| 1 |
<!-- doc/src/sgml/fdwhandler.sgml --> |
| 2 |
|
| 3 |
<chapter id="fdwhandler"> |
| 4 |
<title>Writing a Foreign Data Wrapper</title> |
| 5 |
|
| 6 |
<indexterm zone="fdwhandler"> |
| 7 |
<primary>foreign data wrapper</primary> |
| 8 |
<secondary>handler for</secondary> |
| 9 |
</indexterm> |
| 10 |
|
| 11 |
<para> |
| 12 |
All operations on a foreign table are handled through its foreign data |
| 13 |
wrapper, which consists of a set of functions that the core server |
| 14 |
calls. The foreign data wrapper is responsible for fetching |
| 15 |
data from the remote data source and returning it to the |
| 16 |
<productname>PostgreSQL</productname> executor. If updating foreign |
| 17 |
tables is to be supported, the wrapper must handle that, too. |
| 18 |
This chapter outlines how to write a new foreign data wrapper. |
| 19 |
</para> |
| 20 |
|
| 21 |
<para> |
| 22 |
The foreign data wrappers included in the standard distribution are good |
| 23 |
references when trying to write your own. Look into the |
| 24 |
<filename>contrib</filename> subdirectory of the source tree. |
| 25 |
The <xref linkend="sql-createforeigndatawrapper"/> reference page also has |
| 26 |
some useful details. |
| 27 |
</para> |
| 28 |
|
| 29 |
<note> |
| 30 |
<para> |
| 31 |
The SQL standard specifies an interface for writing foreign data wrappers. |
| 32 |
However, PostgreSQL does not implement that API, because the effort to |
| 33 |
accommodate it into PostgreSQL would be large, and the standard API hasn't |
| 34 |
gained wide adoption anyway. |
| 35 |
</para> |
| 36 |
</note> |
| 37 |
|
| 38 |
<sect1 id="fdw-functions"> |
| 39 |
<title>Foreign Data Wrapper Functions</title> |
| 40 |
|
| 41 |
<para> |
| 42 |
The FDW author needs to implement a handler function, and optionally |
| 43 |
a validator function. Both functions must be written in a compiled |
| 44 |
language such as C, using the version-1 interface. |
| 45 |
For details on C language calling conventions and dynamic loading, |
| 46 |
see <xref linkend="xfunc-c"/>. |
| 47 |
</para> |
| 48 |
|
| 49 |
<para> |
| 50 |
The handler function simply returns a struct of function pointers to |
| 51 |
callback functions that will be called by the planner, executor, and |
| 52 |
various maintenance commands. |
| 53 |
Most of the effort in writing an FDW is in implementing these callback |
| 54 |
functions. |
| 55 |
The handler function must be registered with |
| 56 |
<productname>PostgreSQL</productname> as taking no arguments and |
| 57 |
returning the special pseudo-type <type>fdw_handler</type>. The |
| 58 |
callback functions are plain C functions and are not visible or |
| 59 |
callable at the SQL level. The callback functions are described in |
| 60 |
<xref linkend="fdw-callbacks"/>. |
| 61 |
</para> |
| 62 |
|
| 63 |
<para> |
| 64 |
The validator function is responsible for validating options given in |
| 65 |
<command>CREATE</command> and <command>ALTER</command> commands for its |
| 66 |
foreign data wrapper, as well as foreign servers, user mappings, and |
| 67 |
foreign tables using the wrapper. |
| 68 |
The validator function must be registered as taking two arguments, a |
| 69 |
text array containing the options to be validated, and an OID |
| 70 |
representing the type of object the options are associated with. The |
| 71 |
latter corresponds to the OID of the system catalog the object |
| 72 |
would be stored in, one of: |
| 73 |
<itemizedlist spacing="compact"> |
| 74 |
<listitem><para><literal>AttributeRelationId</literal></para></listitem> |
| 75 |
<listitem><para><literal>ForeignDataWrapperRelationId</literal></para></listitem> |
| 76 |
<listitem><para><literal>ForeignServerRelationId</literal></para></listitem> |
| 77 |
<listitem><para><literal>ForeignTableRelationId</literal></para></listitem> |
| 78 |
<listitem><para><literal>UserMappingRelationId</literal></para></listitem> |
| 79 |
</itemizedlist> |
| 80 |
If no validator function is supplied, options are not checked at object |
| 81 |
creation time or object alteration time. |
| 82 |
</para> |
| 83 |
|
| 84 |
</sect1> |
| 85 |
|
| 86 |
<sect1 id="fdw-callbacks"> |
| 87 |
<title>Foreign Data Wrapper Callback Routines</title> |
| 88 |
|
| 89 |
<para> |
| 90 |
The FDW handler function returns a palloc'd <structname>FdwRoutine</structname> |
| 91 |
struct containing pointers to the callback functions described below. |
| 92 |
The scan-related functions are required, the rest are optional. |
| 93 |
</para> |
| 94 |
|
| 95 |
<para> |
| 96 |
The <structname>FdwRoutine</structname> struct type is declared in |
| 97 |
<filename>src/include/foreign/fdwapi.h</filename>, which see for additional |
| 98 |
details. |
| 99 |
</para> |
| 100 |
|
| 101 |
<sect2 id="fdw-callbacks-scan"> |
| 102 |
<title>FDW Routines for Scanning Foreign Tables</title> |
| 103 |
|
| 104 |
<para> |
| 105 |
<programlisting> |
| 106 |
void |
| 107 |
GetForeignRelSize(PlannerInfo *root, |
| 108 |
RelOptInfo *baserel, |
| 109 |
Oid foreigntableid); |
| 110 |
</programlisting> |
| 111 |
|
| 112 |
Obtain relation size estimates for a foreign table. This is called |
| 113 |
at the beginning of planning for a query that scans a foreign table. |
| 114 |
<literal>root</literal> is the planner's global information about the query; |
| 115 |
<literal>baserel</literal> is the planner's information about this table; and |
| 116 |
<literal>foreigntableid</literal> is the <structname>pg_class</structname> OID of the |
| 117 |
foreign table. (<literal>foreigntableid</literal> could be obtained from the |
| 118 |
planner data structures, but it's passed explicitly to save effort.) |
| 119 |
</para> |
| 120 |
|
| 121 |
<para> |
| 122 |
This function should update <literal>baserel->rows</literal> to be the |
| 123 |
expected number of rows returned by the table scan, after accounting for |
| 124 |
the filtering done by the restriction quals. The initial value of |
| 125 |
<literal>baserel->rows</literal> is just a constant default estimate, which |
| 126 |
should be replaced if at all possible. The function may also choose to |
| 127 |
update <literal>baserel->width</literal> if it can compute a better estimate |
| 128 |
of the average result row width. |
| 129 |
(The initial value is based on column data types and on column |
| 130 |
average-width values measured by the last <command>ANALYZE</command>.) |
| 131 |
Also, this function may update <literal>baserel->tuples</literal> if |
| 132 |
it can compute a better estimate of the foreign table's total row count. |
| 133 |
(The initial value is |
| 134 |
from <structname>pg_class</structname>.<structfield>reltuples</structfield> |
| 135 |
which represents the total row count seen by the |
| 136 |
last <command>ANALYZE</command>; it will be <literal>-1</literal> if |
| 137 |
no <command>ANALYZE</command> has been done on this foreign table.) |
| 138 |
</para> |
| 139 |
|
| 140 |
<para> |
| 141 |
See <xref linkend="fdw-planning"/> for additional information. |
| 142 |
</para> |
| 143 |
|
| 144 |
<para> |
| 145 |
<programlisting> |
| 146 |
void |
| 147 |
GetForeignPaths(PlannerInfo *root, |
| 148 |
RelOptInfo *baserel, |
| 149 |
Oid foreigntableid); |
| 150 |
</programlisting> |
| 151 |
|
| 152 |
Create possible access paths for a scan on a foreign table. |
| 153 |
This is called during query planning. |
| 154 |
The parameters are the same as for <function>GetForeignRelSize</function>, |
| 155 |
which has already been called. |
| 156 |
</para> |
| 157 |
|
| 158 |
<para> |
| 159 |
This function must generate at least one access path |
| 160 |
(<structname>ForeignPath</structname> node) for a scan on the foreign table and |
| 161 |
must call <function>add_path</function> to add each such path to |
| 162 |
<literal>baserel->pathlist</literal>. It's recommended to use |
| 163 |
<function>create_foreignscan_path</function> to build the |
| 164 |
<structname>ForeignPath</structname> nodes. The function can generate multiple |
| 165 |
access paths, e.g., a path which has valid <literal>pathkeys</literal> to |
| 166 |
represent a pre-sorted result. Each access path must contain cost |
| 167 |
estimates, and can contain any FDW-private information that is needed to |
