SerializablePredicateListLock exclusively. |
| 2728 |
*/ |
| 2729 |
static bool |
| 2730 |
TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag, |
| 2731 |
PREDICATELOCKTARGETTAG newtargettag, |
| 2732 |
bool removeOld) |
| 2733 |
{ |
| 2734 |
uint32 oldtargettaghash; |
| 2735 |
LWLock *oldpartitionLock; |
| 2736 |
PREDICATELOCKTARGET *oldtarget; |
| 2737 |
uint32 newtargettaghash; |
| 2738 |
LWLock *newpartitionLock; |
| 2739 |
bool found; |
| 2740 |
bool outOfShmem = false; |
| 2741 |
|
| 2742 |
Assert(LWLockHeldByMeInMode(SerializablePredicateListLock, |
| 2743 |
LW_EXCLUSIVE)); |
| 2744 |
|
| 2745 |
oldtargettaghash = PredicateLockTargetTagHashCode(&oldtargettag); |
| 2746 |
newtargettaghash = PredicateLockTargetTagHashCode(&newtargettag); |
| 2747 |
oldpartitionLock = PredicateLockHashPartitionLock(oldtargettaghash); |
| 2748 |
newpartitionLock = PredicateLockHashPartitionLock(newtargettaghash); |
| 2749 |
|
| 2750 |
if (removeOld) |
| 2751 |
{ |
| 2752 |
/* |
| 2753 |
* Remove the dummy entry to give us scratch space, so we know we'll |
| 2754 |
* be able to create the new lock target. |
| 2755 |
*/ |
| 2756 |
RemoveScratchTarget(false); |
| 2757 |
} |
| 2758 |
|
| 2759 |
/* |
| 2760 |
* We must get the partition locks in ascending sequence to avoid |
| 2761 |
* deadlocks. If old and new partitions are the same, we must request the |
| 2762 |
* lock only once. |
| 2763 |
*/ |
| 2764 |
if (oldpartitionLock < newpartitionLock) |
| 2765 |
{ |
| 2766 |
LWLockAcquire(oldpartitionLock, |
| 2767 |
(removeOld ? LW_EXCLUSIVE : LW_SHARED)); |
| 2768 |
LWLockAcquire(newpartitionLock, LW_EXCLUSIVE); |
| 2769 |
} |
| 2770 |
else if (oldpartitionLock > newpartitionLock) |
| 2771 |
{ |
| 2772 |
LWLockAcquire(newpartitionLock, LW_EXCLUSIVE); |
| 2773 |
LWLockAcquire(oldpartitionLock, |
| 2774 |
(removeOld ? LW_EXCLUSIVE : LW_SHARED)); |
| 2775 |
} |
| 2776 |
else |
| 2777 |
LWLockAcquire(newpartitionLock, LW_EXCLUSIVE); |
| 2778 |
|
| 2779 |
/* |
| 2780 |
* Look for the old target. If not found, that's OK; no predicate locks |
| 2781 |
* are affected, so we can just clean up and return. If it does exist, |
| 2782 |
* walk its list of predicate locks and move or copy them to the new |
| 2783 |
* target. |
| 2784 |
*/ |
| 2785 |
oldtarget = hash_search_with_hash_value(PredicateLockTargetHash, |
| 2786 |
&oldtargettag, |
| 2787 |
oldtargettaghash, |
| 2788 |
HASH_FIND, NULL); |
| 2789 |
|
| 2790 |
if (oldtarget) |
| 2791 |
{ |
| 2792 |
PREDICATELOCKTARGET *newtarget; |
| 2793 |
PREDICATELOCKTAG newpredlocktag; |
| 2794 |
dlist_mutable_iter iter; |
| 2795 |
|
| 2796 |
newtarget = hash_search_with_hash_value(PredicateLockTargetHash, |
| 2797 |
&newtargettag, |
| 2798 |
newtargettaghash, |
| 2799 |
HASH_ENTER_NULL, &found); |
| 2800 |
|
| 2801 |
if (!newtarget) |
| 2802 |
{ |
| 2803 |
/* Failed to allocate due to insufficient shmem */ |
| 2804 |
outOfShmem = true; |
| 2805 |
goto exit; |
| 2806 |
} |
| 2807 |
|
| 2808 |
/* If we created a new entry, initialize it */ |
| 2809 |
if (!found) |
| 2810 |
dlist_init(&newtarget->predicateLocks); |
| 2811 |
|
| 2812 |
newpredlocktag.myTarget = newtarget; |
| 2813 |
|
| 2814 |
/* |
| 2815 |
* Loop through all the locks on the old target, replacing them with |
| 2816 |
* locks on the new target. |
| 2817 |
*/ |
| 2818 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 2819 |
|
| 2820 |
dlist_foreach_modify(iter, &oldtarget->predicateLocks) |
| 2821 |
{ |
| 2822 |
PREDICATELOCK *oldpredlock = |
| 2823 |
dlist_container(PREDICATELOCK, targetLink, iter.cur); |
| 2824 |
PREDICATELOCK *newpredlock; |
| 2825 |
SerCommitSeqNo oldCommitSeqNo = oldpredlock->commitSeqNo; |
| 2826 |
|
| 2827 |
newpredlocktag.myXact = oldpredlock->tag.myXact; |
| 2828 |
|
| 2829 |
if (removeOld) |
| 2830 |
{ |
| 2831 |
dlist_delete(&(oldpredlock->xactLink)); |
| 2832 |
dlist_delete(&(oldpredlock->targetLink)); |
| 2833 |
|
| 2834 |
hash_search_with_hash_value |
| 2835 |
(PredicateLockHash, |
| 2836 |
&oldpredlock->tag, |
| 2837 |
PredicateLockHashCodeFromTargetHashCode(&oldpredlock->tag, |
| 2838 |
oldtargettaghash), |
| 2839 |
HASH_REMOVE, &found); |
| 2840 |
Assert(found); |
| 2841 |
} |
| 2842 |
|
| 2843 |
newpredlock = (PREDICATELOCK *) |
| 2844 |
hash_search_with_hash_value(PredicateLockHash, |
| 2845 |
&newpredlocktag, |
| 2846 |
PredicateLockHashCodeFromTargetHashCode(&newpredlocktag, |
| 2847 |
newtargettaghash), |
| 2848 |
HASH_ENTER_NULL, |
| 2849 |
&found); |
| 2850 |
if (!newpredlock) |
| 2851 |
{ |
| 2852 |
/* Out of shared memory. Undo what we've done so far. */ |
| 2853 |
LWLockRelease(SerializableXactHashLock); |
| 2854 |
DeleteLockTarget(newtarget, newtargettaghash); |
| 2855 |
outOfShmem = true; |
| 2856 |
goto exit; |
| 2857 |
} |
| 2858 |
if (!found) |
| 2859 |
{ |
| 2860 |
dlist_push_tail(&(newtarget->predicateLocks), |
| 2861 |
&(newpredlock->targetLink)); |
| 2862 |
dlist_push_tail(&(newpredlocktag.myXact->predicateLocks), |
| 2863 |
&(newpredlock->xactLink)); |
| 2864 |
newpredlock->commitSeqNo = oldCommitSeqNo; |
| 2865 |
} |
| 2866 |
else |
| 2867 |
{ |
| 2868 |
if (newpredlock->commitSeqNo < oldCommitSeqNo) |
| 2869 |
newpredlock->commitSeqNo = oldCommitSeqNo; |
| 2870 |
} |
| 2871 |
|
| 2872 |
Assert(newpredlock->commitSeqNo != 0); |
| 2873 |
Assert((newpredlock->commitSeqNo == InvalidSerCommitSeqNo) |
| 2874 |
|| (newpredlock->tag.myXact == OldCommittedSxact)); |
| 2875 |
} |
| 2876 |
LWLockRelease(SerializableXactHashLock); |
| 2877 |
|
| 2878 |
if (removeOld) |
| 2879 |
{ |
| 2880 |
Assert(dlist_is_empty(&oldtarget->predicateLocks)); |
| 2881 |
RemoveTargetIfNoLongerUsed(oldtarget, oldtargettaghash); |
| 2882 |
} |
| 2883 |
} |
| 2884 |
|
| 2885 |
|
| 2886 |
exit: |
| 2887 |
/* Release partition locks in reverse order of acquisition. */ |
| 2888 |
if (oldpartitionLock < newpartitionLock) |
| 2889 |
{ |
| 2890 |
LWLockRelease(newpartitionLock); |
| 2891 |
LWLockRelease(oldpartitionLock); |
| 2892 |
} |
| 2893 |
else if (oldpartitionLock > newpartitionLock) |
| 2894 |
{ |
| 2895 |
LWLockRelease(oldpartitionLock); |
| 2896 |
LWLockRelease(newpartitionLock); |
| 2897 |
} |
| 2898 |
else |
| 2899 |
LWLockRelease(newpartitionLock); |
| 2900 |
|
| 2901 |
if (removeOld) |
| 2902 |
{ |
| 2903 |
/* We shouldn't run out of memory if we're moving locks */ |
| 2904 |
Assert(!outOfShmem); |
| 2905 |
|
| 2906 |
/* Put the scratch entry back */ |
| 2907 |
RestoreScratchTarget(false); |
| 2908 |
} |
| 2909 |
|
| 2910 |
return !outOfShmem; |
| 2911 |
} |
| 2912 |
|
| 2913 |
/* |
| 2914 |
* Drop all predicate locks of any granularity from the specified relation, |
| 2915 |
* which can be a heap relation or an index relation. If 'transfer' is true, |
| 2916 |
* acquire a relation lock on the heap for any transactions with any lock(s) |
| 2917 |
* on the specified relation. |
| 2918 |
* |
| 2919 |
* This requires grabbing a lot of LW locks and scanning the entire lock |
| 2920 |
* target table for matches. That makes this more expensive than most |
| 2921 |
* predicate lock management functions, but it will only be called for DDL |
| 2922 |
* type commands that are expensive anyway, and there are fast returns when |
| 2923 |
* no serializable transactions are active or the relation is temporary. |
| 2924 |
* |
| 2925 |
* We don't use the TransferPredicateLocksToNewTarget function because it |
| 2926 |
* acquires its own locks on the partitions of the two targets involved, |
| 2927 |
* and we'll already be holding all partition locks. |
| 2928 |
* |
| 2929 |
* We can't throw an error from here, because the call could be from a |
| 2930 |
* transaction which is not serializable. |
| 2931 |
* |
| 2932 |
* NOTE: This is currently only called with transfer set to true, but that may |
| 2933 |
* change. If we decide to clean up the locks from a table on commit of a |
| 2934 |
* transaction which executed DROP TABLE, the false condition will be useful. |
| 2935 |
*/ |
| 2936 |
static void |
| 2937 |
DropAllPredicateLocksFromTable(Relation relation, bool transfer) |
| 2938 |
{ |
| 2939 |
HASH_SEQ_STATUS seqstat; |
| 2940 |
PREDICATELOCKTARGET *oldtarget; |
| 2941 |
PREDICATELOCKTARGET *heaptarget; |
| 2942 |
Oid dbId; |
| 2943 |
Oid relId; |
| 2944 |
Oid heapId; |
| 2945 |
int i; |
| 2946 |
bool isIndex; |
| 2947 |
bool found; |
| 2948 |
uint32 heaptargettaghash; |
| 2949 |
|
| 2950 |
/* |
| 2951 |
* Bail out quickly if there are no serializable transactions running. |
| 2952 |
* It's safe to check this without taking locks because the caller is |
| 2953 |
* holding an ACCESS EXCLUSIVE lock on the relation. No new locks which |
| 2954 |
* would matter here can be acquired while that is held. |
| 2955 |
*/ |
| 2956 |
if (!TransactionIdIsValid(PredXact->SxactGlobalXmin)) |
| 2957 |
return; |
| 2958 |
|
| 2959 |
if (!PredicateLockingNeededForRelation(relation)) |
| 2960 |
return; |
| 2961 |
|
| 2962 |
dbId = relation->rd_locator.dbOid; |
| 2963 |
relId = relation->rd_id; |
| 2964 |
if (relation->rd_index == NULL) |
| 2965 |
{ |
| 2966 |
isIndex = false; |
| 2967 |
heapId = relId; |
| 2968 |
} |
| 2969 |
else |
| 2970 |
{ |
| 2971 |
isIndex = true; |
| 2972 |
heapId = relation->rd_index->indrelid; |
| 2973 |
} |
| 2974 |
Assert(heapId != InvalidOid); |
| 2975 |
Assert(transfer || !isIndex); /* index OID only makes sense with |
| 2976 |
* transfer */ |
| 2977 |
|
| 2978 |
/* Retrieve first time needed, then keep. */ |
| 2979 |
heaptargettaghash = 0; |
| 2980 |
heaptarget = NULL; |
| 2981 |
|
| 2982 |
/* Acquire locks on all lock partitions */ |
| 2983 |
LWLockAcquire(SerializablePredicateListLock, LW_EXCLUSIVE); |
| 2984 |
for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++) |
| 2985 |
LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_EXCLUSIVE); |
| 2986 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 2987 |
|
| 2988 |
/* |
| 2989 |
* Remove the dummy entry to give us scratch space, so we know we'll be |
| 2990 |
* able to create the new lock target. |
| 2991 |
*/ |
| 2992 |
if (transfer) |
| 2993 |
RemoveScratchTarget(true); |
| 2994 |
|
| 2995 |
/* Scan through target map */ |
| 2996 |
hash_seq_init(&seqstat, PredicateLockTargetHash); |
| 2997 |
|
| 2998 |
while ((oldtarget = (PREDICATELOCKTARGET *) hash_seq_search(&seqstat))) |
| 2999 |
{ |
| 3000 |
dlist_mutable_iter iter; |
| 3001 |
|
| 3002 |
/* |
| 3003 |
* Check whether this is a target which needs attention. |
| 3004 |
*/ |
| 3005 |
if (GET_PREDICATELOCKTARGETTAG_RELATION(oldtarget->tag) != relId) |
| 3006 |
continue; /* wrong relation id */ |
| 3007 |
if (GET_PREDICATELOCKTARGETTAG_DB(oldtarget->tag) != dbId) |
| 3008 |
continue; /* wrong database id */ |
| 3009 |
if (transfer && !isIndex |
| 3010 |
&& GET_PREDICATELOCKTARGETTAG_TYPE(oldtarget->tag) == PREDLOCKTAG_RELATION) |
| 3011 |
continue; /* already the right lock */ |
| 3012 |
|
| 3013 |
/* |
| 3014 |
* If we made it here, we have work to do. We make sure the heap |
| 3015 |
* relation lock exists, then we walk the list of predicate locks for |
| 3016 |
* the old target we found, moving all locks to the heap relation lock |
| 3017 |
* -- unless they already hold that. |
| 3018 |
*/ |
| 3019 |
|
| 3020 |
/* |
| 3021 |
* First make sure we have the heap relation target. We only need to |
| 3022 |
* do this once. |
| 3023 |
*/ |
| 3024 |
if (transfer && heaptarget == NULL) |
| 3025 |
{ |
| 3026 |
PREDICATELOCKTARGETTAG heaptargettag; |
| 3027 |
|
| 3028 |
SET_PREDICATELOCKTARGETTAG_RELATION(heaptargettag, dbId, heapId); |
| 3029 |
heaptargettaghash = PredicateLockTargetTagHashCode(&heaptargettag); |
| 3030 |
heaptarget = hash_search_with_hash_value(PredicateLockTargetHash, |
| 3031 |
&heaptargettag, |
| 3032 |
heaptargettaghash, |
| 3033 |
HASH_ENTER, &found); |
| 3034 |
if (!found) |
| 3035 |
dlist_init(&heaptarget->predicateLocks); |
| 3036 |
} |
| 3037 |
|
| 3038 |