| 168 |
identify the specific scan method intended. |
| 169 |
</para> |
| 170 |
|
| 171 |
<para> |
| 172 |
See <xref linkend="fdw-planning"/> for additional information. |
| 173 |
</para> |
| 174 |
|
| 175 |
<para> |
| 176 |
<programlisting> |
| 177 |
ForeignScan * |
| 178 |
GetForeignPlan(PlannerInfo *root, |
| 179 |
RelOptInfo *baserel, |
| 180 |
Oid foreigntableid, |
| 181 |
ForeignPath *best_path, |
| 182 |
List *tlist, |
| 183 |
List *scan_clauses, |
| 184 |
Plan *outer_plan); |
| 185 |
</programlisting> |
| 186 |
|
| 187 |
Create a <structname>ForeignScan</structname> plan node from the selected foreign |
| 188 |
access path. This is called at the end of query planning. |
| 189 |
The parameters are as for <function>GetForeignRelSize</function>, plus |
| 190 |
the selected <structname>ForeignPath</structname> (previously produced by |
| 191 |
<function>GetForeignPaths</function>, <function>GetForeignJoinPaths</function>, |
| 192 |
or <function>GetForeignUpperPaths</function>), |
| 193 |
the target list to be emitted by the plan node, |
| 194 |
the restriction clauses to be enforced by the plan node, |
| 195 |
and the outer subplan of the <structname>ForeignScan</structname>, |
| 196 |
which is used for rechecks performed by <function>RecheckForeignScan</function>. |
| 197 |
(If the path is for a join rather than a base |
| 198 |
relation, <literal>foreigntableid</literal> is <literal>InvalidOid</literal>.) |
| 199 |
</para> |
| 200 |
|
| 201 |
<para> |
| 202 |
This function must create and return a <structname>ForeignScan</structname> plan |
| 203 |
node; it's recommended to use <function>make_foreignscan</function> to build the |
| 204 |
<structname>ForeignScan</structname> node. |
| 205 |
</para> |
| 206 |
|
| 207 |
<para> |
| 208 |
See <xref linkend="fdw-planning"/> for additional information. |
| 209 |
</para> |
| 210 |
|
| 211 |
<para> |
| 212 |
<programlisting> |
| 213 |
void |
| 214 |
BeginForeignScan(ForeignScanState *node, |
| 215 |
int eflags); |
| 216 |
</programlisting> |
| 217 |
|
| 218 |
Begin executing a foreign scan. This is called during executor startup. |
| 219 |
It should perform any initialization needed before the scan can start, |
| 220 |
but not start executing the actual scan (that should be done upon the |
| 221 |
first call to <function>IterateForeignScan</function>). |
| 222 |
The <structname>ForeignScanState</structname> node has already been created, but |
| 223 |
its <structfield>fdw_state</structfield> field is still NULL. Information about |
| 224 |
the table to scan is accessible through the |
| 225 |
<structname>ForeignScanState</structname> node (in particular, from the underlying |
| 226 |
<structname>ForeignScan</structname> plan node, which contains any FDW-private |
| 227 |
information provided by <function>GetForeignPlan</function>). |
| 228 |
<literal>eflags</literal> contains flag bits describing the executor's |
| 229 |
operating mode for this plan node. |
| 230 |
</para> |
| 231 |
|
| 232 |
<para> |
| 233 |
Note that when <literal>(eflags & EXEC_FLAG_EXPLAIN_ONLY)</literal> is |
| 234 |
true, this function should not perform any externally-visible actions; |
| 235 |
it should only do the minimum required to make the node state valid |
| 236 |
for <function>ExplainForeignScan</function> and <function>EndForeignScan</function>. |
| 237 |
</para> |
| 238 |
|
| 239 |
<para> |
| 240 |
<programlisting> |
| 241 |
TupleTableSlot * |
| 242 |
IterateForeignScan(ForeignScanState *node); |
| 243 |
</programlisting> |
| 244 |
|
| 245 |
Fetch one row from the foreign source, returning it in a tuple table slot |
| 246 |
(the node's <structfield>ScanTupleSlot</structfield> should be used for this |
| 247 |
purpose). Return NULL if no more rows are available. The tuple table |
| 248 |
slot infrastructure allows either a physical or virtual tuple to be |
| 249 |
returned; in most cases the latter choice is preferable from a |
| 250 |
performance standpoint. Note that this is called in a short-lived memory |
| 251 |
context that will be reset between invocations. Create a memory context |
| 252 |
in <function>BeginForeignScan</function> if you need longer-lived storage, or use |
| 253 |
the <structfield>es_query_cxt</structfield> of the node's <structname>EState</structname>. |
| 254 |
</para> |
| 255 |
|
| 256 |
<para> |
| 257 |
The rows returned must match the <structfield>fdw_scan_tlist</structfield> target |
| 258 |
list if one was supplied, otherwise they must match the row type of the |
| 259 |
foreign table being scanned. If you choose to optimize away fetching |
| 260 |
columns that are not needed, you should insert nulls in those column |
| 261 |
positions, or else generate a <structfield>fdw_scan_tlist</structfield> list with |
| 262 |
those columns omitted. |
| 263 |
</para> |
| 264 |
|
| 265 |
<para> |
| 266 |
Note that <productname>PostgreSQL</productname>'s executor doesn't care |
| 267 |
whether the rows returned violate any constraints that were defined on |
| 268 |
the foreign table — but the planner does care, and may optimize |
| 269 |
queries incorrectly if there are rows visible in the foreign table that |
| 270 |
do not satisfy a declared constraint. If a constraint is violated when |
| 271 |
the user has declared that the constraint should hold true, it may be |
| 272 |
appropriate to raise an error (just as you would need to do in the case |
| 273 |
of a data type mismatch). |
| 274 |
</para> |
| 275 |
|
| 276 |
<para> |
| 277 |
<programlisting> |
| 278 |
void |
| 279 |
ReScanForeignScan(ForeignScanState *node); |
| 280 |
</programlisting> |
| 281 |
|
| 282 |
Restart the scan from the beginning. Note that any parameters the |
| 283 |
scan depends on may have changed value, so the new scan does not |
| 284 |
necessarily return exactly the same rows. |
| 285 |
</para> |
| 286 |
|
| 287 |
<para> |
| 288 |
<programlisting> |
| 289 |
void |
| 290 |
EndForeignScan(ForeignScanState *node); |
| 291 |
</programlisting> |
| 292 |
|
| 293 |
End the scan and release resources. It is normally not important |
| 294 |
to release palloc'd memory, but for example open files and connections |
| 295 |
to remote servers should be cleaned up. |
| 296 |
</para> |
| 297 |
|
| 298 |
</sect2> |
| 299 |
|
| 300 |
<sect2 id="fdw-callbacks-join-scan"> |
| 301 |
<title>FDW Routines for Scanning Foreign Joins</title> |
| 302 |
|
| 303 |
<para> |
| 304 |
If an FDW supports performing foreign joins remotely (rather than |
| 305 |
by fetching both tables' data and doing the join locally), it should |
| 306 |
provide this callback function: |
| 307 |
</para> |
| 308 |
|
| 309 |
<para> |
| 310 |
<programlisting> |
| 311 |
void |
| 312 |
GetForeignJoinPaths(PlannerInfo *root, |
| 313 |
RelOptInfo *joinrel, |
| 314 |
RelOptInfo *outerrel, |
| 315 |
RelOptInfo *innerrel, |
| 316 |
JoinType jointype, |
| 317 |
JoinPathExtraData *extra); |
| 318 |
</programlisting> |
| 319 |
Create possible access paths for a join of two (or more) foreign tables |
| 320 |
that all belong to the same foreign server. This optional |
| 321 |
function is called during query planning. As |
| 322 |
with <function>GetForeignPaths</function>, this function should |
| 323 |
generate <structname>ForeignPath</structname> path(s) for the |
| 324 |
supplied <literal>joinrel</literal> |
| 325 |
(use <function>create_foreign_join_path</function> to build them), |
| 326 |
and call <function>add_path</function> to add these |
| 327 |
paths to the set of paths considered for the join. But unlike |
| 328 |
<function>GetForeignPaths</function>, it is not necessary that this function |
| 329 |
succeed in creating at least one path, since paths involving local |
| 330 |
joining are always possible. |
| 331 |
</para> |
| 332 |
|
| 333 |
<para> |
| 334 |
Note that this function will be invoked repeatedly for the same join |
| 335 |
relation, with different combinations of inner and outer relations; it is |
| 336 |
the responsibility of the FDW to minimize duplicated work. |
| 337 |
</para> |
| 338 |
|
| 339 |
<para> |
| 340 |
Note also that the set of join clauses to apply to the join, |
| 341 |
which is passed as <literal>extra->restrictlist</literal>, varies |
| 342 |
depending on the combination of inner and outer relations. A |
| 343 |
<structname>ForeignPath</structname> path generated for the |
| 344 |
<literal>joinrel</literal> must contain the set of join clauses it uses, |
| 345 |
which will be used by the planner to convert the |
| 346 |
<structname>ForeignPath</structname> path into a plan, if it is selected |
| 347 |
by the planner as the best path for the <literal>joinrel</literal>. |
| 348 |
</para> |
| 349 |
|
| 350 |
<para> |
| 351 |
If a <structname>ForeignPath</structname> path is chosen for the join, it will |
| 352 |
represent the entire join process; paths generated for the component |
| 353 |
tables and subsidiary joins will not be used. Subsequent processing of |
| 354 |
the join path proceeds much as it does for a path scanning a single |
| 355 |
foreign table. One difference is that the <structfield>scanrelid</structfield> of |
| 356 |
the resulting <structname>ForeignScan</structname> plan node should be set to zero, |
| 357 |
since there is no single relation that it represents; instead, |
| 358 |
the <structfield>fs_relids</structfield> field of the <structname>ForeignScan</structname> |
| 359 |
node represents the set of relations that were joined. (The latter field |
| 360 |
is set up automatically by the core planner code, and need not be filled |
| 361 |
by the FDW.) Another difference is that, because the column list for a |
| 362 |
remote join cannot be found from the system catalogs, the FDW must |
| 363 |
fill <structfield>fdw_scan_tlist</structfield> with an appropriate list |
| 364 |
of <structfield>TargetEntry</structfield> nodes, representing the set of columns |
| 365 |
it will supply at run time in the tuples it returns. |
| 366 |
</para> |
| 367 |
|
| 368 |
<note> |
| 369 |
<para> |
| 370 |
Beginning with <productname>PostgreSQL</productname> 16, |
| 371 |
<structfield>fs_relids</structfield> includes the rangetable indexes |
| 372 |
of outer joins, if any were involved in this join. The new field |
| 373 |
<structfield>fs_base_relids</structfield> includes only base |
| 374 |
relation indexes, and thus |
| 375 |
mimics <structfield>fs_relids</structfield>'s old semantics. |
| 376 |
</para> |
| 377 |
</note> |
| 378 |
|
| 379 |
<para> |
| 380 |
See <xref linkend="fdw-planning"/> for additional information. |
| 381 |
</para> |
| 382 |
</sect2> |
| 383 |
|
| 384 |
<sect2 id="fdw-callbacks-upper-planning"> |
| 385 |
<title>FDW Routines for Planning Post-Scan/Join Processing</title> |
| 386 |
|
| 387 |
<para> |
| 388 |
If an FDW supports performing remote post-scan/join processing, such as |
| 389 |
remote aggregation, it should provide this callback function: |
| 390 |
</para> |
| 391 |
|
| 392 |
<para> |
| 393 |
<programlisting> |
| 394 |
void |
| 395 |
GetForeignUpperPaths(PlannerInfo *root, |
| 396 |
UpperRelationKind stage, |
| 397 |
RelOptInfo *input_rel, |
| 398 |
RelOptInfo *output_rel, |
| 399 |
void *extra); |
| 400 |
</programlisting> |
| 401 |
Create possible access paths for <firstterm>upper relation</firstterm> processing, |
| 402 |
which is the planner's term for all post-scan/join query processing, such |
| 403 |
as aggregation, window functions, sorting, and table updates. This |
| 404 |
optional function is called during query planning. Currently, it is |
| 405 |
called only if all base relation(s) involved in the query belong to the |
| 406 |
same FDW. This function should generate <structname>ForeignPath</structname> |
| 407 |
path(s) for any post-scan/join processing that the FDW knows how to |
| 408 |
perform remotely |
| 409 |
(use <function>create_foreign_upper_path</function> to build them), |
| 410 |
and call <function>add_path</function> to add these paths to |
| 411 |
the indicated upper relation. As with <function>GetForeignJoinPaths</function>, |
| 412 |
it is not necessary that this function succeed in creating any paths, |
| 413 |
since paths involving local processing are always possible. |
| 414 |
</para> |
| 415 |
|
| 416 |
<para> |
| 417 |
The <literal>stage</literal> parameter identifies which post-scan/join step is |
| 418 |
currently being considered. <literal>output_rel</literal> is the upper relation |
| 419 |
that should receive paths representing computation of this step, |
| 420 |
and <literal>input_rel</literal> is the relation representing the input to this |
| 421 |
step. The <literal>extra</literal> parameter provides additional details, |
| 422 |
currently, it is set only for <literal>UPPERREL_PARTIAL_GROUP_AGG</literal> |
| 423 |
or <literal>UPPERREL_GROUP_AGG</literal>, in which case it points to a |
| 424 |
<literal>GroupPathExtraData</literal> structure; |
| 425 |
or for <literal>UPPERREL_FINAL</literal>, in which case it points to a |
| 426 |
<literal>FinalPathExtraData</literal> structure. |
| 427 |
(Note that <structname>ForeignPath</structname> paths added |
| 428 |
to <literal>output_rel</literal> would typically not have any direct dependency |
| 429 |
on paths of the <literal>input_rel</literal>, since their processing is expected |
| 430 |
to be done externally. However, examining paths previously generated for |
| 431 |
the previous processing step can be useful to avoid redundant planning |
| 432 |
work.) |
| 433 |
</para> |
| 434 |
|
| 435 |
<para> |
| 436 |
See <xref linkend="fdw-planning"/> for additional information. |
| 437 |
</para> |
| 438 |
</sect2> |
| 439 |
|
| 440 |
<sect2 id="fdw-callbacks-update"> |
| 441 |
<title>FDW Routines for Updating Foreign Tables</title> |
| 442 |
|
| 443 |
<para> |
| 444 |
If an FDW supports writable foreign tables, it should provide |
| 445 |
some or all of the following callback functions depending on |
| 446 |
the needs and capabilities of the FDW: |
| 447 |
</para> |
| 448 |
|
| 449 |
<para> |
| 450 |
<programlisting> |
| 451 |
void |
| 452 |
AddForeignUpdateTargets(PlannerInfo *root, |
| 453 |
Index rtindex, |
| 454 |
RangeTblEntry *target_rte, |
| 455 |
Relation target_relation); |
| 456 |
</programlisting> |
| 457 |
|
| 458 |
<command>UPDATE</command> and <command>DELETE</command> operations are performed |
| 459 |
against rows previously fetched by the table-scanning functions. The |
| 460 |
FDW may need extra information, such as a row ID or the values of |
| 461 |
primary-key columns, to ensure that it can identify the exact row to |
| 462 |
update or delete. To support that, this function can add extra hidden, |
| 463 |
or <quote>junk</quote>, target columns to the list of columns that are to be |
| 464 |
retrieved from the foreign table during an <command>UPDATE</command> or |
| 465 |
<command>DELETE</command>. |
| 466 |
</para> |
| 467 |
|
| 468 |
<para> |
| 469 |
To do that, construct a <structname>Var</structname> representing |
| 470 |
an extra value you need, and pass it |
| 471 |
to <function>add_row_identity_var</function>, along with a name for |
| 472 |
the junk column. (You can do this more than once if several columns |
| 473 |
are needed.) You must choose a distinct junk column name for each |
| 474 |
different <structname>Var</structname> you need, except |
| 475 |
that <structname>Var</structname>s that are identical except for |
| 476 |
the <structfield>varno</structfield> field can and should share a |
| 477 |
column name. |
| 478 |
The core system uses the junk column names |
| 479 |
<literal>tableoid</literal> for a |
| 480 |
table's <structfield>tableoid</structfield> column, |
| 481 |
<literal>ctid</literal> |
| 482 |
or <literal>ctid<replaceable>N</replaceable></literal> |
| 483 |
for <structfield>ctid</structfield>, |
| 484 |
<literal>wholerow</literal> |
| 485 |
for a whole-row <structname>Var</structname> marked with |
| 486 |
<structfield>vartype</structfield> = <type>RECORD</type>, |
| 487 |
and <literal>wholerow<replaceable>N</replaceable></literal> |
| 488 |
for a whole-row <structname>Var</structname> with |
| 489 |
<structfield>vartype</structfield> equal to the table's declared row type. |
| 490 |
Re-use these names when you can (the planner will combine duplicate |
| 491 |
requests for identical junk columns). If you need another kind of |