/* |
| 3039 |
* Loop through all the locks on the old target, replacing them with |
| 3040 |
* locks on the new target. |
| 3041 |
*/ |
| 3042 |
dlist_foreach_modify(iter, &oldtarget->predicateLocks) |
| 3043 |
{ |
| 3044 |
PREDICATELOCK *oldpredlock = |
| 3045 |
dlist_container(PREDICATELOCK, targetLink, iter.cur); |
| 3046 |
PREDICATELOCK *newpredlock; |
| 3047 |
SerCommitSeqNo oldCommitSeqNo; |
| 3048 |
SERIALIZABLEXACT *oldXact; |
| 3049 |
|
| 3050 |
/* |
| 3051 |
* Remove the old lock first. This avoids the chance of running |
| 3052 |
* out of lock structure entries for the hash table. |
| 3053 |
*/ |
| 3054 |
oldCommitSeqNo = oldpredlock->commitSeqNo; |
| 3055 |
oldXact = oldpredlock->tag.myXact; |
| 3056 |
|
| 3057 |
dlist_delete(&(oldpredlock->xactLink)); |
| 3058 |
|
| 3059 |
/* |
| 3060 |
* No need for retail delete from oldtarget list, we're removing |
| 3061 |
* the whole target anyway. |
| 3062 |
*/ |
| 3063 |
hash_search(PredicateLockHash, |
| 3064 |
&oldpredlock->tag, |
| 3065 |
HASH_REMOVE, &found); |
| 3066 |
Assert(found); |
| 3067 |
|
| 3068 |
if (transfer) |
| 3069 |
{ |
| 3070 |
PREDICATELOCKTAG newpredlocktag; |
| 3071 |
|
| 3072 |
newpredlocktag.myTarget = heaptarget; |
| 3073 |
newpredlocktag.myXact = oldXact; |
| 3074 |
newpredlock = (PREDICATELOCK *) |
| 3075 |
hash_search_with_hash_value(PredicateLockHash, |
| 3076 |
&newpredlocktag, |
| 3077 |
PredicateLockHashCodeFromTargetHashCode(&newpredlocktag, |
| 3078 |
heaptargettaghash), |
| 3079 |
HASH_ENTER, |
| 3080 |
&found); |
| 3081 |
if (!found) |
| 3082 |
{ |
| 3083 |
dlist_push_tail(&(heaptarget->predicateLocks), |
| 3084 |
&(newpredlock->targetLink)); |
| 3085 |
dlist_push_tail(&(newpredlocktag.myXact->predicateLocks), |
| 3086 |
&(newpredlock->xactLink)); |
| 3087 |
newpredlock->commitSeqNo = oldCommitSeqNo; |
| 3088 |
} |
| 3089 |
else |
| 3090 |
{ |
| 3091 |
if (newpredlock->commitSeqNo < oldCommitSeqNo) |
| 3092 |
newpredlock->commitSeqNo = oldCommitSeqNo; |
| 3093 |
} |
| 3094 |
|
| 3095 |
Assert(newpredlock->commitSeqNo != 0); |
| 3096 |
Assert((newpredlock->commitSeqNo == InvalidSerCommitSeqNo) |
| 3097 |
|| (newpredlock->tag.myXact == OldCommittedSxact)); |
| 3098 |
} |
| 3099 |
} |
| 3100 |
|
| 3101 |
hash_search(PredicateLockTargetHash, &oldtarget->tag, HASH_REMOVE, |
| 3102 |
&found); |
| 3103 |
Assert(found); |
| 3104 |
} |
| 3105 |
|
| 3106 |
/* Put the scratch entry back */ |
| 3107 |
if (transfer) |
| 3108 |
RestoreScratchTarget(true); |
| 3109 |
|
| 3110 |
/* Release locks in reverse order */ |
| 3111 |
LWLockRelease(SerializableXactHashLock); |
| 3112 |
for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--) |
| 3113 |
LWLockRelease(PredicateLockHashPartitionLockByIndex(i)); |
| 3114 |
LWLockRelease(SerializablePredicateListLock); |
| 3115 |
} |
| 3116 |
|
| 3117 |
/* |
| 3118 |
* TransferPredicateLocksToHeapRelation |
| 3119 |
* For all transactions, transfer all predicate locks for the given |
| 3120 |
* relation to a single relation lock on the heap. |
| 3121 |
*/ |
| 3122 |
void |
| 3123 |
TransferPredicateLocksToHeapRelation(Relation relation) |
| 3124 |
{ |
| 3125 |
DropAllPredicateLocksFromTable(relation, true); |
| 3126 |
} |
| 3127 |
|
| 3128 |
|
| 3129 |
/* |
| 3130 |
* PredicateLockPageSplit |
| 3131 |
* |
| 3132 |
* Copies any predicate locks for the old page to the new page. |
| 3133 |
* Skip if this is a temporary table or toast table. |
| 3134 |
* |
| 3135 |
* NOTE: A page split (or overflow) affects all serializable transactions, |
| 3136 |
* even if it occurs in the context of another transaction isolation level. |
| 3137 |
* |
| 3138 |
* NOTE: This currently leaves the local copy of the locks without |
| 3139 |
* information on the new lock which is in shared memory. This could cause |
| 3140 |
* problems if enough page splits occur on locked pages without the processes |
| 3141 |
* which hold the locks getting in and noticing. |
| 3142 |
*/ |
| 3143 |
void |
| 3144 |
PredicateLockPageSplit(Relation relation, BlockNumber oldblkno, |
| 3145 |
BlockNumber newblkno) |
| 3146 |
{ |
| 3147 |
PREDICATELOCKTARGETTAG oldtargettag; |
| 3148 |
PREDICATELOCKTARGETTAG newtargettag; |
| 3149 |
bool success; |
| 3150 |
|
| 3151 |
/* |
| 3152 |
* Bail out quickly if there are no serializable transactions running. |
| 3153 |
* |
| 3154 |
* It's safe to do this check without taking any additional locks. Even if |
| 3155 |
* a serializable transaction starts concurrently, we know it can't take |
| 3156 |
* any SIREAD locks on the page being split because the caller is holding |
| 3157 |
* the associated buffer page lock. Memory reordering isn't an issue; the |
| 3158 |
* memory barrier in the LWLock acquisition guarantees that this read |
| 3159 |
* occurs while the buffer page lock is held. |
| 3160 |
*/ |
| 3161 |
if (!TransactionIdIsValid(PredXact->SxactGlobalXmin)) |
| 3162 |
return; |
| 3163 |
|
| 3164 |
if (!PredicateLockingNeededForRelation(relation)) |
| 3165 |
return; |
| 3166 |
|
| 3167 |
Assert(oldblkno != newblkno); |
| 3168 |
Assert(BlockNumberIsValid(oldblkno)); |
| 3169 |
Assert(BlockNumberIsValid(newblkno)); |
| 3170 |
|
| 3171 |
SET_PREDICATELOCKTARGETTAG_PAGE(oldtargettag, |
| 3172 |
relation->rd_locator.dbOid, |
| 3173 |
relation->rd_id, |
| 3174 |
oldblkno); |
| 3175 |
SET_PREDICATELOCKTARGETTAG_PAGE(newtargettag, |
| 3176 |
relation->rd_locator.dbOid, |
| 3177 |
relation->rd_id, |
| 3178 |
newblkno); |
| 3179 |
|
| 3180 |
LWLockAcquire(SerializablePredicateListLock, LW_EXCLUSIVE); |
| 3181 |
|
| 3182 |
/* |
| 3183 |
* Try copying the locks over to the new page's tag, creating it if |
| 3184 |
* necessary. |
| 3185 |
*/ |
| 3186 |
success = TransferPredicateLocksToNewTarget(oldtargettag, |
| 3187 |
newtargettag, |
| 3188 |
false); |
| 3189 |
|
| 3190 |
if (!success) |
| 3191 |
{ |
| 3192 |
/* |
| 3193 |
* No more predicate lock entries are available. Failure isn't an |
| 3194 |
* option here, so promote the page lock to a relation lock. |
| 3195 |
*/ |
| 3196 |
|
| 3197 |
/* Get the parent relation lock's lock tag */ |
| 3198 |
success = GetParentPredicateLockTag(&oldtargettag, |
| 3199 |
&newtargettag); |
| 3200 |
Assert(success); |
| 3201 |
|
| 3202 |
/* |
| 3203 |
* Move the locks to the parent. This shouldn't fail. |
| 3204 |
* |
| 3205 |
* Note that here we are removing locks held by other backends, |
| 3206 |
* leading to a possible inconsistency in their local lock hash table. |
| 3207 |
* This is OK because we're replacing it with a lock that covers the |
| 3208 |
* old one. |
| 3209 |
*/ |
| 3210 |
success = TransferPredicateLocksToNewTarget(oldtargettag, |
| 3211 |
newtargettag, |
| 3212 |
true); |
| 3213 |
Assert(success); |
| 3214 |
} |
| 3215 |
|
| 3216 |
LWLockRelease(SerializablePredicateListLock); |
| 3217 |
} |
| 3218 |
|
| 3219 |
/* |
| 3220 |
* PredicateLockPageCombine |
| 3221 |
* |
| 3222 |
* Combines predicate locks for two existing pages. |
| 3223 |
* Skip if this is a temporary table or toast table. |
| 3224 |
* |
| 3225 |
* NOTE: A page combine affects all serializable transactions, even if it |
| 3226 |
* occurs in the context of another transaction isolation level. |
| 3227 |
*/ |
| 3228 |
void |
| 3229 |
PredicateLockPageCombine(Relation relation, BlockNumber oldblkno, |
| 3230 |
BlockNumber newblkno) |
| 3231 |
{ |
| 3232 |
/* |
| 3233 |
* Page combines differ from page splits in that we ought to be able to |
| 3234 |
* remove the locks on the old page after transferring them to the new |
| 3235 |
* page, instead of duplicating them. However, because we can't edit other |
| 3236 |
* backends' local lock tables, removing the old lock would leave them |
| 3237 |
* with an entry in their LocalPredicateLockHash for a lock they're not |
| 3238 |
* holding, which isn't acceptable. So we wind up having to do the same |
| 3239 |
* work as a page split, acquiring a lock on the new page and keeping the |
| 3240 |
* old page locked too. That can lead to some false positives, but should |
| 3241 |
* be rare in practice. |
| 3242 |
*/ |
| 3243 |
PredicateLockPageSplit(relation, oldblkno, newblkno); |
| 3244 |
} |
| 3245 |
|
| 3246 |
/* |
| 3247 |
* Walk the list of in-progress serializable transactions and find the new |
| 3248 |
* xmin. |
| 3249 |
*/ |
| 3250 |
static void |
| 3251 |
SetNewSxactGlobalXmin(void) |
| 3252 |
{ |
| 3253 |
dlist_iter iter; |
| 3254 |
|
| 3255 |
Assert(LWLockHeldByMe(SerializableXactHashLock)); |
| 3256 |
|
| 3257 |
PredXact->SxactGlobalXmin = InvalidTransactionId; |
| 3258 |
PredXact->SxactGlobalXminCount = 0; |
| 3259 |
|
| 3260 |
dlist_foreach(iter, &PredXact->activeList) |
| 3261 |
{ |
| 3262 |
SERIALIZABLEXACT *sxact = |
| 3263 |
dlist_container(SERIALIZABLEXACT, xactLink, iter.cur); |
| 3264 |
|
| 3265 |
if (!SxactIsRolledBack(sxact) |
| 3266 |
&& !SxactIsCommitted(sxact) |
| 3267 |
&& sxact != OldCommittedSxact) |
| 3268 |
{ |
| 3269 |
Assert(sxact->xmin != InvalidTransactionId); |
| 3270 |
if (!TransactionIdIsValid(PredXact->SxactGlobalXmin) |
| 3271 |
|| TransactionIdPrecedes(sxact->xmin, |
| 3272 |
PredXact->SxactGlobalXmin)) |
| 3273 |
{ |
| 3274 |
PredXact->SxactGlobalXmin = sxact->xmin; |
| 3275 |
PredXact->SxactGlobalXminCount = 1; |
| 3276 |
} |
| 3277 |
else if (TransactionIdEquals(sxact->xmin, |
| 3278 |
PredXact->SxactGlobalXmin)) |
| 3279 |
PredXact->SxactGlobalXminCount++; |
| 3280 |
} |
| 3281 |
} |
| 3282 |
|
| 3283 |
SerialSetActiveSerXmin(PredXact->SxactGlobalXmin); |
| 3284 |
} |
| 3285 |
|
| 3286 |
/* |
| 3287 |
* ReleasePredicateLocks |
| 3288 |
* |
| 3289 |
* Releases predicate locks based on completion of the current transaction, |
| 3290 |
* whether committed or rolled back. It can also be called for a read only |
| 3291 |
* transaction when it becomes impossible for the transaction to become |
| 3292 |
* part of a dangerous structure. |
| 3293 |
* |
| 3294 |
* We do nothing unless this is a serializable transaction. |
| 3295 |
* |
| 3296 |
* This method must ensure that shared memory hash tables are cleaned |
| 3297 |
* up in some relatively timely fashion. |
| 3298 |
* |
| 3299 |
* If this transaction is committing and is holding any predicate locks, |
| 3300 |
* it must be added to a list of completed serializable transactions still |
| 3301 |
* holding locks. |
| 3302 |
* |
| 3303 |
* If isReadOnlySafe is true, then predicate locks are being released before |
| 3304 |
* the end of the transaction because MySerializableXact has been determined |
| 3305 |
* to be RO_SAFE. In non-parallel mode we can release it completely, but it |
| 3306 |
* in parallel mode we partially release the SERIALIZABLEXACT and keep it |
| 3307 |
* around until the end of the transaction, allowing each backend to clear its |
| 3308 |
* MySerializableXact variable and benefit from the optimization in its own |
| 3309 |
* time. |
| 3310 |
*/ |
| 3311 |
void |
| 3312 |
ReleasePredicateLocks(bool isCommit, bool isReadOnlySafe) |
| 3313 |
{ |
| 3314 |
bool partiallyReleasing = false; |
| 3315 |
bool needToClear; |
| 3316 |
SERIALIZABLEXACT *roXact; |
| 3317 |
dlist_mutable_iter iter; |
| 3318 |
|
| 3319 |
/* |
| 3320 |
* We can't trust XactReadOnly here, because a transaction which started |
| 3321 |
* as READ WRITE can show as READ ONLY later, e.g., within |
| 3322 |
* subtransactions. We want to flag a transaction as READ ONLY if it |
| 3323 |
* commits without writing so that de facto READ ONLY transactions get the |
| 3324 |
* benefit of some RO optimizations, so we will use this local variable to |
| 3325 |
* get some cleanup logic right which is based on whether the transaction |
| 3326 |
* was declared READ ONLY at the top level. |
| 3327 |
*/ |
| 3328 |
bool topLevelIsDeclaredReadOnly; |
| 3329 |
|
| 3330 |
/* We can't be both committing and releasing early due to RO_SAFE. */ |
| 3331 |
Assert(!(isCommit && isReadOnlySafe)); |