| 492 |
junk column besides these, it might be wise to choose a name prefixed |
| 493 |
with your extension name, to avoid conflicts against other FDWs. |
| 494 |
</para> |
| 495 |
|
| 496 |
<para> |
| 497 |
If the <function>AddForeignUpdateTargets</function> pointer is set to |
| 498 |
<literal>NULL</literal>, no extra target expressions are added. |
| 499 |
(This will make it impossible to implement <command>DELETE</command> |
| 500 |
operations, though <command>UPDATE</command> may still be feasible if the FDW |
| 501 |
relies on an unchanging primary key to identify rows.) |
| 502 |
</para> |
| 503 |
|
| 504 |
<para> |
| 505 |
<programlisting> |
| 506 |
List * |
| 507 |
PlanForeignModify(PlannerInfo *root, |
| 508 |
ModifyTable *plan, |
| 509 |
Index resultRelation, |
| 510 |
int subplan_index); |
| 511 |
</programlisting> |
| 512 |
|
| 513 |
Perform any additional planning actions needed for an insert, update, or |
| 514 |
delete on a foreign table. This function generates the FDW-private |
| 515 |
information that will be attached to the <structname>ModifyTable</structname> plan |
| 516 |
node that performs the update action. This private information must |
| 517 |
have the form of a <literal>List</literal>, and will be delivered to |
| 518 |
<function>BeginForeignModify</function> during the execution stage. |
| 519 |
</para> |
| 520 |
|
| 521 |
<para> |
| 522 |
<literal>root</literal> is the planner's global information about the query. |
| 523 |
<literal>plan</literal> is the <structname>ModifyTable</structname> plan node, which is |
| 524 |
complete except for the <structfield>fdwPrivLists</structfield> field. |
| 525 |
<literal>resultRelation</literal> identifies the target foreign table by its |
| 526 |
range table index. <literal>subplan_index</literal> identifies which target of |
| 527 |
the <structname>ModifyTable</structname> plan node this is, counting from zero; |
| 528 |
use this if you want to index into per-target-relation substructures of the |
| 529 |
<literal>plan</literal> node. |
| 530 |
</para> |
| 531 |
|
| 532 |
<para> |
| 533 |
See <xref linkend="fdw-planning"/> for additional information. |
| 534 |
</para> |
| 535 |
|
| 536 |
<para> |
| 537 |
If the <function>PlanForeignModify</function> pointer is set to |
| 538 |
<literal>NULL</literal>, no additional plan-time actions are taken, and the |
| 539 |
<literal>fdw_private</literal> list delivered to |
| 540 |
<function>BeginForeignModify</function> will be NIL. |
| 541 |
</para> |
| 542 |
|
| 543 |
<para> |
| 544 |
<programlisting> |
| 545 |
void |
| 546 |
BeginForeignModify(ModifyTableState *mtstate, |
| 547 |
ResultRelInfo *rinfo, |
| 548 |
List *fdw_private, |
| 549 |
int subplan_index, |
| 550 |
int eflags); |
| 551 |
</programlisting> |
| 552 |
|
| 553 |
Begin executing a foreign table modification operation. This routine is |
| 554 |
called during executor startup. It should perform any initialization |
| 555 |
needed prior to the actual table modifications. Subsequently, |
| 556 |
<function>ExecForeignInsert/ExecForeignBatchInsert</function>, |
| 557 |
<function>ExecForeignUpdate</function> or |
| 558 |
<function>ExecForeignDelete</function> will be called for tuple(s) to be |
| 559 |
inserted, updated, or deleted. |
| 560 |
</para> |
| 561 |
|
| 562 |
<para> |
| 563 |
<literal>mtstate</literal> is the overall state of the |
| 564 |
<structname>ModifyTable</structname> plan node being executed; global data about |
| 565 |
the plan and execution state is available via this structure. |
| 566 |
<literal>rinfo</literal> is the <structname>ResultRelInfo</structname> struct describing |
| 567 |
the target foreign table. (The <structfield>ri_FdwState</structfield> field of |
| 568 |
<structname>ResultRelInfo</structname> is available for the FDW to store any |
| 569 |
private state it needs for this operation.) |
| 570 |
<literal>fdw_private</literal> contains the private data generated by |
| 571 |
<function>PlanForeignModify</function>, if any. |
| 572 |
<literal>subplan_index</literal> identifies which target of |
| 573 |
the <structname>ModifyTable</structname> plan node this is. |
| 574 |
<literal>eflags</literal> contains flag bits describing the executor's |
| 575 |
operating mode for this plan node. |
| 576 |
</para> |
| 577 |
|
| 578 |
<para> |
| 579 |
Note that when <literal>(eflags & EXEC_FLAG_EXPLAIN_ONLY)</literal> is |
| 580 |
true, this function should not perform any externally-visible actions; |
| 581 |
it should only do the minimum required to make the node state valid |
| 582 |
for <function>ExplainForeignModify</function> and <function>EndForeignModify</function>. |
| 583 |
</para> |
| 584 |
|
| 585 |
<para> |
| 586 |
If the <function>BeginForeignModify</function> pointer is set to |
| 587 |
<literal>NULL</literal>, no action is taken during executor startup. |
| 588 |
</para> |
| 589 |
|
| 590 |
<para> |
| 591 |
<programlisting> |
| 592 |
TupleTableSlot * |
| 593 |
ExecForeignInsert(EState *estate, |
| 594 |
ResultRelInfo *rinfo, |
| 595 |
TupleTableSlot *slot, |
| 596 |
TupleTableSlot *planSlot); |
| 597 |
</programlisting> |
| 598 |
|
| 599 |
Insert one tuple into the foreign table. |
| 600 |
<literal>estate</literal> is global execution state for the query. |
| 601 |
<literal>rinfo</literal> is the <structname>ResultRelInfo</structname> struct describing |
| 602 |
the target foreign table. |
| 603 |
<literal>slot</literal> contains the tuple to be inserted; it will match the |
| 604 |
row-type definition of the foreign table. |
| 605 |
<literal>planSlot</literal> contains the tuple that was generated by the |
| 606 |
<structname>ModifyTable</structname> plan node's subplan; it differs from |
| 607 |
<literal>slot</literal> in possibly containing additional <quote>junk</quote> |
| 608 |
columns. (The <literal>planSlot</literal> is typically of little interest |
| 609 |
for <command>INSERT</command> cases, but is provided for completeness.) |
| 610 |
</para> |
| 611 |
|
| 612 |
<para> |
| 613 |
The return value is either a slot containing the data that was actually |
| 614 |
inserted (this might differ from the data supplied, for example as a |
| 615 |
result of trigger actions), or NULL if no row was actually inserted |
| 616 |
(again, typically as a result of triggers). The passed-in |
| 617 |
<literal>slot</literal> can be re-used for this purpose. |
| 618 |
</para> |
| 619 |
|
| 620 |
<para> |
| 621 |
The data in the returned slot is used only if the <command>INSERT</command> |
| 622 |
statement has a <literal>RETURNING</literal> clause or involves a view |
| 623 |
<literal>WITH CHECK OPTION</literal>; or if the foreign table has |
| 624 |
an <literal>AFTER ROW</literal> trigger. Triggers require all columns, |
| 625 |
but the FDW could choose to optimize away returning some or all columns |
| 626 |
depending on the contents of the <literal>RETURNING</literal> clause or |
| 627 |
<literal>WITH CHECK OPTION</literal> constraints. Regardless, some slot |
| 628 |
must be returned to indicate success, or the query's reported row count |
| 629 |
will be wrong. |
| 630 |
</para> |
| 631 |
|
| 632 |
<para> |
| 633 |
If the <function>ExecForeignInsert</function> pointer is set to |
| 634 |
<literal>NULL</literal>, attempts to insert into the foreign table will fail |
| 635 |
with an error message. |
| 636 |
</para> |
| 637 |
|
| 638 |
<para> |
| 639 |
Note that this function is also called when inserting routed tuples into |
| 640 |
a foreign-table partition or executing <command>COPY FROM</command> on |
| 641 |
a foreign table, in which case it is called in a different way than it |
| 642 |
is in the <command>INSERT</command> case. See the callback functions |
| 643 |
described below that allow the FDW to support that. |
| 644 |
</para> |
| 645 |
|
| 646 |
<para> |
| 647 |
<programlisting> |
| 648 |
TupleTableSlot ** |
| 649 |
ExecForeignBatchInsert(EState *estate, |
| 650 |
ResultRelInfo *rinfo, |
| 651 |
TupleTableSlot **slots, |
| 652 |
TupleTableSlot **planSlots, |
| 653 |
int *numSlots); |
| 654 |
</programlisting> |
| 655 |
|
| 656 |
Insert multiple tuples in bulk into the foreign table. |
| 657 |