| 3332 |
|
| 3333 |
/* Are we at the end of a transaction, that is, a commit or abort? */ |
| 3334 |
if (!isReadOnlySafe) |
| 3335 |
{ |
| 3336 |
/* |
| 3337 |
* Parallel workers mustn't release predicate locks at the end of |
| 3338 |
* their transaction. The leader will do that at the end of its |
| 3339 |
* transaction. |
| 3340 |
*/ |
| 3341 |
if (IsParallelWorker()) |
| 3342 |
{ |
| 3343 |
ReleasePredicateLocksLocal(); |
| 3344 |
return; |
| 3345 |
} |
| 3346 |
|
| 3347 |
/* |
| 3348 |
* By the time the leader in a parallel query reaches end of |
| 3349 |
* transaction, it has waited for all workers to exit. |
| 3350 |
*/ |
| 3351 |
Assert(!ParallelContextActive()); |
| 3352 |
|
| 3353 |
/* |
| 3354 |
* If the leader in a parallel query earlier stashed a partially |
| 3355 |
* released SERIALIZABLEXACT for final clean-up at end of transaction |
| 3356 |
* (because workers might still have been accessing it), then it's |
| 3357 |
* time to restore it. |
| 3358 |
*/ |
| 3359 |
if (SavedSerializableXact != InvalidSerializableXact) |
| 3360 |
{ |
| 3361 |
Assert(MySerializableXact == InvalidSerializableXact); |
| 3362 |
MySerializableXact = SavedSerializableXact; |
| 3363 |
SavedSerializableXact = InvalidSerializableXact; |
| 3364 |
Assert(SxactIsPartiallyReleased(MySerializableXact)); |
| 3365 |
} |
| 3366 |
} |
| 3367 |
|
| 3368 |
if (MySerializableXact == InvalidSerializableXact) |
| 3369 |
{ |
| 3370 |
Assert(LocalPredicateLockHash == NULL); |
| 3371 |
return; |
| 3372 |
} |
| 3373 |
|
| 3374 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 3375 |
|
| 3376 |
/* |
| 3377 |
* If the transaction is committing, but it has been partially released |
| 3378 |
* already, then treat this as a roll back. It was marked as rolled back. |
| 3379 |
*/ |
| 3380 |
if (isCommit && SxactIsPartiallyReleased(MySerializableXact)) |
| 3381 |
isCommit = false; |
| 3382 |
|
| 3383 |
/* |
| 3384 |
* If we're called in the middle of a transaction because we discovered |
| 3385 |
* that the SXACT_FLAG_RO_SAFE flag was set, then we'll partially release |
| 3386 |
* it (that is, release the predicate locks and conflicts, but not the |
| 3387 |
* SERIALIZABLEXACT itself) if we're the first backend to have noticed. |
| 3388 |
*/ |
| 3389 |
if (isReadOnlySafe && IsInParallelMode()) |
| 3390 |
{ |
| 3391 |
/* |
| 3392 |
* The leader needs to stash a pointer to it, so that it can |
| 3393 |
* completely release it at end-of-transaction. |
| 3394 |
*/ |
| 3395 |
if (!IsParallelWorker()) |
| 3396 |
SavedSerializableXact = MySerializableXact; |
| 3397 |
|
| 3398 |
/* |
| 3399 |
* The first backend to reach this condition will partially release |
| 3400 |
* the SERIALIZABLEXACT. All others will just clear their |
| 3401 |
* backend-local state so that they stop doing SSI checks for the rest |
| 3402 |
* of the transaction. |
| 3403 |
*/ |
| 3404 |
if (SxactIsPartiallyReleased(MySerializableXact)) |
| 3405 |
{ |
| 3406 |
LWLockRelease(SerializableXactHashLock); |
| 3407 |
ReleasePredicateLocksLocal(); |
| 3408 |
return; |
| 3409 |
} |
| 3410 |
else |
| 3411 |
{ |
| 3412 |
MySerializableXact->flags |= SXACT_FLAG_PARTIALLY_RELEASED; |
| 3413 |
partiallyReleasing = true; |
| 3414 |
/* ... and proceed to perform the partial release below. */ |
| 3415 |
} |
| 3416 |
} |
| 3417 |
Assert(!isCommit || SxactIsPrepared(MySerializableXact)); |
| 3418 |
Assert(!isCommit || !SxactIsDoomed(MySerializableXact)); |
| 3419 |
Assert(!SxactIsCommitted(MySerializableXact)); |
| 3420 |
Assert(SxactIsPartiallyReleased(MySerializableXact) |
| 3421 |
|| !SxactIsRolledBack(MySerializableXact)); |
| 3422 |
|
| 3423 |
/* may not be serializable during COMMIT/ROLLBACK PREPARED */ |
| 3424 |
Assert(MySerializableXact->pid == 0 || IsolationIsSerializable()); |
| 3425 |
|
| 3426 |
/* We'd better not already be on the cleanup list. */ |
| 3427 |
Assert(!SxactIsOnFinishedList(MySerializableXact)); |
| 3428 |
|
| 3429 |
topLevelIsDeclaredReadOnly = SxactIsReadOnly(MySerializableXact); |
| 3430 |
|
| 3431 |
/* |
| 3432 |
* We don't hold XidGenLock lock here, assuming that TransactionId is |
| 3433 |
* atomic! |
| 3434 |
* |
| 3435 |
* If this value is changing, we don't care that much whether we get the |
| 3436 |
* old or new value -- it is just used to determine how far |
| 3437 |
* SxactGlobalXmin must advance before this transaction can be fully |
| 3438 |
* cleaned up. The worst that could happen is we wait for one more |
| 3439 |
* transaction to complete before freeing some RAM; correctness of visible |
| 3440 |
* behavior is not affected. |
| 3441 |
*/ |
| 3442 |
MySerializableXact->finishedBefore = XidFromFullTransactionId(TransamVariables->nextXid); |
| 3443 |
|
| 3444 |
/* |
| 3445 |
* If it's not a commit it's either a rollback or a read-only transaction |
| 3446 |
* flagged SXACT_FLAG_RO_SAFE, and we can clear our locks immediately. |
| 3447 |
*/ |
| 3448 |
if (isCommit) |
| 3449 |
{ |
| 3450 |
MySerializableXact->flags |= SXACT_FLAG_COMMITTED; |
| 3451 |
MySerializableXact->commitSeqNo = ++(PredXact->LastSxactCommitSeqNo); |
| 3452 |
/* Recognize implicit read-only transaction (commit without write). */ |
| 3453 |
if (!MyXactDidWrite) |
| 3454 |
MySerializableXact->flags |= SXACT_FLAG_READ_ONLY; |
| 3455 |
} |
| 3456 |
else |
| 3457 |
{ |
| 3458 |
/* |
| 3459 |
* The DOOMED flag indicates that we intend to roll back this |
| 3460 |
* transaction and so it should not cause serialization failures for |
| 3461 |
* other transactions that conflict with it. Note that this flag might |
| 3462 |
* already be set, if another backend marked this transaction for |
| 3463 |
* abort. |
| 3464 |
* |
| 3465 |
* The ROLLED_BACK flag further indicates that ReleasePredicateLocks |
| 3466 |
* has been called, and so the SerializableXact is eligible for |
| 3467 |
* cleanup. This means it should not be considered when calculating |
| 3468 |
* SxactGlobalXmin. |
| 3469 |
*/ |
| 3470 |
MySerializableXact->flags |= SXACT_FLAG_DOOMED; |
| 3471 |
MySerializableXact->flags |= SXACT_FLAG_ROLLED_BACK; |
| 3472 |
|
| 3473 |
/* |
| 3474 |
* If the transaction was previously prepared, but is now failing due |
| 3475 |
* to a ROLLBACK PREPARED or (hopefully very rare) error after the |
| 3476 |
* prepare, clear the prepared flag. This simplifies conflict |
| 3477 |
* checking. |
| 3478 |
*/ |
| 3479 |
MySerializableXact->flags &= ~SXACT_FLAG_PREPARED; |
| 3480 |
} |
| 3481 |
|
| 3482 |
if (!topLevelIsDeclaredReadOnly) |
| 3483 |
{ |
| 3484 |
Assert(PredXact->WritableSxactCount > 0); |
| 3485 |
if (--(PredXact->WritableSxactCount) == 0) |
| 3486 |
{ |
| 3487 |
/* |
| 3488 |
* Release predicate locks and rw-conflicts in for all committed |
| 3489 |
* transactions. There are no longer any transactions which might |
| 3490 |
* conflict with the locks and no chance for new transactions to |
| 3491 |
* overlap. Similarly, existing conflicts in can't cause pivots, |
| 3492 |
* and any conflicts in which could have completed a dangerous |
| 3493 |
* structure would already have caused a rollback, so any |
| 3494 |
* remaining ones must be benign. |
| 3495 |
*/ |
| 3496 |
PredXact->CanPartialClearThrough = PredXact->LastSxactCommitSeqNo; |
| 3497 |
} |
| 3498 |
} |
| 3499 |
else |
| 3500 |
{ |
| 3501 |
/* |
| 3502 |
* Read-only transactions: clear the list of transactions that might |
| 3503 |
* make us unsafe. Note that we use 'inLink' for the iteration as |
| 3504 |
* opposed to 'outLink' for the r/w xacts. |
| 3505 |
*/ |
| 3506 |
dlist_foreach_modify(iter, &MySerializableXact->possibleUnsafeConflicts) |
| 3507 |
{ |
| 3508 |
RWConflict possibleUnsafeConflict = |
| 3509 |
dlist_container(RWConflictData, inLink, iter.cur); |
| 3510 |
|
| 3511 |
Assert(!SxactIsReadOnly(possibleUnsafeConflict->sxactOut)); |
| 3512 |
Assert(MySerializableXact == possibleUnsafeConflict->sxactIn); |
| 3513 |
|
| 3514 |
ReleaseRWConflict(possibleUnsafeConflict); |
| 3515 |
} |
| 3516 |
} |
| 3517 |
|
| 3518 |
/* Check for conflict out to old committed transactions. */ |
| 3519 |
if (isCommit |
| 3520 |
&& !SxactIsReadOnly(MySerializableXact) |
| 3521 |
&& SxactHasSummaryConflictOut(MySerializableXact)) |
| 3522 |
{ |
| 3523 |
/* |
| 3524 |
* we don't know which old committed transaction we conflicted with, |
| 3525 |
* so be conservative and use FirstNormalSerCommitSeqNo here |
| 3526 |
*/ |
| 3527 |
MySerializableXact->SeqNo.earliestOutConflictCommit = |
| 3528 |
FirstNormalSerCommitSeqNo; |
| 3529 |
MySerializableXact->flags |= SXACT_FLAG_CONFLICT_OUT; |
| 3530 |
} |
| 3531 |
|
| 3532 |
/* |
| 3533 |
* Release all outConflicts to committed transactions. If we're rolling |
| 3534 |
* back clear them all. Set SXACT_FLAG_CONFLICT_OUT if any point to |
| 3535 |
* previously committed transactions. |
| 3536 |
*/ |
| 3537 |
dlist_foreach_modify(iter, &MySerializableXact->outConflicts) |
| 3538 |
{ |
| 3539 |
RWConflict conflict = |
| 3540 |
dlist_container(RWConflictData, outLink, iter.cur); |
| 3541 |
|
| 3542 |
if (isCommit |
| 3543 |
&& !SxactIsReadOnly(MySerializableXact) |
| 3544 |
&& SxactIsCommitted(conflict->sxactIn)) |
| 3545 |
{ |
| 3546 |
if ((MySerializableXact->flags & SXACT_FLAG_CONFLICT_OUT) == 0 |
| 3547 |
|| conflict->sxactIn->prepareSeqNo < MySerializableXact->SeqNo.earliestOutConflictCommit) |
| 3548 |
MySerializableXact->SeqNo.earliestOutConflictCommit = conflict->sxactIn->prepareSeqNo; |
| 3549 |
MySerializableXact->flags |= SXACT_FLAG_CONFLICT_OUT; |
| 3550 |
} |
| 3551 |
|
| 3552 |
if (!isCommit |
| 3553 |
|| SxactIsCommitted(conflict->sxactIn) |
| 3554 |
|| (conflict->sxactIn->SeqNo.lastCommitBeforeSnapshot >= PredXact->LastSxactCommitSeqNo)) |
| 3555 |
ReleaseRWConflict(conflict); |
| 3556 |
} |
| 3557 |
|
| 3558 |
/* |
| 3559 |
* Release all inConflicts from committed and read-only transactions. If |
| 3560 |
* we're rolling back, clear them all. |
| 3561 |
*/ |
| 3562 |
dlist_foreach_modify(iter, &MySerializableXact->inConflicts) |
| 3563 |
{ |
| 3564 |
RWConflict conflict = |
| 3565 |
dlist_container(RWConflictData, inLink, iter.cur); |
| 3566 |
|
| 3567 |
if (!isCommit |
| 3568 |
|| SxactIsCommitted(conflict->sxactOut) |
| 3569 |
|| SxactIsReadOnly(conflict->sxactOut)) |
| 3570 |
ReleaseRWConflict(conflict); |
| 3571 |
} |
| 3572 |
|
| 3573 |
if (!topLevelIsDeclaredReadOnly) |
| 3574 |
{ |
| 3575 |
/* |
| 3576 |
* Remove ourselves from the list of possible conflicts for concurrent |
| 3577 |
* READ ONLY transactions, flagging them as unsafe if we have a |
| 3578 |
* conflict out. If any are waiting DEFERRABLE transactions, wake them |
| 3579 |
* up if they are known safe or known unsafe. |
| 3580 |
*/ |
| 3581 |
dlist_foreach_modify(iter, &MySerializableXact->possibleUnsafeConflicts) |
| 3582 |
{ |
| 3583 |
RWConflict possibleUnsafeConflict = |
| 3584 |
dlist_container(RWConflictData, outLink, iter.cur); |
| 3585 |
|
| 3586 |
roXact = possibleUnsafeConflict->sxactIn; |
| 3587 |
Assert(MySerializableXact == possibleUnsafeConflict->sxactOut); |
| 3588 |
Assert(SxactIsReadOnly(roXact)); |
| 3589 |
|
| 3590 |
/* Mark conflicted if necessary. */ |
| 3591 |
if (isCommit |
| 3592 |
&& MyXactDidWrite |
| 3593 |
&& SxactHasConflictOut(MySerializableXact) |
| 3594 |
&& (MySerializableXact->SeqNo.earliestOutConflictCommit |
| 3595 |
<= roXact->SeqNo.lastCommitBeforeSnapshot)) |
| 3596 |
{ |
| 3597 |
/* |
| 3598 |
* This releases possibleUnsafeConflict (as well as all other |
| 3599 |
* possible conflicts for roXact) |
| 3600 |
*/ |
| 3601 |
FlagSxactUnsafe(roXact); |
| 3602 |
} |
| 3603 |
else |
| 3604 |
{ |
| 3605 |
ReleaseRWConflict(possibleUnsafeConflict); |
| 3606 |
|
| 3607 |
/* |
| 3608 |
* If we were the last possible conflict, flag it safe. The |
| 3609 |
* transaction can now safely release its predicate locks (but |
| 3610 |
* that transaction's backend has to do that itself). |
| 3611 |
*/ |
| 3612 |
if (dlist_is_empty(&roXact->possibleUnsafeConflicts)) |
| 3613 |
roXact->flags |= SXACT_FLAG_RO_SAFE; |
| 3614 |
} |
| 3615 |
|
| 3616 |
/* |
| 3617 |
* Wake up the process for a waiting DEFERRABLE transaction if we |
| 3618 |
* now know it's either safe or conflicted. |
| 3619 |
*/ |
| 3620 |
if (SxactIsDeferrableWaiting(roXact) && |
| 3621 |
(SxactIsROUnsafe(roXact) || SxactIsROSafe(roXact))) |
| 3622 |
ProcSendSignal(roXact->pgprocno); |
| 3623 |
} |
| 3624 |
} |
| 3625 |
|
| 3626 |
/* |
| 3627 |
* Check whether it's time to clean up old transactions. This can only be |
| 3628 |
* done when the last serializable transaction with the oldest xmin among |
| 3629 |
* serializable transactions completes. We then find the "new oldest" |
| 3630 |
* xmin and purge any transactions which finished before this transaction |
| 3631 |
* was launched. |
| 3632 |
* |
| 3633 |
* For parallel queries in read-only transactions, it might run twice. We |
| 3634 |
* only release the reference on the first call. |
| 3635 |
*/ |
| 3636 |
needToClear = false; |
| 3637 |
if ((partiallyReleasing || |
| 3638 |
!SxactIsPartiallyReleased(MySerializableXact)) && |
| 3639 |
TransactionIdEquals(MySerializableXact->xmin, |
| 3640 |
PredXact->SxactGlobalXmin)) |
| 3641 |
{ |
| 3642 |
Assert(PredXact->SxactGlobalXminCount > 0); |
| 3643 |
if (--(PredXact->SxactGlobalXminCount) == 0) |
| 3644 |
{ |
| 3645 |
SetNewSxactGlobalXmin(); |
| 3646 |
needToClear = true; |
| 3647 |
} |
| 3648 |
} |
| 3649 |
|
| 3650 |
LWLockRelease(SerializableXactHashLock); |
| 3651 |
|
| 3652 |
LWLockAcquire(SerializableFinishedListLock, LW_EXCLUSIVE); |
| 3653 |
|
| 3654 |
/* Add this to the list of transactions to check for later cleanup. */ |
| 3655 |
if (isCommit) |
| 3656 |
dlist_push_tail(FinishedSerializableTransactions, |
| 3657 |
&MySerializableXact->finishedLink); |
| 3658 |
|
| 3659 |
/* |
| 3660 |
* If we're releasing a RO_SAFE transaction in parallel mode, we'll only |
| 3661 |
* partially release it. That's necessary because other backends may have |
| 3662 |
* a reference to it. The leader will release the SERIALIZABLEXACT itself |
| 3663 |
* at the end of the transaction after workers have stopped running. |
| 3664 |
*/ |
| 3665 |
if (!isCommit) |
| 3666 |
ReleaseOneSerializableXact(MySerializableXact, |
| 3667 |
isReadOnlySafe && IsInParallelMode(), |
| 3668 |
false); |
| 3669 |
|
| 3670 |
LWLockRelease(SerializableFinishedListLock); |
| 3671 |
|
| 3672 |
if (needToClear) |
| 3673 |
ClearOldPredicateLocks(); |
| 3674 |
|
| 3675 |
ReleasePredicateLocksLocal(); |
| 3676 |
} |
| 3677 |
|
| 3678 |
static void |
| 3679 |
ReleasePredicateLocksLocal(void) |
| 3680 |
{ |
| 3681 |
MySerializableXact = InvalidSerializableXact; |
| 3682 |
MyXactDidWrite = false; |
| 3683 |
|
| 3684 |
/* Delete per-transaction lock table */ |
| 3685 |
if (LocalPredicateLockHash != NULL) |
| 3686 |
{ |
| 3687 |
hash_destroy(LocalPredicateLockHash); |
| 3688 |
LocalPredicateLockHash = NULL; |
| 3689 |
} |
| 3690 |
} |
| 3691 |
|
| 3692 |
/* |
| 3693 |
* Clear old predicate locks, belonging to committed transactions that are no |
| 3694 |
* longer interesting to any in-progress transaction. |
| 3695 |
*/ |
| 3696 |
static void |
| 3697 |
ClearOldPredicateLocks(void) |
| 3698 |
{ |
| 3699 |
dlist_mutable_iter iter; |
| 3700 |
|
| 3701 |
/* |
| 3702 |
* Loop through finished transactions. They are in commit order, so we can |
| 3703 |
* stop as soon as we find one that's still interesting. |
| 3704 |
*/ |
| 3705 |
LWLockAcquire(SerializableFinishedListLock, LW_EXCLUSIVE); |
| 3706 |
LWLockAcquire(SerializableXactHashLock, LW_SHARED); |
| 3707 |
dlist_foreach_modify(iter, FinishedSerializableTransactions) |
| 3708 |
{ |
| 3709 |
SERIALIZABLEXACT *finishedSxact = |
| 3710 |
dlist_container(SERIALIZABLEXACT, finishedLink, iter.cur); |
| 3711 |
|
| 3712 |
if (!TransactionIdIsValid(PredXact->SxactGlobalXmin) |
| 3713 |
|| TransactionIdPrecedesOrEquals(finishedSxact->finishedBefore, |
| 3714 |
PredXact->SxactGlobalXmin)) |
| 3715 |
{ |
| 3716 |
/* |
| 3717 |
* This transaction committed before any in-progress transaction |
| 3718 |
* took its snapshot. It's no longer interesting. |
| 3719 |
*/ |
| 3720 |
LWLockRelease(SerializableXactHashLock); |
| 3721 |
dlist_delete_thoroughly(&finishedSxact->finishedLink); |
| 3722 |
ReleaseOneSerializableXact(finishedSxact, false, false); |
| 3723 |
LWLockAcquire(SerializableXactHashLock, LW_SHARED); |
| 3724 |
} |
| 3725 |
else if (finishedSxact->commitSeqNo > PredXact->HavePartialClearedThrough |
| 3726 |
&& finishedSxact->commitSeqNo <= PredXact->CanPartialClearThrough) |
| 3727 |
{ |
| 3728 |
/* |
| 3729 |
* Any active transactions that took their snapshot before this |
| 3730 |
* transaction committed are read-only, so we can clear part of |
| 3731 |
* its state. |
| 3732 |
*/ |
| 3733 |
LWLockRelease(SerializableXactHashLock); |
| 3734 |
|
| 3735 |
if (SxactIsReadOnly(finishedSxact)) |
| 3736 |
{ |
| 3737 |
/* A read-only transaction can be removed entirely */ |
| 3738 |
dlist_delete_thoroughly(&(finishedSxact->finishedLink)); |
| 3739 |
ReleaseOneSerializableXact(finishedSxact, false, false); |
| 3740 |
} |
| 3741 |
else |
| 3742 |
{ |
| 3743 |
/* |
| 3744 |
* A read-write transaction can only be partially cleared. We |
| 3745 |
* need to keep the SERIALIZABLEXACT but can release the |
| 3746 |
* SIREAD locks and conflicts in. |
| 3747 |
*/ |
| 3748 |
ReleaseOneSerializableXact(finishedSxact, true, false); |
| 3749 |
} |
| 3750 |
|
| 3751 |
PredXact->HavePartialClearedThrough = finishedSxact->commitSeqNo; |
| 3752 |
LWLockAcquire(SerializableXactHashLock, LW_SHARED); |
| 3753 |
} |
| 3754 |
else |
| 3755 |
{ |
| 3756 |
/* Still interesting. */ |
| 3757 |
break; |
| 3758 |
} |
| 3759 |
} |
| 3760 |
LWLockRelease(SerializableXactHashLock); |
| 3761 |
|
| 3762 |
/* |
| 3763 |
* Loop through predicate locks on dummy transaction for summarized data. |
| 3764 |
*/ |
| 3765 |
LWLockAcquire(SerializablePredicateListLock, LW_SHARED); |
| 3766 |
dlist_foreach_modify(iter, &OldCommittedSxact->predicateLocks) |
| 3767 |
{ |
| 3768 |
PREDICATELOCK *predlock = |
| 3769 |
dlist_container(PREDICATELOCK, xactLink, iter.cur); |
| 3770 |
bool canDoPartialCleanup; |
| 3771 |
|
| 3772 |
LWLockAcquire(SerializableXactHashLock, LW_SHARED); |
| 3773 |
Assert(predlock->commitSeqNo != 0); |
| 3774 |
Assert(predlock->commitSeqNo != InvalidSerCommitSeqNo); |
| 3775 |
canDoPartialCleanup = (predlock->commitSeqNo <= PredXact->CanPartialClearThrough); |
| 3776 |
LWLockRelease(SerializableXactHashLock); |
| 3777 |
|
| 3778 |
/* |
| 3779 |
* If this lock originally belonged to an old enough transaction, we |
| 3780 |
* can release it. |
| 3781 |
*/ |
| 3782 |
if (canDoPartialCleanup) |
| 3783 |
{ |
| 3784 |
PREDICATELOCKTAG tag; |
| 3785 |
PREDICATELOCKTARGET *target; |
| 3786 |
PREDICATELOCKTARGETTAG targettag; |
| 3787 |
uint32 targettaghash; |
| 3788 |
LWLock *partitionLock; |
| 3789 |
|
| 3790 |
tag = predlock->tag; |
| 3791 |
target = tag.myTarget; |
| 3792 |
targettag = target->tag; |
| 3793 |
targettaghash = PredicateLockTargetTagHashCode(&targettag); |
| 3794 |
partitionLock = PredicateLockHashPartitionLock(targettaghash); |
| 3795 |
|
| 3796 |
LWLockAcquire(partitionLock, LW_EXCLUSIVE); |
| 3797 |
|
| 3798 |
dlist_delete(&(predlock->targetLink)); |
| 3799 |
dlist_delete(&(predlock->xactLink)); |
| 3800 |
|
| 3801 |
hash_search_with_hash_value(PredicateLockHash, &tag, |
| 3802 |
PredicateLockHashCodeFromTargetHashCode(&tag, |
| 3803 |
targettaghash), |
| 3804 |
HASH_REMOVE, NULL); |
| 3805 |
RemoveTargetIfNoLongerUsed(target, targettaghash); |
| 3806 |
|
| 3807 |
LWLockRelease(partitionLock); |
| 3808 |
} |
| 3809 |
} |
| 3810 |
|
| 3811 |
LWLockRelease(SerializablePredicateListLock); |
| 3812 |
LWLockRelease(SerializableFinishedListLock); |
| 3813 |
} |
| 3814 |
|
| 3815 |
/* |
| 3816 |
* This is the normal way to delete anything from any of the predicate |
| 3817 |
* locking hash tables. Given a transaction which we know can be deleted: |
| 3818 |
* delete all predicate locks held by that transaction and any predicate |
| 3819 |
* lock targets which are now unreferenced by a lock; delete all conflicts |
| 3820 |
* for the transaction; delete all xid values for the transaction; then |
| 3821 |
* delete the transaction. |
| 3822 |
* |
| 3823 |
* When the partial flag is set, we can release all predicate locks and |
| 3824 |
* in-conflict information -- we've established that there are no longer |
| 3825 |
* any overlapping read write transactions for which this transaction could |
| 3826 |
* matter -- but keep the transaction entry itself and any outConflicts. |
| 3827 |
* |
| 3828 |
* When the summarize flag is set, we've run short of room for sxact data |
| 3829 |
* and must summarize to the SLRU. Predicate locks are transferred to a |
| 3830 |
* dummy "old" transaction, with duplicate locks on a single target |
| 3831 |
* collapsing to a single lock with the "latest" commitSeqNo from among |
| 3832 |
* the conflicting locks.. |
| 3833 |
*/ |
| 3834 |
static void |
| 3835 |
ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial, |
| 3836 |
bool summarize) |
| 3837 |
{ |
| 3838 |
SERIALIZABLEXIDTAG sxidtag; |
| 3839 |
dlist_mutable_iter iter; |
| 3840 |
|
| 3841 |
Assert(sxact != NULL); |
| 3842 |
Assert(SxactIsRolledBack(sxact) || SxactIsCommitted(sxact)); |
| 3843 |
Assert(partial || !SxactIsOnFinishedList(sxact)); |
| 3844 |
Assert(LWLockHeldByMe(SerializableFinishedListLock)); |
| 3845 |
|
| 3846 |
/* |
| 3847 |
* First release all the predicate locks held by this xact (or transfer |
| 3848 |
* them to OldCommittedSxact if summarize is true) |
| 3849 |
*/ |
| 3850 |
LWLockAcquire(SerializablePredicateListLock, LW_SHARED); |
| 3851 |
if (IsInParallelMode()) |
| 3852 |
LWLockAcquire(&sxact->perXactPredicateListLock, LW_EXCLUSIVE); |
| 3853 |
dlist_foreach_modify(iter, &sxact->predicateLocks) |
| 3854 |
{ |
| 3855 |
PREDICATELOCK *predlock = |
| 3856 |
dlist_container(PREDICATELOCK, xactLink, iter.cur); |
| 3857 |
PREDICATELOCKTAG tag; |
| 3858 |
PREDICATELOCKTARGET *target; |
| 3859 |
PREDICATELOCKTARGETTAG targettag; |
| 3860 |
uint32 targettaghash; |
| 3861 |
LWLock *partitionLock; |
| 3862 |
|
| 3863 |
tag = predlock->tag; |
| 3864 |
target = tag.myTarget; |
| 3865 |
targettag = target->tag; |
| 3866 |
targettaghash = PredicateLockTargetTagHashCode(&targettag); |
| 3867 |
partitionLock = PredicateLockHashPartitionLock(targettaghash); |
| 3868 |
|
| 3869 |
LWLockAcquire(partitionLock, LW_EXCLUSIVE); |
| 3870 |
|
| 3871 |
dlist_delete(&predlock->targetLink); |
| 3872 |
|
| 3873 |
hash_search_with_hash_value(PredicateLockHash, &tag, |
| 3874 |
PredicateLockHashCodeFromTargetHashCode(&tag, |
| 3875 |
targettaghash), |
| 3876 |
HASH_REMOVE, NULL); |
| 3877 |
if (summarize) |
| 3878 |
{ |
| 3879 |
bool found; |
| 3880 |
|
| 3881 |
/* Fold into dummy transaction list. */ |
| 3882 |
tag.myXact = OldCommittedSxact; |
| 3883 |
predlock = hash_search_with_hash_value(PredicateLockHash, &tag, |
| 3884 |
PredicateLockHashCodeFromTargetHashCode(&tag, |
| 3885 |
targettaghash), |
| 3886 |
HASH_ENTER_NULL, &found); |
| 3887 |
if (!predlock) |
| 3888 |
ereport(ERROR, |
| 3889 |
(errcode(ERRCODE_OUT_OF_MEMORY), |
| 3890 |
errmsg("out of shared memory"), |
| 3891 |
errhint("You might need to increase \"%s\".", "max_pred_locks_per_transaction"))); |
| 3892 |
if (found) |
| 3893 |
{ |
| 3894 |
Assert(predlock->commitSeqNo != 0); |
| 3895 |
Assert(predlock->commitSeqNo != InvalidSerCommitSeqNo); |
| 3896 |
if (predlock->commitSeqNo < sxact->commitSeqNo) |
| 3897 |
predlock->commitSeqNo = sxact->commitSeqNo; |
| 3898 |
} |
| 3899 |
else |
| 3900 |
{ |
| 3901 |
dlist_push_tail(&target->predicateLocks, |
| 3902 |
&predlock->targetLink); |
| 3903 |