The parameters are the same for <function>ExecForeignInsert</function> |
| 658 |
except <literal>slots</literal> and <literal>planSlots</literal> contain |
| 659 |
multiple tuples and <literal>*numSlots</literal> specifies the number of |
| 660 |
tuples in those arrays. |
| 661 |
</para> |
| 662 |
|
| 663 |
<para> |
| 664 |
The return value is an array of slots containing the data that was |
| 665 |
actually inserted (this might differ from the data supplied, for |
| 666 |
example as a result of trigger actions.) |
| 667 |
The passed-in <literal>slots</literal> can be re-used for this purpose. |
| 668 |
The number of successfully inserted tuples is returned in |
| 669 |
<literal>*numSlots</literal>. |
| 670 |
</para> |
| 671 |
|
| 672 |
<para> |
| 673 |
The data in the returned slot is used only if the <command>INSERT</command> |
| 674 |
statement involves a view |
| 675 |
<literal>WITH CHECK OPTION</literal>; or if the foreign table has |
| 676 |
an <literal>AFTER ROW</literal> trigger. Triggers require all columns, |
| 677 |
but the FDW could choose to optimize away returning some or all columns |
| 678 |
depending on the contents of the |
| 679 |
<literal>WITH CHECK OPTION</literal> constraints. |
| 680 |
</para> |
| 681 |
|
| 682 |
<para> |
| 683 |
If the <function>ExecForeignBatchInsert</function> or |
| 684 |
<function>GetForeignModifyBatchSize</function> pointer is set to |
| 685 |
<literal>NULL</literal>, attempts to insert into the foreign table will |
| 686 |
use <function>ExecForeignInsert</function>. |
| 687 |
This function is not used if the <command>INSERT</command> has the |
| 688 |
<literal>RETURNING</literal> clause. |
| 689 |
</para> |
| 690 |
|
| 691 |
<para> |
| 692 |
Note that this function is also called when inserting routed tuples into |
| 693 |
a foreign-table partition or executing <command>COPY FROM</command> on |
| 694 |
a foreign table, in which case it is called in a different way than it |
| 695 |
is in the <command>INSERT</command> case. See the callback functions |
| 696 |
described below that allow the FDW to support that. |
| 697 |
</para> |
| 698 |
|
| 699 |
<para> |
| 700 |
<programlisting> |
| 701 |
int |
| 702 |
GetForeignModifyBatchSize(ResultRelInfo *rinfo); |
| 703 |
</programlisting> |
| 704 |
|
| 705 |
Report the maximum number of tuples that a single |
| 706 |
<function>ExecForeignBatchInsert</function> call can handle for |
| 707 |
the specified foreign table. The executor passes at most |
| 708 |
the given number of tuples to <function>ExecForeignBatchInsert</function>. |
| 709 |
<literal>rinfo</literal> is the <structname>ResultRelInfo</structname> struct describing |
| 710 |
the target foreign table. |
| 711 |
The FDW is expected to provide a foreign server and/or foreign |
| 712 |
table option for the user to set this value, or some hard-coded value. |
| 713 |
</para> |
| 714 |
|
| 715 |
<para> |
| 716 |
If the <function>ExecForeignBatchInsert</function> or |
| 717 |
<function>GetForeignModifyBatchSize</function> pointer is set to |
| 718 |
<literal>NULL</literal>, attempts to insert into the foreign table will |
| 719 |
use <function>ExecForeignInsert</function>. |
| 720 |
</para> |
| 721 |
|
| 722 |
<para> |
| 723 |
<programlisting> |
| 724 |
TupleTableSlot * |
| 725 |
ExecForeignUpdate(EState *estate, |
| 726 |
ResultRelInfo *rinfo, |
| 727 |
TupleTableSlot *slot, |
| 728 |
TupleTableSlot *planSlot); |
| 729 |
</programlisting> |
| 730 |
|
| 731 |
Update one tuple in the foreign table. |
| 732 |
<literal>estate</literal> is global execution state for the query. |
| 733 |
<literal>rinfo</literal> is the <structname>ResultRelInfo</structname> struct describing |
| 734 |
the target foreign table. |
| 735 |
<literal>slot</literal> contains the new data for the tuple; it will match the |
| 736 |
row-type definition of the foreign table. |
| 737 |
<literal>planSlot</literal> contains the tuple that was generated by the |
| 738 |
<structname>ModifyTable</structname> plan node's subplan. Unlike |
| 739 |
<literal>slot</literal>, this tuple contains only the new values for |
| 740 |
columns changed by the query, so do not rely on attribute numbers of the |
| 741 |
foreign table to index into <literal>planSlot</literal>. |
| 742 |
Also, <literal>planSlot</literal> typically contains |
| 743 |
additional <quote>junk</quote> columns. In particular, any junk columns |
| 744 |
that were requested by <function>AddForeignUpdateTargets</function> will |
| 745 |
be available from this slot. |
| 746 |
</para> |
| 747 |
|
| 748 |
<para> |
| 749 |
The return value is either a slot containing the row as it was actually |
| 750 |
updated (this might differ from the data supplied, for example as a |
| 751 |
result of trigger actions), or NULL if no row was actually updated |
| 752 |
(again, typically as a result of triggers). The passed-in |
| 753 |
<literal>slot</literal> can be re-used for this purpose. |
| 754 |
</para> |
| 755 |
|
| 756 |
<para> |
| 757 |
The data in the returned slot is used only if the <command>UPDATE</command> |
| 758 |
statement has a <literal>RETURNING</literal> clause or involves a view |
| 759 |
<literal>WITH CHECK OPTION</literal>; or if the foreign table has |
| 760 |
an <literal>AFTER ROW</literal> trigger. Triggers require all columns, |
| 761 |
but the FDW could choose to optimize away returning some or all columns |
| 762 |
depending on the contents of the <literal>RETURNING</literal> clause or |
| 763 |
<literal>WITH CHECK OPTION</literal> constraints. Regardless, some slot |
| 764 |
must be returned to indicate success, or the query's reported row count |
| 765 |
will be wrong. |
| 766 |
</para> |
| 767 |
|
| 768 |
<para> |
| 769 |
If the <function>ExecForeignUpdate</function> pointer is set to |
| 770 |
<literal>NULL</literal>, attempts to update the foreign table will fail |
| 771 |
with an error message. |
| 772 |
</para> |
| 773 |
|
| 774 |
<para> |
| 775 |
<programlisting> |
| 776 |
TupleTableSlot * |
| 777 |
ExecForeignDelete(EState *estate, |
| 778 |
ResultRelInfo *rinfo, |
| 779 |
TupleTableSlot *slot, |
| 780 |
TupleTableSlot *planSlot); |
| 781 |
</programlisting> |
| 782 |
|
| 783 |
Delete one tuple from the foreign table. |
| 784 |
<literal>estate</literal> is global execution state for the query. |
| 785 |
<literal>rinfo</literal> is the <structname>ResultRelInfo</structname> struct describing |
| 786 |
the target foreign table. |
| 787 |
<literal>slot</literal> contains nothing useful upon call, but can be used to |
| 788 |
hold the returned tuple. |
| 789 |
<literal>planSlot</literal> contains the tuple that was generated by the |
| 790 |
<structname>ModifyTable</structname> plan node's subplan; in particular, it will |
| 791 |
carry any junk columns that were requested by |
| 792 |
<function>AddForeignUpdateTargets</function>. The junk column(s) must be used |
| 793 |
to identify the tuple to be deleted. |
| 794 |
</para> |
| 795 |
|
| 796 |
<para> |
| 797 |
The return value is either a slot containing the row that was deleted, |
| 798 |
or NULL if no row was deleted (typically as a result of triggers). The |
| 799 |
passed-in <literal>slot</literal> can be used to hold the tuple to be returned. |
| 800 |
</para> |
| 801 |
|
| 802 |
<para> |
| 803 |
The data in the returned slot is used only if the <command>DELETE</command> |
| 804 |
query has a <literal>RETURNING</literal> clause or the foreign table has |
| 805 |
an <literal>AFTER ROW</literal> trigger. Triggers require all columns, but the |
| 806 |
FDW could choose to optimize away returning some or all columns depending |
| 807 |
on the contents of the <literal>RETURNING</literal> clause. Regardless, some |
| 808 |
slot must be returned to indicate success, or the query's reported row |
| 809 |
count will be wrong. |
| 810 |
</para> |
| 811 |
|
| 812 |
<para> |
| 813 |
If the <function>ExecForeignDelete</function> pointer is set to |
| 814 |
<literal>NULL</literal>, attempts to delete from the foreign table will fail |
| 815 |
with an error message. |
| 816 |
</para> |
| 817 |
|
| 818 |
<para> |
| 819 |
<programlisting> |
| 820 |
void |
| 821 |
EndForeignModify(EState *estate, |
| 822 |
ResultRelInfo *rinfo); |
| 823 |
</programlisting> |
| 824 |
|
| 825 |
End the table update and release resources. It is normally not important |
| 826 |
to release palloc'd memory, but for example open files and connections |
| 827 |
to remote servers should be cleaned up. |
| 828 |
</para> |
| 829 |
|
| 830 |
<para> |
| 831 |
If the <function>EndForeignModify</function> pointer is set to |
| 832 |
<literal>NULL</literal>, no action is taken during executor shutdown. |
| 833 |
</para> |
| 834 |
|
| 835 |
<para> |
| 836 |
Tuples inserted into a partitioned table by <command>INSERT</command> or |
| 837 |
<command>COPY FROM</command> are routed to partitions. If an FDW |
| 838 |
supports routable foreign-table partitions, it should also provide the |
| 839 |
following callback functions. These functions are also called when |
| 840 |
<command>COPY FROM</command> is executed on a foreign table. |
| 841 |
</para> |
| 842 |
|
| 843 |
<para> |
| 844 |
<programlisting> |
| 845 |
void |
| 846 |
BeginForeignInsert(ModifyTableState *mtstate, |
| 847 |
ResultRelInfo *rinfo); |
| 848 |
</programlisting> |
| 849 |
|
| 850 |
Begin executing an insert operation on a foreign table. This routine is |
| 851 |
called right before the first tuple is inserted into the foreign table |
| 852 |
in both cases when it is the partition chosen for tuple routing and the |
| 853 |
target specified in a <command>COPY FROM</command> command. It should |
| 854 |
perform any initialization needed prior to the actual insertion. |
| 855 |
Subsequently, <function>ExecForeignInsert</function> or |
| 856 |
<function>ExecForeignBatchInsert</function> will be called for |
| 857 |
tuple(s) to be inserted into the foreign table. |
| 858 |
</para> |
| 859 |
|
| 860 |
<para> |
| 861 |
<literal>mtstate</literal> is the overall state of the |
| 862 |
<structname>ModifyTable</structname> plan node being executed; global data about |
| 863 |
the plan and execution state is available via this structure. |
| 864 |
<literal>rinfo</literal> is the <structname>ResultRelInfo</structname> struct describing |
| 865 |
the target foreign table. (The <structfield>ri_FdwState</structfield> field of |
| 866 |
<structname>ResultRelInfo</structname> is available for the FDW to store any |
| 867 |
private state it needs for this operation.) |
| 868 |
</para> |
| 869 |
|
| 870 |
<para> |
| 871 |
When this is called by a <command>COPY FROM</command> command, the |
| 872 |
plan-related global data in <literal>mtstate</literal> is not provided |
| 873 |
and the <literal>planSlot</literal> parameter of |
| 874 |
<function>ExecForeignInsert</function> subsequently called for each |
| 875 |
inserted tuple is <literal>NULL</literal>, whether the foreign table is |
| 876 |
the partition chosen for tuple routing or the target specified in the |
| 877 |
command. |
| 878 |
</para> |
| 879 |
|
| 880 |
<para> |
| 881 |
If the <function>BeginForeignInsert</function> pointer is set to |
| 882 |
<literal>NULL</literal>, no action is taken for the initialization. |
| 883 |
</para> |
| 884 |
|
| 885 |
<para> |
| 886 |
Note that if the FDW does not support routable foreign-table partitions |
| 887 |
and/or executing <command>COPY FROM</command> on foreign tables, this |
| 888 |
function or <function>ExecForeignInsert/ExecForeignBatchInsert</function> |
| 889 |
subsequently called must throw error as needed. |
| 890 |
</para> |
| 891 |
|
| 892 |
<para> |
| 893 |
<programlisting> |
| 894 |
void |
| 895 |
EndForeignInsert(EState *estate, |
| 896 |
ResultRelInfo *rinfo); |
| 897 |
</programlisting> |
| 898 |
|
| 899 |
End the insert operation and release resources. It is normally not important |
| 900 |
to release palloc'd memory, but for example open files and connections |
| 901 |
to remote servers should be cleaned up. |
| 902 |
</para> |
| 903 |
|
| 904 |
<para> |
| 905 |
If the <function>EndForeignInsert</function> pointer is set to |
| 906 |
<literal>NULL</literal>, no action is taken for the termination. |
| 907 |
</para> |
| 908 |
|
| 909 |
<para> |
| 910 |
<programlisting> |
| 911 |
int |
| 912 |
IsForeignRelUpdatable(Relation rel); |
| 913 |
</programlisting> |
| 914 |
|
| 915 |
Report which update operations the specified foreign table supports. |
| 916 |
The return value should be a bit mask of rule event numbers indicating |
| 917 |
which operations are supported by the foreign table, using the |
| 918 |
<literal>CmdType</literal> enumeration; that is, |
| 919 |
<literal>(1 << CMD_UPDATE) = 4</literal> for <command>UPDATE</command>, |
| 920 |
<literal>(1 << CMD_INSERT) = 8</literal> for <command>INSERT</command>, and |
| 921 |
<literal>(1 << CMD_DELETE) = 16</literal> for <command>DELETE</command>. |
| 922 |
</para> |
| 923 |
|
| 924 |
<para> |
| 925 |
If the <function>IsForeignRelUpdatable</function> pointer is set to |
| 926 |
<literal>NULL</literal>, foreign tables are assumed to be insertable, updatable, |
| 927 |
or deletable if the FDW provides <function>ExecForeignInsert</function>, |
| 928 |
<function>ExecForeignUpdate</function>, or <function>ExecForeignDelete</function> |
| 929 |
respectively. This function is only needed if the FDW supports some |
| 930 |
tables that are updatable and some that are not. (Even then, it's |
| 931 |
permissible to throw an error in the execution routine instead of |
| 932 |
checking in this function. However, this function is used to determine |
| 933 |
updatability for display in the <literal>information_schema</literal> views.) |
| 934 |
</para> |
| 935 |
|
| 936 |
<para> |
| 937 |
Some inserts, updates, and deletes to foreign tables can be optimized |
| 938 |
by implementing an alternative set of interfaces. The ordinary |
| 939 |
interfaces for inserts, updates, and deletes fetch rows from the remote |
| 940 |
server and then modify those rows one at a time. In some cases, this |
| 941 |
row-by-row approach is necessary, but it can be inefficient. If it is |
| 942 |
possible for the foreign server to determine which rows should be |
| 943 |
modified without actually retrieving them, and if there are no local |
| 944 |
structures which would affect the operation (row-level local triggers, |
| 945 |
stored generated columns, or <literal>WITH CHECK OPTION</literal> |
| 946 |
constraints from parent views), then it is possible to arrange things |
| 947 |
so that the entire operation is performed on the remote server. The |
| 948 |
interfaces described below make this possible. |
| 949 |
</para> |
| 950 |
|
| 951 |
<para> |
| 952 |
<programlisting> |
| 953 |
bool |
| 954 |
PlanDirectModify(PlannerInfo *root, |
| 955 |
ModifyTable *plan, |
| 956 |
Index resultRelation, |
| 957 |
int subplan_index); |
| 958 |
</programlisting> |
| 959 |
|
| 960 |
Decide whether it is safe to execute a direct modification |
| 961 |
on the remote server. If so, return <literal>true</literal> after performing |
| 962 |
planning actions needed for that. Otherwise, return <literal>false</literal>. |
| 963 |
This optional function is called during query planning. |
| 964 |
If this function succeeds, <function>BeginDirectModify</function>, |
| 965 |
<function>IterateDirectModify</function> and <function>EndDirectModify</function> will |
| 966 |
be called at the execution stage, instead. Otherwise, the table |
| 967 |
modification will be executed using the table-updating functions |
| 968 |
described above. |
| 969 |
The parameters are the same as for <function>PlanForeignModify</function>. |
| 970 |
</para> |
| 971 |
|
| 972 |
<para> |
| 973 |
To execute the direct modification on the remote server, this function |
| 974 |
must rewrite the target subplan with a <structname>ForeignScan</structname> plan |
| 975 |
node that executes the direct modification on the remote server. The |
| 976 |
<structfield>operation</structfield> and <structfield>resultRelation</structfield> fields |
| 977 |
of the <structname>ForeignScan</structname> must be set appropriately. |
| 978 |
<structfield>operation</structfield> must be set to the <literal>CmdType</literal> |
| 979 |
enumeration corresponding to the statement kind (that is, |
| 980 |
<literal>CMD_UPDATE</literal> for <command>UPDATE</command>, |
| 981 |