dlist_push_tail(&OldCommittedSxact->predicateLocks, |
| 3904 |
&predlock->xactLink); |
| 3905 |
predlock->commitSeqNo = sxact->commitSeqNo; |
| 3906 |
} |
| 3907 |
} |
| 3908 |
else |
| 3909 |
RemoveTargetIfNoLongerUsed(target, targettaghash); |
| 3910 |
|
| 3911 |
LWLockRelease(partitionLock); |
| 3912 |
} |
| 3913 |
|
| 3914 |
/* |
| 3915 |
* Rather than retail removal, just re-init the head after we've run |
| 3916 |
* through the list. |
| 3917 |
*/ |
| 3918 |
dlist_init(&sxact->predicateLocks); |
| 3919 |
|
| 3920 |
if (IsInParallelMode()) |
| 3921 |
LWLockRelease(&sxact->perXactPredicateListLock); |
| 3922 |
LWLockRelease(SerializablePredicateListLock); |
| 3923 |
|
| 3924 |
sxidtag.xid = sxact->topXid; |
| 3925 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 3926 |
|
| 3927 |
/* Release all outConflicts (unless 'partial' is true) */ |
| 3928 |
if (!partial) |
| 3929 |
{ |
| 3930 |
dlist_foreach_modify(iter, &sxact->outConflicts) |
| 3931 |
{ |
| 3932 |
RWConflict conflict = |
| 3933 |
dlist_container(RWConflictData, outLink, iter.cur); |
| 3934 |
|
| 3935 |
if (summarize) |
| 3936 |
conflict->sxactIn->flags |= SXACT_FLAG_SUMMARY_CONFLICT_IN; |
| 3937 |
ReleaseRWConflict(conflict); |
| 3938 |
} |
| 3939 |
} |
| 3940 |
|
| 3941 |
/* Release all inConflicts. */ |
| 3942 |
dlist_foreach_modify(iter, &sxact->inConflicts) |
| 3943 |
{ |
| 3944 |
RWConflict conflict = |
| 3945 |
dlist_container(RWConflictData, inLink, iter.cur); |
| 3946 |
|
| 3947 |
if (summarize) |
| 3948 |
conflict->sxactOut->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT; |
| 3949 |
ReleaseRWConflict(conflict); |
| 3950 |
} |
| 3951 |
|
| 3952 |
/* Finally, get rid of the xid and the record of the transaction itself. */ |
| 3953 |
if (!partial) |
| 3954 |
{ |
| 3955 |
if (sxidtag.xid != InvalidTransactionId) |
| 3956 |
hash_search(SerializableXidHash, &sxidtag, HASH_REMOVE, NULL); |
| 3957 |
ReleasePredXact(sxact); |
| 3958 |
} |
| 3959 |
|
| 3960 |
LWLockRelease(SerializableXactHashLock); |
| 3961 |
} |
| 3962 |
|
| 3963 |
/* |
| 3964 |
* Tests whether the given top level transaction is concurrent with |
| 3965 |
* (overlaps) our current transaction. |
| 3966 |
* |
| 3967 |
* We need to identify the top level transaction for SSI, anyway, so pass |
| 3968 |
* that to this function to save the overhead of checking the snapshot's |
| 3969 |
* subxip array. |
| 3970 |
*/ |
| 3971 |
static bool |
| 3972 |
XidIsConcurrent(TransactionId xid) |
| 3973 |
{ |
| 3974 |
Snapshot snap; |
| 3975 |
|
| 3976 |
Assert(TransactionIdIsValid(xid)); |
| 3977 |
Assert(!TransactionIdEquals(xid, GetTopTransactionIdIfAny())); |
| 3978 |
|
| 3979 |
snap = GetTransactionSnapshot(); |
| 3980 |
|
| 3981 |
if (TransactionIdPrecedes(xid, snap->xmin)) |
| 3982 |
return false; |
| 3983 |
|
| 3984 |
if (TransactionIdFollowsOrEquals(xid, snap->xmax)) |
| 3985 |
return true; |
| 3986 |
|
| 3987 |
return pg_lfind32(xid, snap->xip, snap->xcnt); |
| 3988 |
} |
| 3989 |
|
| 3990 |
bool |
| 3991 |
CheckForSerializableConflictOutNeeded(Relation relation, Snapshot snapshot) |
| 3992 |
{ |
| 3993 |
if (!SerializationNeededForRead(relation, snapshot)) |
| 3994 |
return false; |
| 3995 |
|
| 3996 |
/* Check if someone else has already decided that we need to die */ |
| 3997 |
if (SxactIsDoomed(MySerializableXact)) |
| 3998 |
{ |
| 3999 |
ereport(ERROR, |
| 4000 |
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE), |
| 4001 |
errmsg("could not serialize access due to read/write dependencies among transactions"), |
| 4002 |
errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."), |
| 4003 |
errhint("The transaction might succeed if retried."))); |
| 4004 |
} |
| 4005 |
|
| 4006 |
return true; |
| 4007 |
} |
| 4008 |
|
| 4009 |
/* |
| 4010 |
* CheckForSerializableConflictOut |
| 4011 |
* A table AM is reading a tuple that has been modified. If it determines |
| 4012 |
* that the tuple version it is reading is not visible to us, it should |
| 4013 |
* pass in the top level xid of the transaction that created it. |
| 4014 |
* Otherwise, if it determines that it is visible to us but it has been |
| 4015 |
* deleted or there is a newer version available due to an update, it |
| 4016 |
* should pass in the top level xid of the modifying transaction. |
| 4017 |
* |
| 4018 |
* This function will check for overlap with our own transaction. If the given |
| 4019 |
* xid is also serializable and the transactions overlap (i.e., they cannot see |
| 4020 |
* each other's writes), then we have a conflict out. |
| 4021 |
*/ |
| 4022 |
void |
| 4023 |
CheckForSerializableConflictOut(Relation relation, TransactionId xid, Snapshot snapshot) |
| 4024 |
{ |
| 4025 |
SERIALIZABLEXIDTAG sxidtag; |
| 4026 |
SERIALIZABLEXID *sxid; |
| 4027 |
SERIALIZABLEXACT *sxact; |
| 4028 |
|
| 4029 |
if (!SerializationNeededForRead(relation, snapshot)) |
| 4030 |
return; |
| 4031 |
|
| 4032 |
/* Check if someone else has already decided that we need to die */ |
| 4033 |
if (SxactIsDoomed(MySerializableXact)) |
| 4034 |
{ |
| 4035 |
ereport(ERROR, |
| 4036 |
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE), |
| 4037 |
errmsg("could not serialize access due to read/write dependencies among transactions"), |
| 4038 |
errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."), |
| 4039 |
errhint("The transaction might succeed if retried."))); |
| 4040 |
} |
| 4041 |
Assert(TransactionIdIsValid(xid)); |
| 4042 |
|
| 4043 |
if (TransactionIdEquals(xid, GetTopTransactionIdIfAny())) |
| 4044 |
return; |
| 4045 |
|
| 4046 |
/* |
| 4047 |
* Find sxact or summarized info for the top level xid. |
| 4048 |
*/ |
| 4049 |
sxidtag.xid = xid; |
| 4050 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 4051 |
sxid = (SERIALIZABLEXID *) |
| 4052 |
hash_search(SerializableXidHash, &sxidtag, HASH_FIND, NULL); |
| 4053 |
if (!sxid) |
| 4054 |
{ |
| 4055 |
/* |
| 4056 |
* Transaction not found in "normal" SSI structures. Check whether it |
| 4057 |
* got pushed out to SLRU storage for "old committed" transactions. |
| 4058 |
*/ |
| 4059 |
SerCommitSeqNo conflictCommitSeqNo; |
| 4060 |
|
| 4061 |
conflictCommitSeqNo = SerialGetMinConflictCommitSeqNo(xid); |
| 4062 |
if (conflictCommitSeqNo != 0) |
| 4063 |
{ |
| 4064 |
if (conflictCommitSeqNo != InvalidSerCommitSeqNo |
| 4065 |
&& (!SxactIsReadOnly(MySerializableXact) |
| 4066 |
|| conflictCommitSeqNo |
| 4067 |
<= MySerializableXact->SeqNo.lastCommitBeforeSnapshot)) |
| 4068 |
ereport(ERROR, |
| 4069 |
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE), |
| 4070 |
errmsg("could not serialize access due to read/write dependencies among transactions"), |
| 4071 |
errdetail_internal("Reason code: Canceled on conflict out to old pivot %u.", xid), |
| 4072 |
errhint("The transaction might succeed if retried."))); |
| 4073 |
|
| 4074 |
if (SxactHasSummaryConflictIn(MySerializableXact) |
| 4075 |
|| !dlist_is_empty(&MySerializableXact->inConflicts)) |
| 4076 |
ereport(ERROR, |
| 4077 |
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE), |
| 4078 |
errmsg("could not serialize access due to read/write dependencies among transactions"), |
| 4079 |
errdetail_internal("Reason code: Canceled on identification as a pivot, with conflict out to old committed transaction %u.", xid), |
| 4080 |
errhint("The transaction might succeed if retried."))); |
| 4081 |
|
| 4082 |
MySerializableXact->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT; |
| 4083 |
} |
| 4084 |
|
| 4085 |
/* It's not serializable or otherwise not important. */ |
| 4086 |
LWLockRelease(SerializableXactHashLock); |
| 4087 |
return; |
| 4088 |
} |
| 4089 |
sxact = sxid->myXact; |
| 4090 |
Assert(TransactionIdEquals(sxact->topXid, xid)); |
| 4091 |
if (sxact == MySerializableXact || SxactIsDoomed(sxact)) |
| 4092 |
{ |
| 4093 |
/* Can't conflict with ourself or a transaction that will roll back. */ |
| 4094 |
LWLockRelease(SerializableXactHashLock); |
| 4095 |
return; |
| 4096 |
} |
| 4097 |
|
| 4098 |
/* |
| 4099 |
* We have a conflict out to a transaction which has a conflict out to a |
| 4100 |
* summarized transaction. That summarized transaction must have |
| 4101 |
* committed first, and we can't tell when it committed in relation to our |
| 4102 |
* snapshot acquisition, so something needs to be canceled. |
| 4103 |
*/ |
| 4104 |
if (SxactHasSummaryConflictOut(sxact)) |
| 4105 |
{ |
| 4106 |
if (!SxactIsPrepared(sxact)) |
| 4107 |
{ |
| 4108 |
sxact->flags |= SXACT_FLAG_DOOMED; |
| 4109 |
LWLockRelease(SerializableXactHashLock); |
| 4110 |
return; |
| 4111 |
} |
| 4112 |
else |
| 4113 |
{ |
| 4114 |
LWLockRelease(SerializableXactHashLock); |
| 4115 |
ereport(ERROR, |
| 4116 |
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE), |
| 4117 |
errmsg("could not serialize access due to read/write dependencies among transactions"), |
| 4118 |
errdetail_internal("Reason code: Canceled on conflict out to old pivot."), |
| 4119 |
errhint("The transaction might succeed if retried."))); |
| 4120 |
} |
| 4121 |
} |
| 4122 |
|
| 4123 |
/* |
| 4124 |
* If this is a read-only transaction and the writing transaction has |
| 4125 |
* committed, and it doesn't have a rw-conflict to a transaction which |
| 4126 |
* committed before it, no conflict. |
| 4127 |
*/ |
| 4128 |
if (SxactIsReadOnly(MySerializableXact) |
| 4129 |
&& SxactIsCommitted(sxact) |
| 4130 |
&& !SxactHasSummaryConflictOut(sxact) |
| 4131 |
&& (!SxactHasConflictOut(sxact) |
| 4132 |
|| MySerializableXact->SeqNo.lastCommitBeforeSnapshot < sxact->SeqNo.earliestOutConflictCommit)) |
| 4133 |
{ |
| 4134 |
/* Read-only transaction will appear to run first. No conflict. */ |
| 4135 |
LWLockRelease(SerializableXactHashLock); |
| 4136 |
return; |
| 4137 |
} |
| 4138 |
|
| 4139 |
if (!XidIsConcurrent(xid)) |
| 4140 |
{ |
| 4141 |
/* This write was already in our snapshot; no conflict. */ |
| 4142 |
LWLockRelease(SerializableXactHashLock); |
| 4143 |
return; |
| 4144 |
} |
| 4145 |
|
| 4146 |
if (RWConflictExists(MySerializableXact, sxact)) |
| 4147 |
{ |
| 4148 |
/* We don't want duplicate conflict records in the list. */ |
| 4149 |
LWLockRelease(SerializableXactHashLock); |
| 4150 |
return; |
| 4151 |
} |
| 4152 |
|
| 4153 |
/* |
| 4154 |
* Flag the conflict. But first, if this conflict creates a dangerous |
| 4155 |
* structure, ereport an error. |
| 4156 |
*/ |
| 4157 |
FlagRWConflict(MySerializableXact, sxact); |
| 4158 |
LWLockRelease(SerializableXactHashLock); |
| 4159 |
} |
| 4160 |
|
| 4161 |
/* |
| 4162 |
* Check a particular target for rw-dependency conflict in. A subroutine of |
| 4163 |
* CheckForSerializableConflictIn(). |
| 4164 |
*/ |
| 4165 |
static void |
| 4166 |
CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag) |
| 4167 |
{ |
| 4168 |
uint32 targettaghash; |
| 4169 |
LWLock *partitionLock; |
| 4170 |
PREDICATELOCKTARGET *target; |
| 4171 |
PREDICATELOCK *mypredlock = NULL; |
| 4172 |
PREDICATELOCKTAG mypredlocktag; |
| 4173 |
dlist_mutable_iter iter; |
| 4174 |
|
| 4175 |
Assert(MySerializableXact != InvalidSerializableXact); |
| 4176 |
|
| 4177 |
/* |
| 4178 |
* The same hash and LW lock apply to the lock target and the lock itself. |
| 4179 |
*/ |
| 4180 |
targettaghash = PredicateLockTargetTagHashCode(targettag); |
| 4181 |
partitionLock = PredicateLockHashPartitionLock(targettaghash); |
| 4182 |
LWLockAcquire(partitionLock, LW_SHARED); |
| 4183 |
target = (PREDICATELOCKTARGET *) |
| 4184 |
hash_search_with_hash_value(PredicateLockTargetHash, |
| 4185 |
targettag, targettaghash, |
| 4186 |
HASH_FIND, NULL); |
| 4187 |
if (!target) |
| 4188 |
{ |
| 4189 |
/* Nothing has this target locked; we're done here. */ |
| 4190 |
LWLockRelease(partitionLock); |
| 4191 |
return; |
| 4192 |
} |
| 4193 |
|
| 4194 |
/* |
| 4195 |
* Each lock for an overlapping transaction represents a conflict: a |
| 4196 |
* rw-dependency in to this transaction. |