<literal>CMD_INSERT</literal> for <command>INSERT</command>, and |
| 982 |
<literal>CMD_DELETE</literal> for <command>DELETE</command>), and the |
| 983 |
<literal>resultRelation</literal> argument must be copied to the |
| 984 |
<structfield>resultRelation</structfield> field. |
| 985 |
</para> |
| 986 |
|
| 987 |
<para> |
| 988 |
See <xref linkend="fdw-planning"/> for additional information. |
| 989 |
</para> |
| 990 |
|
| 991 |
<para> |
| 992 |
If the <function>PlanDirectModify</function> pointer is set to |
| 993 |
<literal>NULL</literal>, no attempts to execute a direct modification on the |
| 994 |
remote server are taken. |
| 995 |
</para> |
| 996 |
|
| 997 |
<para> |
| 998 |
<programlisting> |
| 999 |
void |
| 1000 |
BeginDirectModify(ForeignScanState *node, |
| 1001 |
int eflags); |
| 1002 |
</programlisting> |
| 1003 |
|
| 1004 |
Prepare to execute a direct modification on the remote server. |
| 1005 |
This is called during executor startup. It should perform any |
| 1006 |
initialization needed prior to the direct modification (that should be |
| 1007 |
done upon the first call to <function>IterateDirectModify</function>). |
| 1008 |
The <structname>ForeignScanState</structname> node has already been created, but |
| 1009 |
its <structfield>fdw_state</structfield> field is still NULL. Information about |
| 1010 |
the table to modify is accessible through the |
| 1011 |
<structname>ForeignScanState</structname> node (in particular, from the underlying |
| 1012 |
<structname>ForeignScan</structname> plan node, which contains any FDW-private |
| 1013 |
information provided by <function>PlanDirectModify</function>). |
| 1014 |
<literal>eflags</literal> contains flag bits describing the executor's |
| 1015 |
operating mode for this plan node. |
| 1016 |
</para> |
| 1017 |
|
| 1018 |
<para> |
| 1019 |
Note that when <literal>(eflags & EXEC_FLAG_EXPLAIN_ONLY)</literal> is |
| 1020 |
true, this function should not perform any externally-visible actions; |
| 1021 |
it should only do the minimum required to make the node state valid |
| 1022 |
for <function>ExplainDirectModify</function> and <function>EndDirectModify</function>. |
| 1023 |
</para> |
| 1024 |
|
| 1025 |
<para> |
| 1026 |
If the <function>BeginDirectModify</function> pointer is set to |
| 1027 |
<literal>NULL</literal>, no attempts to execute a direct modification on the |
| 1028 |
remote server are taken. |
| 1029 |
</para> |
| 1030 |
|
| 1031 |
<para> |
| 1032 |
<programlisting> |
| 1033 |
TupleTableSlot * |
| 1034 |
IterateDirectModify(ForeignScanState *node); |
| 1035 |
</programlisting> |
| 1036 |
|
| 1037 |
When the <command>INSERT</command>, <command>UPDATE</command> or <command>DELETE</command> |
| 1038 |
query doesn't have a <literal>RETURNING</literal> clause, just return NULL |
| 1039 |
after a direct modification on the remote server. |
| 1040 |
When the query has the clause, fetch one result containing the data |
| 1041 |
needed for the <literal>RETURNING</literal> calculation, returning it in a |
| 1042 |
tuple table slot (the node's <structfield>ScanTupleSlot</structfield> should be |
| 1043 |
used for this purpose). The data that was actually inserted, updated |
| 1044 |
or deleted must be stored in |
| 1045 |
<literal>node->resultRelInfo->ri_projectReturning->pi_exprContext->ecxt_scantuple</literal>. |
| 1046 |
Return NULL if no more rows are available. |
| 1047 |
Note that this is called in a short-lived memory context that will be |
| 1048 |
reset between invocations. Create a memory context in |
| 1049 |
<function>BeginDirectModify</function> if you need longer-lived storage, or use |
| 1050 |
the <structfield>es_query_cxt</structfield> of the node's <structname>EState</structname>. |
| 1051 |
</para> |
| 1052 |
|
| 1053 |
<para> |
| 1054 |
The rows returned must match the <structfield>fdw_scan_tlist</structfield> target |
| 1055 |
list if one was supplied, otherwise they must match the row type of the |
| 1056 |
foreign table being updated. If you choose to optimize away fetching |
| 1057 |
columns that are not needed for the <literal>RETURNING</literal> calculation, |
| 1058 |
you should insert nulls in those column positions, or else generate a |
| 1059 |
<structfield>fdw_scan_tlist</structfield> list with those columns omitted. |
| 1060 |
</para> |
| 1061 |
|
| 1062 |
<para> |
| 1063 |
Whether the query has the clause or not, the query's reported row count |
| 1064 |
must be incremented by the FDW itself. When the query doesn't have the |
| 1065 |
clause, the FDW must also increment the row count for the |
| 1066 |
<structname>ForeignScanState</structname> node in the <command>EXPLAIN ANALYZE</command> |
| 1067 |
case. |
| 1068 |
</para> |
| 1069 |
|
| 1070 |
<para> |
| 1071 |
If the <function>IterateDirectModify</function> pointer is set to |
| 1072 |
<literal>NULL</literal>, no attempts to execute a direct modification on the |
| 1073 |
remote server are taken. |
| 1074 |
</para> |
| 1075 |
|
| 1076 |
<para> |
| 1077 |
<programlisting> |
| 1078 |
void |
| 1079 |
EndDirectModify(ForeignScanState *node); |
| 1080 |
</programlisting> |
| 1081 |
|
| 1082 |
Clean up following a direct modification on the remote server. It is |
| 1083 |
normally not important to release palloc'd memory, but for example open |
| 1084 |
files and connections to the remote server should be cleaned up. |
| 1085 |
</para> |
| 1086 |
|
| 1087 |
<para> |
| 1088 |
If the <function>EndDirectModify</function> pointer is set to |
| 1089 |
<literal>NULL</literal>, no attempts to execute a direct modification on the |
| 1090 |
remote server are taken. |
| 1091 |
</para> |
| 1092 |
|
| 1093 |
</sect2> |
| 1094 |
|
| 1095 |
<sect2 id="fdw-callbacks-truncate"> |
| 1096 |
<title>FDW Routines for <command>TRUNCATE</command></title> |
| 1097 |
|
| 1098 |
<para> |
| 1099 |
<programlisting> |
| 1100 |
void |
| 1101 |
ExecForeignTruncate(List *rels, |
| 1102 |
DropBehavior behavior, |
| 1103 |
bool restart_seqs); |
| 1104 |
</programlisting> |
| 1105 |
|
| 1106 |
Truncate foreign tables. This function is called when |
| 1107 |
<xref linkend="sql-truncate"/> is executed on a foreign table. |
| 1108 |
<literal>rels</literal> is a list of <structname>Relation</structname> |
| 1109 |
data structures of foreign tables to truncate. |
| 1110 |
</para> |
| 1111 |
|
| 1112 |
<para> |
| 1113 |
<literal>behavior</literal> is either <literal>DROP_RESTRICT</literal> |
| 1114 |
or <literal>DROP_CASCADE</literal> indicating that the |
| 1115 |
<literal>RESTRICT</literal> or <literal>CASCADE</literal> option was |
| 1116 |
requested in the original <command>TRUNCATE</command> command, |
| 1117 |
respectively. |
| 1118 |
</para> |
| 1119 |
|
| 1120 |
<para> |
| 1121 |
If <literal>restart_seqs</literal> is <literal>true</literal>, |
| 1122 |
the original <command>TRUNCATE</command> command requested the |
| 1123 |
<literal>RESTART IDENTITY</literal> behavior, otherwise the |
| 1124 |
<literal>CONTINUE IDENTITY</literal> behavior was requested. |
| 1125 |
</para> |
| 1126 |
|
| 1127 |
<para> |
| 1128 |
Note that the <literal>ONLY</literal> options specified |
| 1129 |
in the original <command>TRUNCATE</command> command are not passed to |
| 1130 |
<function>ExecForeignTruncate</function>. This behavior is similar to |
| 1131 |
the callback functions of <command>SELECT</command>, |
| 1132 |
<command>UPDATE</command> and <command>DELETE</command> on |
| 1133 |
a foreign table. |
| 1134 |
</para> |
| 1135 |
|
| 1136 |
<para> |
| 1137 |
<function>ExecForeignTruncate</function> is invoked once per |
| 1138 |
foreign server for which foreign tables are to be truncated. |
| 1139 |
This means that all foreign tables included in <literal>rels</literal> |
| 1140 |
must belong to the same server. |
| 1141 |
</para> |
| 1142 |
|
| 1143 |
<para> |
| 1144 |
If the <function>ExecForeignTruncate</function> pointer is set to |
| 1145 |
<literal>NULL</literal>, attempts to truncate foreign tables will |
| 1146 |
fail with an error message. |
| 1147 |
</para> |
| 1148 |
</sect2> |
| 1149 |
|
| 1150 |
<sect2 id="fdw-callbacks-row-locking"> |
| 1151 |
<title>FDW Routines for Row Locking</title> |
| 1152 |
|