| 4197 |
*/ |
| 4198 |
LWLockAcquire(SerializableXactHashLock, LW_SHARED); |
| 4199 |
|
| 4200 |
dlist_foreach_modify(iter, &target->predicateLocks) |
| 4201 |
{ |
| 4202 |
PREDICATELOCK *predlock = |
| 4203 |
dlist_container(PREDICATELOCK, targetLink, iter.cur); |
| 4204 |
SERIALIZABLEXACT *sxact = predlock->tag.myXact; |
| 4205 |
|
| 4206 |
if (sxact == MySerializableXact) |
| 4207 |
{ |
| 4208 |
/* |
| 4209 |
* If we're getting a write lock on a tuple, we don't need a |
| 4210 |
* predicate (SIREAD) lock on the same tuple. We can safely remove |
| 4211 |
* our SIREAD lock, but we'll defer doing so until after the loop |
| 4212 |
* because that requires upgrading to an exclusive partition lock. |
| 4213 |
* |
| 4214 |
* We can't use this optimization within a subtransaction because |
| 4215 |
* the subtransaction could roll back, and we would be left |
| 4216 |
* without any lock at the top level. |
| 4217 |
*/ |
| 4218 |
if (!IsSubTransaction() |
| 4219 |
&& GET_PREDICATELOCKTARGETTAG_OFFSET(*targettag)) |
| 4220 |
{ |
| 4221 |
mypredlock = predlock; |
| 4222 |
mypredlocktag = predlock->tag; |
| 4223 |
} |
| 4224 |
} |
| 4225 |
else if (!SxactIsDoomed(sxact) |
| 4226 |
&& (!SxactIsCommitted(sxact) |
| 4227 |
|| TransactionIdPrecedes(GetTransactionSnapshot()->xmin, |
| 4228 |
sxact->finishedBefore)) |
| 4229 |
&& !RWConflictExists(sxact, MySerializableXact)) |
| 4230 |
{ |
| 4231 |
LWLockRelease(SerializableXactHashLock); |
| 4232 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 4233 |
|
| 4234 |
/* |
| 4235 |
* Re-check after getting exclusive lock because the other |
| 4236 |
* transaction may have flagged a conflict. |
| 4237 |
*/ |
| 4238 |
if (!SxactIsDoomed(sxact) |
| 4239 |
&& (!SxactIsCommitted(sxact) |
| 4240 |
|| TransactionIdPrecedes(GetTransactionSnapshot()->xmin, |
| 4241 |
sxact->finishedBefore)) |
| 4242 |
&& !RWConflictExists(sxact, MySerializableXact)) |
| 4243 |
{ |
| 4244 |
FlagRWConflict(sxact, MySerializableXact); |
| 4245 |
} |
| 4246 |
|
| 4247 |
LWLockRelease(SerializableXactHashLock); |
| 4248 |
LWLockAcquire(SerializableXactHashLock, LW_SHARED); |
| 4249 |
} |
| 4250 |
} |
| 4251 |
LWLockRelease(SerializableXactHashLock); |
| 4252 |
LWLockRelease(partitionLock); |
| 4253 |
|
| 4254 |
/* |
| 4255 |
* If we found one of our own SIREAD locks to remove, remove it now. |
| 4256 |
* |
| 4257 |
* At this point our transaction already has a RowExclusiveLock on the |
| 4258 |
* relation, so we are OK to drop the predicate lock on the tuple, if |
| 4259 |
* found, without fearing that another write against the tuple will occur |
| 4260 |
* before the MVCC information makes it to the buffer. |
| 4261 |
*/ |
| 4262 |
if (mypredlock != NULL) |
| 4263 |
{ |
| 4264 |
uint32 predlockhashcode; |
| 4265 |
PREDICATELOCK *rmpredlock; |
| 4266 |
|
| 4267 |
LWLockAcquire(SerializablePredicateListLock, LW_SHARED); |
| 4268 |
if (IsInParallelMode()) |
| 4269 |
LWLockAcquire(&MySerializableXact->perXactPredicateListLock, LW_EXCLUSIVE); |
| 4270 |
LWLockAcquire(partitionLock, LW_EXCLUSIVE); |
| 4271 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 4272 |
|
| 4273 |
/* |
| 4274 |
* Remove the predicate lock from shared memory, if it wasn't removed |
| 4275 |
* while the locks were released. One way that could happen is from |
| 4276 |
* autovacuum cleaning up an index. |
| 4277 |
*/ |
| 4278 |
predlockhashcode = PredicateLockHashCodeFromTargetHashCode |
| 4279 |
(&mypredlocktag, targettaghash); |
| 4280 |
rmpredlock = (PREDICATELOCK *) |
| 4281 |
hash_search_with_hash_value(PredicateLockHash, |
| 4282 |
&mypredlocktag, |
| 4283 |
predlockhashcode, |
| 4284 |
HASH_FIND, NULL); |
| 4285 |
if (rmpredlock != NULL) |
| 4286 |
{ |
| 4287 |
Assert(rmpredlock == mypredlock); |
| 4288 |
|
| 4289 |
dlist_delete(&(mypredlock->targetLink)); |
| 4290 |
dlist_delete(&(mypredlock->xactLink)); |
| 4291 |
|
| 4292 |
rmpredlock = (PREDICATELOCK *) |
| 4293 |
hash_search_with_hash_value(PredicateLockHash, |
| 4294 |
&mypredlocktag, |
| 4295 |
predlockhashcode, |
| 4296 |
HASH_REMOVE, NULL); |
| 4297 |
Assert(rmpredlock == mypredlock); |
| 4298 |
|
| 4299 |
RemoveTargetIfNoLongerUsed(target, targettaghash); |
| 4300 |
} |
| 4301 |
|
| 4302 |
LWLockRelease(SerializableXactHashLock); |
| 4303 |
LWLockRelease(partitionLock); |
| 4304 |
if (IsInParallelMode()) |
| 4305 |
LWLockRelease(&MySerializableXact->perXactPredicateListLock); |
| 4306 |
LWLockRelease(SerializablePredicateListLock); |
| 4307 |
|
| 4308 |
if (rmpredlock != NULL) |
| 4309 |
{ |
| 4310 |
/* |
| 4311 |
* Remove entry in local lock table if it exists. It's OK if it |
| 4312 |
* doesn't exist; that means the lock was transferred to a new |
| 4313 |
* target by a different backend. |
| 4314 |
*/ |
| 4315 |
hash_search_with_hash_value(LocalPredicateLockHash, |
| 4316 |
targettag, targettaghash, |
| 4317 |
HASH_REMOVE, NULL); |
| 4318 |
|
| 4319 |
DecrementParentLocks(targettag); |
| 4320 |
} |
| 4321 |
} |
| 4322 |
} |
| 4323 |
|
| 4324 |
/* |
| 4325 |
* CheckForSerializableConflictIn |
| 4326 |
* We are writing the given tuple. If that indicates a rw-conflict |
| 4327 |
* in from another serializable transaction, take appropriate action. |
| 4328 |
* |
| 4329 |
* Skip checking for any granularity for which a parameter is missing. |
| 4330 |
* |
| 4331 |
* A tuple update or delete is in conflict if we have a predicate lock |
| 4332 |
* against the relation or page in which the tuple exists, or against the |
| 4333 |
* tuple itself. |
| 4334 |
*/ |
| 4335 |
void |
| 4336 |
CheckForSerializableConflictIn(Relation relation, ItemPointer tid, BlockNumber blkno) |
| 4337 |
{ |
| 4338 |
PREDICATELOCKTARGETTAG targettag; |
| 4339 |
|
| 4340 |
if (!SerializationNeededForWrite(relation)) |
| 4341 |
return; |
| 4342 |
|
| 4343 |
/* Check if someone else has already decided that we need to die */ |
| 4344 |
if (SxactIsDoomed(MySerializableXact)) |
| 4345 |
ereport(ERROR, |
| 4346 |
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE), |
| 4347 |
errmsg("could not serialize access due to read/write dependencies among transactions"), |
| 4348 |
errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict in checking."), |
| 4349 |
errhint("The transaction might succeed if retried."))); |
| 4350 |
|
| 4351 |
/* |
| 4352 |
* We're doing a write which might cause rw-conflicts now or later. |
| 4353 |
* Memorize that fact. |
| 4354 |
*/ |
| 4355 |
MyXactDidWrite = true; |
| 4356 |
|
| 4357 |
/* |
| 4358 |
* It is important that we check for locks from the finest granularity to |
| 4359 |
* the coarsest granularity, so that granularity promotion doesn't cause |
| 4360 |
* us to miss a lock. The new (coarser) lock will be acquired before the |
| 4361 |
* old (finer) locks are released. |
| 4362 |
* |
| 4363 |
* It is not possible to take and hold a lock across the checks for all |
| 4364 |
* granularities because each target could be in a separate partition. |
| 4365 |
*/ |
| 4366 |
if (tid != NULL) |
| 4367 |
{ |
| 4368 |
SET_PREDICATELOCKTARGETTAG_TUPLE(targettag, |
| 4369 |
relation->rd_locator.dbOid, |
| 4370 |
relation->rd_id, |
| 4371 |
ItemPointerGetBlockNumber(tid), |
| 4372 |
ItemPointerGetOffsetNumber(tid)); |
| 4373 |
CheckTargetForConflictsIn(&targettag); |
| 4374 |
} |
| 4375 |
|
| 4376 |
if (blkno != InvalidBlockNumber) |
| 4377 |
{ |
| 4378 |
SET_PREDICATELOCKTARGETTAG_PAGE(targettag, |
| 4379 |
relation->rd_locator.dbOid, |
| 4380 |
relation->rd_id, |
| 4381 |
blkno); |
| 4382 |
CheckTargetForConflictsIn(&targettag); |
| 4383 |
} |
| 4384 |
|
| 4385 |
SET_PREDICATELOCKTARGETTAG_RELATION(targettag, |
| 4386 |
relation->rd_locator.dbOid, |
| 4387 |
relation->rd_id); |
| 4388 |
CheckTargetForConflictsIn(&targettag); |
| 4389 |
} |
| 4390 |
|
| 4391 |
/* |
| 4392 |
* CheckTableForSerializableConflictIn |
| 4393 |
* The entire table is going through a DDL-style logical mass delete |
| 4394 |
* like TRUNCATE or DROP TABLE. If that causes a rw-conflict in from |
| 4395 |
* another serializable transaction, take appropriate action. |
| 4396 |
* |
| 4397 |
* While these operations do not operate entirely within the bounds of |
| 4398 |
* snapshot isolation, they can occur inside a serializable transaction, and |
| 4399 |
* will logically occur after any reads which saw rows which were destroyed |
| 4400 |
* by these operations, so we do what we can to serialize properly under |
| 4401 |
* SSI. |
| 4402 |
* |
| 4403 |
* The relation passed in must be a heap relation. Any predicate lock of any |
| 4404 |
* granularity on the heap will cause a rw-conflict in to this transaction. |
| 4405 |
* Predicate locks on indexes do not matter because they only exist to guard |
| 4406 |
* against conflicting inserts into the index, and this is a mass *delete*. |
| 4407 |
* When a table is truncated or dropped, the index will also be truncated |
| 4408 |
* or dropped, and we'll deal with locks on the index when that happens. |
| 4409 |
* |
| 4410 |
* Dropping or truncating a table also needs to drop any existing predicate |
| 4411 |
* locks on heap tuples or pages, because they're about to go away. This |
| 4412 |
* should be done before altering the predicate locks because the transaction |
| 4413 |
* could be rolled back because of a conflict, in which case the lock changes |
| 4414 |
* are not needed. (At the moment, we don't actually bother to drop the |
| 4415 |
* existing locks on a dropped or truncated table at the moment. That might |
| 4416 |
* lead to some false positives, but it doesn't seem worth the trouble.) |
| 4417 |
*/ |
| 4418 |
void |
| 4419 |
CheckTableForSerializableConflictIn(Relation relation) |
| 4420 |
{ |
| 4421 |
HASH_SEQ_STATUS seqstat; |
| 4422 |
PREDICATELOCKTARGET *target; |
| 4423 |
Oid dbId; |
| 4424 |
Oid heapId; |
| 4425 |
int i; |
| 4426 |
|
| 4427 |
/* |
| 4428 |
* Bail out quickly if there are no serializable transactions running. |
| 4429 |
* It's safe to check this without taking locks because the caller is |
| 4430 |
* holding an ACCESS EXCLUSIVE lock on the relation. No new locks which |
| 4431 |
* would matter here can be acquired while that is held. |
| 4432 |
*/ |
| 4433 |
if (!TransactionIdIsValid(PredXact->SxactGlobalXmin)) |
| 4434 |
return; |
| 4435 |
|
| 4436 |
if (!SerializationNeededForWrite(relation)) |
| 4437 |
return; |
| 4438 |
|
| 4439 |
/* |
| 4440 |
* We're doing a write which might cause rw-conflicts now or later. |
| 4441 |
* Memorize that fact. |
| 4442 |
*/ |
| 4443 |
MyXactDidWrite = true; |
| 4444 |
|
| 4445 |
Assert(relation->rd_index == NULL); /* not an index relation */ |
| 4446 |
|
| 4447 |
dbId = relation->rd_locator.dbOid; |
| 4448 |
heapId = relation->rd_id; |
| 4449 |
|
| 4450 |
LWLockAcquire(SerializablePredicateListLock, LW_EXCLUSIVE); |
| 4451 |
for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++) |
| 4452 |
LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_SHARED); |
| 4453 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 4454 |
|
| 4455 |
/* Scan through target list */ |
| 4456 |
hash_seq_init(&seqstat, PredicateLockTargetHash); |
| 4457 |
|
| 4458 |
while ((target = (PREDICATELOCKTARGET *) hash_seq_search(&seqstat))) |
| 4459 |
{ |
| 4460 |
dlist_mutable_iter iter; |
| 4461 |
|
| 4462 |
/* |
| 4463 |
* Check whether this is a target which needs attention. |
| 4464 |
*/ |
| 4465 |
if (GET_PREDICATELOCKTARGETTAG_RELATION(target->tag) != heapId) |
| 4466 |
continue; /* wrong relation id */ |
| 4467 |
if (GET_PREDICATELOCKTARGETTAG_DB(target->tag) != dbId) |
| 4468 |
continue; /* wrong database id */ |
| 4469 |
|
| 4470 |
/* |
| 4471 |
* Loop through locks for this target and flag conflicts. |
| 4472 |
*/ |
| 4473 |
dlist_foreach_modify(iter, &target->predicateLocks) |
| 4474 |
{ |
| 4475 |
PREDICATELOCK *predlock = |
| 4476 |
dlist_container(PREDICATELOCK, targetLink, iter.cur); |
| 4477 |
|
| 4478 |
if (predlock->tag.myXact != MySerializableXact |
| 4479 |
&& !RWConflictExists(predlock->tag.myXact, MySerializableXact)) |
| 4480 |
{ |
| 4481 |
FlagRWConflict(predlock->tag.myXact, MySerializableXact); |
| 4482 |
} |
| 4483 |
} |
| 4484 |
} |
| 4485 |
|
| 4486 |
/* Release locks in reverse order */ |
| 4487 |
LWLockRelease(SerializableXactHashLock); |
| 4488 |
for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--) |
| 4489 |
LWLockRelease(PredicateLockHashPartitionLockByIndex(i)); |
| 4490 |
LWLockRelease(SerializablePredicateListLock); |
| 4491 |
} |
| 4492 |
|
| 4493 |
|
| 4494 |
/* |
| 4495 |
* Flag a rw-dependency between two serializable transactions. |
| 4496 |
* |
| 4497 |
* The caller is responsible for ensuring that we have a LW lock on |
| 4498 |
* the transaction hash table. |
| 4499 |
*/ |
| 4500 |
static void |
| 4501 |
FlagRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer) |
| 4502 |
{ |
| 4503 |
Assert(reader != writer); |
| 4504 |
|
| 4505 |
/* First, see if this conflict causes failure. */ |
| 4506 |
OnConflict_CheckForSerializationFailure(reader, writer); |
| 4507 |
|
| 4508 |
/* Actually do the conflict flagging. */ |
| 4509 |
if (reader == OldCommittedSxact) |
| 4510 |
writer->flags |= SXACT_FLAG_SUMMARY_CONFLICT_IN; |
| 4511 |
else if (writer == OldCommittedSxact) |
| 4512 |
reader->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT; |
| 4513 |
else |
| 4514 |
SetRWConflict(reader, writer); |
| 4515 |
} |
| 4516 |
|
| 4517 |
/*---------------------------------------------------------------------------- |
| 4518 |
* We are about to add a RW-edge to the dependency graph - check that we don't |
| 4519 |
* introduce a dangerous structure by doing so, and abort one of the |
| 4520 |
* transactions if so. |
| 4521 |
* |
| 4522 |
* A serialization failure can only occur if there is a dangerous structure |
| 4523 |
* in the dependency graph: |
| 4524 |
* |
| 4525 |
* Tin ------> Tpivot ------> Tout |
| 4526 |
* rw rw |
| 4527 |
* |
| 4528 |
* Furthermore, Tout must commit first. |
| 4529 |
* |
| 4530 |
* One more optimization is that if Tin is declared READ ONLY (or commits |
| 4531 |
* without writing), we can only have a problem if Tout committed before Tin |
| 4532 |
* acquired its snapshot. |
| 4533 |
*---------------------------------------------------------------------------- |
| 4534 |
*/ |
| 4535 |
static void |
| 4536 |
OnConflict_CheckForSerializationFailure(const SERIALIZABLEXACT *reader, |
| 4537 |
SERIALIZABLEXACT *writer) |
| 4538 |
{ |
| 4539 |
bool failure; |
| 4540 |
|
| 4541 |
Assert(LWLockHeldByMe(SerializableXactHashLock)); |
| 4542 |
|
| 4543 |
failure = false; |
| 4544 |
|
| 4545 |
/*------------------------------------------------------------------------ |
| 4546 |
* Check for already-committed writer with rw-conflict out flagged |
| 4547 |
* (conflict-flag on W means that T2 committed before W): |
| 4548 |
* |
| 4549 |
* R ------> W ------> T2 |
| 4550 |
* rw rw |
| 4551 |
* |
| 4552 |
* That is a dangerous structure, so we must abort. (Since the writer |
| 4553 |
* has already committed, we must be the reader) |
| 4554 |
*------------------------------------------------------------------------ |
| 4555 |
*/ |
| 4556 |
if (SxactIsCommitted(writer) |
| 4557 |
&& (SxactHasConflictOut(writer) || SxactHasSummaryConflictOut(writer))) |
| 4558 |
failure = true; |
| 4559 |
|
| 4560 |
/*------------------------------------------------------------------------ |
| 4561 |
* Check whether the writer has become a pivot with an out-conflict |
| 4562 |
* committed transaction (T2), and T2 committed first: |
| 4563 |
* |
| 4564 |
* R ------> W ------> T2 |
| 4565 |
* rw rw |
| 4566 |
* |
| 4567 |
* Because T2 must've committed first, there is no anomaly if: |
| 4568 |
* - the reader committed before T2 |
| 4569 |
* - the writer committed before T2 |
| 4570 |
* - the reader is a READ ONLY transaction and the reader was concurrent |
| 4571 |
* with T2 (= reader acquired its snapshot before T2 committed) |
| 4572 |
* |
| 4573 |
* We also handle the case that T2 is prepared but not yet committed |
| 4574 |
* here. In that case T2 has already checked for conflicts, so if it |
| 4575 |
* commits first, making the above conflict real, it's too late for it |
| 4576 |
* to abort. |
| 4577 |
*------------------------------------------------------------------------ |
| 4578 |
*/ |
| 4579 |
if (!failure && SxactHasSummaryConflictOut(writer)) |
| 4580 |
failure = true; |
| 4581 |
else if (!failure) |
| 4582 |
{ |
| 4583 |
dlist_iter iter; |
| 4584 |
|
| 4585 |
dlist_foreach(iter, &writer->outConflicts) |
| 4586 |
{ |
| 4587 |
RWConflict conflict = |
| 4588 |
dlist_container(RWConflictData, outLink, iter.cur); |
| 4589 |
SERIALIZABLEXACT *t2 = conflict->sxactIn; |
| 4590 |
|
| 4591 |
if (SxactIsPrepared(t2) |
| 4592 |
&& (!SxactIsCommitted(reader) |
| 4593 |
|| t2->prepareSeqNo <= reader->commitSeqNo) |
| 4594 |
&& (!SxactIsCommitted(writer) |
| 4595 |
|| t2->prepareSeqNo <= writer->commitSeqNo) |
| 4596 |
&& (!SxactIsReadOnly(reader) |
| 4597 |
|| t2->prepareSeqNo <= reader->SeqNo.lastCommitBeforeSnapshot)) |
| 4598 |
{ |
| 4599 |
failure = true; |
| 4600 |
break; |
| 4601 |
} |
| 4602 |
} |
| 4603 |
} |
| 4604 |
|
| 4605 |
/*------------------------------------------------------------------------ |
| 4606 |
* Check whether the reader has become a pivot with a writer |
| 4607 |
* that's committed (or prepared): |
| 4608 |
* |
| 4609 |
* T0 ------> R ------> W |
| 4610 |
* rw rw |
| 4611 |
* |
| 4612 |
* Because W must've committed first for an anomaly to occur, there is no |
| 4613 |
* anomaly if: |
| 4614 |
* - T0 committed before the writer |
| 4615 |
* - T0 is READ ONLY, and overlaps the writer |
| 4616 |
*------------------------------------------------------------------------ |
| 4617 |
*/ |
| 4618 |
if (!failure && SxactIsPrepared(writer) && !SxactIsReadOnly(reader)) |
| 4619 |
{ |
| 4620 |
if (SxactHasSummaryConflictIn(reader)) |
| 4621 |
{ |
| 4622 |
failure = true; |
| 4623 |
} |
| 4624 |
else |
| 4625 |
{ |
| 4626 |
dlist_iter iter; |
| 4627 |
|
| 4628 |
/* |
| 4629 |
* The unconstify is needed as we have no const version of |
| 4630 |
* dlist_foreach(). |
| 4631 |
*/ |
| 4632 |
dlist_foreach(iter, &unconstify(SERIALIZABLEXACT *, reader)->inConflicts) |
| 4633 |
{ |
| 4634 |
const RWConflict conflict = |
| 4635 |
dlist_container(RWConflictData, inLink, iter.cur); |
| 4636 |
const SERIALIZABLEXACT *t0 = conflict->sxactOut; |
| 4637 |
|
| 4638 |
if (!SxactIsDoomed(t0) |
| 4639 |
&& (!SxactIsCommitted(t0) |
| 4640 |
|| t0->commitSeqNo >= writer->prepareSeqNo) |
| 4641 |
&& (!SxactIsReadOnly(t0) |
| 4642 |
|| t0->SeqNo.lastCommitBeforeSnapshot >= writer->prepareSeqNo)) |
| 4643 |
{ |
| 4644 |
failure = true; |
| 4645 |
break; |
| 4646 |
} |
| 4647 |
} |
| 4648 |
} |
| 4649 |
} |
| 4650 |
|
| 4651 |
if (failure) |
| 4652 |
{ |
| 4653 |
/* |
| 4654 |
* We have to kill a transaction to avoid a possible anomaly from |
| 4655 |
* occurring. If the writer is us, we can just ereport() to cause a |
| 4656 |
* transaction abort. Otherwise we flag the writer for termination, |
| 4657 |
* causing it to abort when it tries to commit. However, if the writer |
| 4658 |
* is a prepared transaction, already prepared, we can't abort it |
| 4659 |
* anymore, so we have to kill the reader instead. |
| 4660 |
*/ |
| 4661 |
if (MySerializableXact == writer) |
| 4662 |
{ |
| 4663 |
LWLockRelease(SerializableXactHashLock); |
| 4664 |
ereport(ERROR, |
| 4665 |
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE), |
| 4666 |
errmsg("could not serialize access due to read/write dependencies among transactions"), |
| 4667 |
errdetail_internal("Reason code: Canceled on identification as a pivot, during write."), |
| 4668 |
errhint("The transaction might succeed if retried."))); |
| 4669 |
} |
| 4670 |
else if (SxactIsPrepared(writer)) |
| 4671 |
{ |
| 4672 |
LWLockRelease(SerializableXactHashLock); |
| 4673 |
|
| 4674 |
/* if we're not the writer, we have to be the reader */ |
| 4675 |
Assert(MySerializableXact == reader); |
| 4676 |
ereport(ERROR, |
| 4677 |
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE), |
| 4678 |
errmsg("could not serialize access due to read/write dependencies among transactions"), |
| 4679 |
errdetail_internal("Reason code: Canceled on conflict out to pivot %u, during read.", writer->topXid), |
| 4680 |
errhint("The transaction might succeed if retried."))); |
| 4681 |
} |
| 4682 |
writer->flags |= SXACT_FLAG_DOOMED; |
| 4683 |
} |
| 4684 |
} |
| 4685 |
|
| 4686 |
/* |
| 4687 |
* PreCommit_CheckForSerializationFailure |
| 4688 |
* Check for dangerous structures in a serializable transaction |
| 4689 |
* at commit. |
| 4690 |
* |
| 4691 |
* We're checking for a dangerous structure as each conflict is recorded. |
| 4692 |
* The only way we could have a problem at commit is if this is the "out" |
| 4693 |
* side of a pivot, and neither the "in" side nor the pivot has yet |
| 4694 |
* committed. |
| 4695 |
* |
| 4696 |
* If a dangerous structure is found, the pivot (the near conflict) is |
| 4697 |
* marked for death, because rolling back another transaction might mean |
| 4698 |
* that we fail without ever making progress. This transaction is |
| 4699 |
* committing writes, so letting it commit ensures progress. If we |
| 4700 |
* canceled the far conflict, it might immediately fail again on retry. |
| 4701 |
*/ |
| 4702 |
void |
| 4703 |
PreCommit_CheckForSerializationFailure(void) |
| 4704 |
{ |
| 4705 |
dlist_iter near_iter; |
| 4706 |
|
| 4707 |
if (MySerializableXact == InvalidSerializableXact) |
| 4708 |
return; |
| 4709 |
|
| 4710 |
Assert(IsolationIsSerializable()); |
| 4711 |
|
| 4712 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 4713 |
|
| 4714 |
/* |
| 4715 |
* Check if someone else has already decided that we need to die. Since |
| 4716 |
* we set our own DOOMED flag when partially releasing, ignore in that |
| 4717 |
* case. |
| 4718 |
*/ |
| 4719 |
if (SxactIsDoomed(MySerializableXact) && |
| 4720 |
!SxactIsPartiallyReleased(MySerializableXact)) |
| 4721 |
{ |
| 4722 |
LWLockRelease(SerializableXactHashLock); |
| 4723 |
ereport(ERROR, |
| 4724 |
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE), |
| 4725 |
errmsg("could not serialize access due to read/write dependencies among transactions"), |
| 4726 |
errdetail_internal("Reason code: Canceled on identification as a pivot, during commit attempt."), |
| 4727 |
errhint("The transaction might succeed if retried."))); |
| 4728 |
} |
| 4729 |
|
| 4730 |
dlist_foreach(near_iter, &MySerializableXact->inConflicts) |
| 4731 |
{ |
| 4732 |
RWConflict nearConflict = |
| 4733 |
dlist_container(RWConflictData, inLink, near_iter.cur); |
| 4734 |
|
| 4735 |
if (!SxactIsCommitted(nearConflict->sxactOut) |
| 4736 |
&& !SxactIsDoomed(nearConflict->sxactOut)) |
| 4737 |
{ |
| 4738 |
dlist_iter far_iter; |
| 4739 |
|
| 4740 |
dlist_foreach(far_iter, &nearConflict->sxactOut->inConflicts) |
| 4741 |
{ |
| 4742 |
RWConflict farConflict = |
| 4743 |
dlist_container(RWConflictData, inLink, far_iter.cur); |
| 4744 |
|
| 4745 |
if (farConflict->sxactOut == MySerializableXact |
| 4746 |
|| (!SxactIsCommitted(farConflict->sxactOut) |
| 4747 |
&& !SxactIsReadOnly(farConflict->sxactOut) |
| 4748 |
&& !SxactIsDoomed(farConflict->sxactOut))) |
| 4749 |
{ |
| 4750 |
/* |
| 4751 |
* Normally, we kill the pivot transaction to make sure we |
| 4752 |
* make progress if the failing transaction is retried. |
| 4753 |
* However, we can't kill it if it's already prepared, so |
| 4754 |
* in that case we commit suicide instead. |
| 4755 |
*/ |
| 4756 |
if (SxactIsPrepared(nearConflict->sxactOut)) |
| 4757 |
{ |
| 4758 |
LWLockRelease(SerializableXactHashLock); |
| 4759 |
ereport(ERROR, |
| 4760 |
(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE), |
| 4761 |
errmsg("could not serialize access due to read/write dependencies among transactions"), |
| 4762 |
errdetail_internal("Reason code: Canceled on commit attempt with conflict in from prepared pivot."), |
| 4763 |
errhint("The transaction might succeed if retried."))); |
| 4764 |
} |
| 4765 |
nearConflict->sxactOut->flags |= SXACT_FLAG_DOOMED; |
| 4766 |
break; |
| 4767 |
} |
| 4768 |
} |
| 4769 |
} |
| 4770 |
} |
| 4771 |
|
| 4772 |
MySerializableXact->prepareSeqNo = ++(PredXact->LastSxactCommitSeqNo); |
| 4773 |
MySerializableXact->flags |= SXACT_FLAG_PREPARED; |
| 4774 |
|
| 4775 |
LWLockRelease(SerializableXactHashLock); |
| 4776 |
} |
| 4777 |
|
| 4778 |
/*------------------------------------------------------------------------*/ |
| 4779 |
|
| 4780 |
/* |
| 4781 |
* Two-phase commit support |
| 4782 |
*/ |
| 4783 |
|
| 4784 |
/* |
| 4785 |
* AtPrepare_Locks |
| 4786 |
* Do the preparatory work for a PREPARE: make 2PC state file |
| 4787 |
* records for all predicate locks currently held. |
| 4788 |
*/ |
| 4789 |
void |
| 4790 |
AtPrepare_PredicateLocks(void) |
| 4791 |
{ |
| 4792 |
SERIALIZABLEXACT *sxact; |
| 4793 |
TwoPhasePredicateRecord record; |
| 4794 |
TwoPhasePredicateXactRecord *xactRecord; |
| 4795 |
TwoPhasePredicateLockRecord *lockRecord; |
| 4796 |
dlist_iter iter; |
| 4797 |
|
| 4798 |
sxact = MySerializableXact; |
| 4799 |
xactRecord = &(record.data.xactRecord); |
| 4800 |
lockRecord = &(record.data.lockRecord); |
| 4801 |
|
| 4802 |
if (MySerializableXact == InvalidSerializableXact) |
| 4803 |
return; |
| 4804 |
|
| 4805 |
/* Generate an xact record for our SERIALIZABLEXACT */ |
| 4806 |
record.type = TWOPHASEPREDICATERECORD_XACT; |
| 4807 |
xactRecord->xmin = MySerializableXact->xmin; |
| 4808 |
xactRecord->flags = MySerializableXact->flags; |
| 4809 |
|
| 4810 |
/* |
| 4811 |
* Note that we don't include the list of conflicts in our out in the |
| 4812 |
* statefile, because new conflicts can be added even after the |
| 4813 |
* transaction prepares. We'll just make a conservative assumption during |
| 4814 |
* recovery instead. |
| 4815 |
*/ |
| 4816 |
|
| 4817 |
RegisterTwoPhaseRecord(TWOPHASE_RM_PREDICATELOCK_ID, 0, |
| 4818 |
&record, sizeof(record)); |
| 4819 |
|
| 4820 |
/* |
| 4821 |
* Generate a lock record for each lock. |
| 4822 |
* |
| 4823 |
* To do this, we need to walk the predicate lock list in our sxact rather |
| 4824 |
* than using the local predicate lock table because the latter is not |
| 4825 |
* guaranteed to be accurate. |
| 4826 |
*/ |
| 4827 |
LWLockAcquire(SerializablePredicateListLock, LW_SHARED); |
| 4828 |
|
| 4829 |
/* |
| 4830 |
* No need to take sxact->perXactPredicateListLock in parallel mode |
| 4831 |
* because there cannot be any parallel workers running while we are |
| 4832 |
* preparing a transaction. |
| 4833 |
*/ |
| 4834 |
Assert(!IsParallelWorker() && !ParallelContextActive()); |
| 4835 |
|
| 4836 |
dlist_foreach(iter, &sxact->predicateLocks) |
| 4837 |
{ |
| 4838 |
PREDICATELOCK *predlock = |
| 4839 |
dlist_container(PREDICATELOCK, xactLink, iter.cur); |
| 4840 |
|
| 4841 |
record.type = TWOPHASEPREDICATERECORD_LOCK; |
| 4842 |
lockRecord->target = predlock->tag.myTarget->tag; |
| 4843 |
|
| 4844 |
RegisterTwoPhaseRecord(TWOPHASE_RM_PREDICATELOCK_ID, 0, |
| 4845 |
&record, sizeof(record)); |
| 4846 |
} |
| 4847 |
|
| 4848 |
LWLockRelease(SerializablePredicateListLock); |
| 4849 |
} |
| 4850 |
|
| 4851 |
/* |
| 4852 |
* PostPrepare_Locks |
| 4853 |
* Clean up after successful PREPARE. Unlike the non-predicate |
| 4854 |
* lock manager, we do not need to transfer locks to a dummy |
| 4855 |
* PGPROC because our SERIALIZABLEXACT will stay around |
| 4856 |
* anyway. We only need to clean up our local state. |
| 4857 |
*/ |
| 4858 |
void |
| 4859 |
PostPrepare_PredicateLocks(TransactionId xid) |
| 4860 |
{ |
| 4861 |
if (MySerializableXact == InvalidSerializableXact) |
| 4862 |
return; |
| 4863 |
|
| 4864 |
Assert(SxactIsPrepared(MySerializableXact)); |
| 4865 |
|
| 4866 |
MySerializableXact->pid = 0; |
| 4867 |
MySerializableXact->pgprocno = INVALID_PROC_NUMBER; |
| 4868 |
|
| 4869 |
hash_destroy(LocalPredicateLockHash); |
| 4870 |
LocalPredicateLockHash = NULL; |
| 4871 |
|
| 4872 |
MySerializableXact = InvalidSerializableXact; |
| 4873 |
MyXactDidWrite = false; |
| 4874 |
} |
| 4875 |
|
| 4876 |
/* |
| 4877 |
* PredicateLockTwoPhaseFinish |
| 4878 |
* Release a prepared transaction's predicate locks once it |
| 4879 |
* commits or aborts. |
| 4880 |
*/ |
| 4881 |
void |
| 4882 |
PredicateLockTwoPhaseFinish(TransactionId xid, bool isCommit) |
| 4883 |
{ |
| 4884 |
SERIALIZABLEXID *sxid; |
| 4885 |
SERIALIZABLEXIDTAG sxidtag; |
| 4886 |
|
| 4887 |
sxidtag.xid = xid; |
| 4888 |
|
| 4889 |
LWLockAcquire(SerializableXactHashLock, LW_SHARED); |
| 4890 |
sxid = (SERIALIZABLEXID *) |
| 4891 |
hash_search(SerializableXidHash, &sxidtag, HASH_FIND, NULL); |
| 4892 |
LWLockRelease(SerializableXactHashLock); |
| 4893 |
|
| 4894 |
/* xid will not be found if it wasn't a serializable transaction */ |
| 4895 |
if (sxid == NULL) |
| 4896 |
return; |
| 4897 |
|
| 4898 |
/* Release its locks */ |
| 4899 |
MySerializableXact = sxid->myXact; |
| 4900 |
MyXactDidWrite = true; /* conservatively assume that we wrote |
| 4901 |
* something */ |
| 4902 |
ReleasePredicateLocks(isCommit, false); |
| 4903 |
} |
| 4904 |
|
| 4905 |
/* |
| 4906 |
* Re-acquire a predicate lock belonging to a transaction that was prepared. |
| 4907 |
*/ |
| 4908 |
void |
| 4909 |
predicatelock_twophase_recover(TransactionId xid, uint16 info, |
| 4910 |
void *recdata, uint32 len) |
| 4911 |
{ |
| 4912 |
TwoPhasePredicateRecord *record; |
| 4913 |
|
| 4914 |
Assert(len == sizeof(TwoPhasePredicateRecord)); |
| 4915 |
|
| 4916 |
record = (TwoPhasePredicateRecord *) recdata; |
| 4917 |
|
| 4918 |
Assert((record->type == TWOPHASEPREDICATERECORD_XACT) || |
| 4919 |
(record->type == TWOPHASEPREDICATERECORD_LOCK)); |
| 4920 |
|
| 4921 |
if (record->type == TWOPHASEPREDICATERECORD_XACT) |
| 4922 |
{ |
| 4923 |
/* Per-transaction record. Set up a SERIALIZABLEXACT. */ |
| 4924 |
TwoPhasePredicateXactRecord *xactRecord; |
| 4925 |
SERIALIZABLEXACT *sxact; |
| 4926 |
SERIALIZABLEXID *sxid; |
| 4927 |
SERIALIZABLEXIDTAG sxidtag; |
| 4928 |
bool found; |
| 4929 |
|
| 4930 |
xactRecord = (TwoPhasePredicateXactRecord *) &record->data.xactRecord; |
| 4931 |
|
| 4932 |
LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE); |
| 4933 |
sxact = CreatePredXact(); |
| 4934 |
if (!sxact) |
| 4935 |
ereport(ERROR, |
| 4936 |
(errcode(ERRCODE_OUT_OF_MEMORY), |
| 4937 |
errmsg("out of shared memory"))); |
| 4938 |
|
| 4939 |
/* vxid for a prepared xact is INVALID_PROC_NUMBER/xid; no pid */ |
| 4940 |
sxact->vxid.procNumber = INVALID_PROC_NUMBER; |
| 4941 |
sxact->vxid.localTransactionId = (LocalTransactionId) xid; |
| 4942 |
sxact->pid = 0; |
| 4943 |
sxact->pgprocno = INVALID_PROC_NUMBER; |
| 4944 |
|
| 4945 |
/* a prepared xact hasn't committed yet */ |
| 4946 |
sxact->prepareSeqNo = RecoverySerCommitSeqNo; |
| 4947 |
sxact->commitSeqNo = InvalidSerCommitSeqNo; |
| 4948 |
sxact->finishedBefore = InvalidTransactionId; |
| 4949 |
|
| 4950 |
sxact->SeqNo.lastCommitBeforeSnapshot = RecoverySerCommitSeqNo; |
| 4951 |
|
| 4952 |
/* |
| 4953 |
* Don't need to track this; no transactions running at the time the |
| 4954 |
* recovered xact started are still active, except possibly other |
| 4955 |
* prepared xacts and we don't care whether those are RO_SAFE or not. |
| 4956 |
*/ |
| 4957 |
dlist_init(&(sxact->possibleUnsafeConflicts)); |
| 4958 |
|
| 4959 |
dlist_init(&(sxact->predicateLocks)); |
| 4960 |
dlist_node_init(&sxact->finishedLink); |
| 4961 |
|
| 4962 |
sxact->topXid = xid; |
| 4963 |
sxact->xmin = xactRecord->xmin; |
| 4964 |
sxact->flags = xactRecord->flags; |
| 4965 |
Assert(SxactIsPrepared(sxact)); |
| 4966 |
if (!SxactIsReadOnly(sxact)) |
| 4967 |
{ |
| 4968 |
++(PredXact->WritableSxactCount); |
| 4969 |
Assert(PredXact->WritableSxactCount <= |
| 4970 |
(MaxBackends + max_prepared_xacts)); |
| 4971 |
} |
| 4972 |
|
| 4973 |
/* |
| 4974 |
* We don't know whether the transaction had any conflicts or not, so |
| 4975 |
* we'll conservatively assume that it had both a conflict in and a |
| 4976 |
* conflict out, and represent that with the summary conflict flags. |
| 4977 |
*/ |
| 4978 |
dlist_init(&(sxact->outConflicts)); |
| 4979 |
dlist_init(&(sxact->inConflicts)); |
| 4980 |
sxact->flags |= SXACT_FLAG_SUMMARY_CONFLICT_IN; |
| 4981 |
sxact->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT; |
| 4982 |
|
| 4983 |
/* Register the transaction's xid */ |
| 4984 |
sxidtag.xid = xid; |
| 4985 |
sxid = (SERIALIZABLEXID *) hash_search(SerializableXidHash, |
| 4986 |
&sxidtag, |
| 4987 |
HASH_ENTER, &found); |
| 4988 |
Assert(sxid != NULL); |
| 4989 |
Assert(!found); |
| 4990 |
sxid->myXact = (SERIALIZABLEXACT *) sxact; |
| 4991 |
|
| 4992 |
/* |
| 4993 |
* Update global xmin. Note that this is a special case compared to |
| 4994 |
* registering a normal transaction, because the global xmin might go |
| 4995 |
* backwards. That's OK, because until recovery is over we're not |
| 4996 |
* going to complete any transactions or create any non-prepared |
| 4997 |
* transactions, so there's no danger of throwing away. |
| 4998 |
*/ |
| 4999 |
if ((!TransactionIdIsValid(PredXact->SxactGlobalXmin)) || |
| 5000 |
(TransactionIdFollows(PredXact->SxactGlobalXmin, sxact->xmin))) |
| 5001 |
{ |
| 5002 |
PredXact->SxactGlobalXmin = sxact->xmin; |
| 5003 |
PredXact->SxactGlobalXminCount = 1; |
| 5004 |
SerialSetActiveSerXmin(sxact->xmin); |
| 5005 |
} |
| 5006 |
else if (TransactionIdEquals(sxact->xmin, PredXact->SxactGlobalXmin)) |
| 5007 |
{ |
| 5008 |
Assert(PredXact->SxactGlobalXminCount > 0); |
| 5009 |
PredXact->SxactGlobalXminCount++; |
| 5010 |
} |
| 5011 |
|
| 5012 |
LWLockRelease(SerializableXactHashLock); |
| 5013 |
} |
| 5014 |
else if (record->type == TWOPHASEPREDICATERECORD_LOCK) |
| 5015 |
{ |
| 5016 |
/* Lock record. Recreate the PREDICATELOCK */ |
| 5017 |
TwoPhasePredicateLockRecord *lockRecord; |
| 5018 |
SERIALIZABLEXID *sxid; |
| 5019 |
SERIALIZABLEXACT *sxact; |
| 5020 |
SERIALIZABLEXIDTAG sxidtag; |
| 5021 |
uint32 targettaghash; |
| 5022 |
|
| 5023 |
lockRecord = (TwoPhasePredicateLockRecord *) &record->data.lockRecord; |
| 5024 |
targettaghash = PredicateLockTargetTagHashCode(&lockRecord->target); |
| 5025 |
|
| 5026 |
LWLockAcquire(SerializableXactHashLock, LW_SHARED); |
| 5027 |
sxidtag.xid = xid; |
| 5028 |
sxid = (SERIALIZABLEXID *) |
| 5029 |
hash_search(SerializableXidHash, &sxidtag, HASH_FIND, NULL); |
| 5030 |
LWLockRelease(SerializableXactHashLock); |
| 5031 |
|
| 5032 |
Assert(sxid != NULL); |
| 5033 |
sxact = sxid->myXact; |
| 5034 |
Assert(sxact != InvalidSerializableXact); |
| 5035 |
|
| 5036 |
CreatePredicateLock(&lockRecord->target, targettaghash, sxact); |
| 5037 |
} |
| 5038 |
} |
| 5039 |
|
| 5040 |
/* |
| 5041 |
* Prepare to share the current SERIALIZABLEXACT with parallel workers. |
| 5042 |
* Return a handle object that can be used by AttachSerializableXact() in a |
| 5043 |
* parallel worker. |
| 5044 |
*/ |
| 5045 |
SerializableXactHandle |
| 5046 |
ShareSerializableXact(void) |
| 5047 |
{ |
| 5048 |
return MySerializableXact; |
| 5049 |
} |
| 5050 |
|
| 5051 |
/* |
| 5052 |
* Allow parallel workers to import the leader's SERIALIZABLEXACT. |
| 5053 |
*/ |
| 5054 |
void |
| 5055 |
AttachSerializableXact(SerializableXactHandle handle) |
| 5056 |
{ |
| 5057 |
|
| 5058 |
Assert(MySerializableXact == InvalidSerializableXact); |
| 5059 |
|
| 5060 |
MySerializableXact = (SERIALIZABLEXACT *) handle; |
| 5061 |
if (MySerializableXact != InvalidSerializableXact) |
| 5062 |
CreateLocalPredicateLockHash(); |
| 5063 |
} |