| 1153 |
<para> |
| 1154 |
If an FDW wishes to support <firstterm>late row locking</firstterm> (as described |
| 1155 |
in <xref linkend="fdw-row-locking"/>), it must provide the following |
| 1156 |
callback functions: |
| 1157 |
</para> |
| 1158 |
|
| 1159 |
<para> |
| 1160 |
<programlisting> |
| 1161 |
RowMarkType |
| 1162 |
GetForeignRowMarkType(RangeTblEntry *rte, |
| 1163 |
LockClauseStrength strength); |
| 1164 |
</programlisting> |
| 1165 |
|
| 1166 |
Report which row-marking option to use for a foreign table. |
| 1167 |
<literal>rte</literal> is the <structname>RangeTblEntry</structname> node for the table |
| 1168 |
and <literal>strength</literal> describes the lock strength requested by the |
| 1169 |
relevant <literal>FOR UPDATE/SHARE</literal> clause, if any. The result must be |
| 1170 |
a member of the <literal>RowMarkType</literal> enum type. |
| 1171 |
</para> |
| 1172 |
|
| 1173 |
<para> |
| 1174 |
This function is called during query planning for each foreign table that |
| 1175 |
appears in an <command>UPDATE</command>, <command>DELETE</command>, or <command>SELECT |
| 1176 |
FOR UPDATE/SHARE</command> query and is not the target of <command>UPDATE</command> |
| 1177 |
or <command>DELETE</command>. |
| 1178 |
</para> |
| 1179 |
|
| 1180 |
<para> |
| 1181 |
If the <function>GetForeignRowMarkType</function> pointer is set to |
| 1182 |
<literal>NULL</literal>, the <literal>ROW_MARK_COPY</literal> option is always used. |
| 1183 |
(This implies that <function>RefetchForeignRow</function> will never be called, |
| 1184 |
so it need not be provided either.) |
| 1185 |
</para> |
| 1186 |
|
| 1187 |
<para> |
| 1188 |
See <xref linkend="fdw-row-locking"/> for more information. |
| 1189 |
</para> |
| 1190 |
|
| 1191 |
<para> |
| 1192 |
<programlisting> |
| 1193 |
void |
| 1194 |
RefetchForeignRow(EState *estate, |
| 1195 |
ExecRowMark *erm, |
| 1196 |
Datum rowid, |
| 1197 |
TupleTableSlot *slot, |
| 1198 |
bool *updated); |
| 1199 |
</programlisting> |
| 1200 |
|
| 1201 |
Re-fetch one tuple slot from the foreign table, after locking it if required. |
| 1202 |
<literal>estate</literal> is global execution state for the query. |
| 1203 |
<literal>erm</literal> is the <structname>ExecRowMark</structname> struct describing |
| 1204 |
the target foreign table and the row lock type (if any) to acquire. |
| 1205 |
<literal>rowid</literal> identifies the tuple to be fetched. |
| 1206 |
<literal>slot</literal> contains nothing useful upon call, but can be used to |
| 1207 |
hold the returned tuple. <literal>updated</literal> is an output parameter. |
| 1208 |
</para> |
| 1209 |
|
| 1210 |
<para> |
| 1211 |
This function should store the tuple into the provided slot, or clear it if |
| 1212 |
the row lock couldn't be obtained. The row lock type to acquire is |
| 1213 |
defined by <literal>erm->markType</literal>, which is the value |
| 1214 |
previously returned by <function>GetForeignRowMarkType</function>. |
| 1215 |
(<literal>ROW_MARK_REFERENCE</literal> means to just re-fetch the tuple |
| 1216 |
without acquiring any lock, and <literal>ROW_MARK_COPY</literal> will |
| 1217 |
never be seen by this routine.) |
| 1218 |
</para> |
| 1219 |
|
| 1220 |
<para> |
| 1221 |
In addition, <literal>*updated</literal> should be set to <literal>true</literal> |
| 1222 |
if what was fetched was an updated version of the tuple rather than |
| 1223 |
the same version previously obtained. (If the FDW cannot be sure about |
| 1224 |
this, always returning <literal>true</literal> is recommended.) |
| 1225 |
</para> |
| 1226 |
|
| 1227 |
<para> |
| 1228 |
Note that by default, failure to acquire a row lock should result in |
| 1229 |
raising an error; returning with an empty slot is only appropriate if |
| 1230 |
the <literal>SKIP LOCKED</literal> option is specified |
| 1231 |
by <literal>erm->waitPolicy</literal>. |
| 1232 |
</para> |
| 1233 |
|
| 1234 |
<para> |
| 1235 |
The <literal>rowid</literal> is the <structfield>ctid</structfield> value previously read |
| 1236 |
for the row to be re-fetched. Although the <literal>rowid</literal> value is |
| 1237 |
passed as a <type>Datum</type>, it can currently only be a <type>tid</type>. The |
| 1238 |
function API is chosen in hopes that it may be possible to allow other |
| 1239 |
data types for row IDs in future. |
| 1240 |
</para> |
| 1241 |
|
| 1242 |
<para> |
| 1243 |
If the <function>RefetchForeignRow</function> pointer is set to |
| 1244 |
<literal>NULL</literal>, attempts to re-fetch rows will fail |
| 1245 |
with an error message. |
| 1246 |
</para> |
| 1247 |
|
| 1248 |
<para> |
| 1249 |
See <xref linkend="fdw-row-locking"/> for more information. |
| 1250 |
</para> |
| 1251 |
|
| 1252 |
<para> |
| 1253 |
<programlisting> |
| 1254 |
bool |
| 1255 |
RecheckForeignScan(ForeignScanState *node, |
| 1256 |
TupleTableSlot *slot); |
| 1257 |
</programlisting> |
| 1258 |
Recheck that a previously-returned tuple still matches the relevant |
| 1259 |
scan and join qualifiers, and possibly provide a modified version of |
| 1260 |
the tuple. For foreign data wrappers which do not perform join pushdown, |
| 1261 |
it will typically be more convenient to set this to <literal>NULL</literal> and |
| 1262 |
instead set <structfield>fdw_recheck_quals</structfield> appropriately. |
| 1263 |
When outer joins are pushed down, however, it isn't sufficient to |
| 1264 |
reapply the checks relevant to all the base tables to the result tuple, |
| 1265 |
even if all needed attributes are present, because failure to match some |
| 1266 |
qualifier might result in some attributes going to NULL, rather than in |
| 1267 |
no tuple being returned. <literal>RecheckForeignScan</literal> can recheck |
| 1268 |
qualifiers and return true if they are still satisfied and false |
| 1269 |
otherwise, but it can also store a replacement tuple into the supplied |
| 1270 |
slot. |
| 1271 |
</para> |
| 1272 |
|
| 1273 |
<para> |
| 1274 |
To implement join pushdown, a foreign data wrapper will typically |
| 1275 |
construct an alternative local join plan which is used only for |
| 1276 |
rechecks; this will become the outer subplan of the |
| 1277 |
<literal>ForeignScan</literal>. When a recheck is required, this subplan |
| 1278 |
can be executed and the resulting tuple can be stored in the slot. |
| 1279 |
This plan need not be efficient since no base table will return more |
| 1280 |
than one row; for example, it may implement all joins as nested loops. |
| 1281 |
The function <literal>GetExistingLocalJoinPath</literal> may be used to search |
| 1282 |
existing paths for a suitable local join path, which can be used as the |
| 1283 |
alternative local join plan. <literal>GetExistingLocalJoinPath</literal> |
| 1284 |
searches for an unparameterized path in the path list of the specified |
| 1285 |
join relation. (If it does not find such a path, it returns NULL, in |
| 1286 |
which case a foreign data wrapper may build the local path by itself or |
| 1287 |
may choose not to create access paths for that join.) |
| 1288 |
</para> |
| 1289 |
</sect2> |
| 1290 |
|
| 1291 |
<sect2 id="fdw-callbacks-explain"> |
| 1292 |
<title>FDW Routines for <command>EXPLAIN</command></title> |
| 1293 |
|
| 1294 |
<para> |
| 1295 |
<programlisting> |
| 1296 |
void |
| 1297 |
ExplainForeignScan(ForeignScanState *node, |
| 1298 |
ExplainState *es); |
| 1299 |
</programlisting> |
| 1300 |
|
| 1301 |
Print additional <command>EXPLAIN</command> output for a foreign table scan. |
| 1302 |
This function can call <function>ExplainPropertyText</function> and |
| 1303 |
related functions to add fields to the <command>EXPLAIN</command> output. |
| 1304 |
The flag fields in <literal>es</literal> can be used to determine what to |
| 1305 |
print, and the state of the <structname>ForeignScanState</structname> node |
| 1306 |
can be inspected to provide run-time statistics in the <command>EXPLAIN |
| 1307 |
ANALYZE</command> case. |
| 1308 |
</para> |
| 1309 |
|
| 1310 |
<para> |
| 1311 |
If the <function>ExplainForeignScan</function> pointer is set to |
| 1312 |
<literal>NULL</literal>, no additional information is printed during |
| 1313 |
<command>EXPLAIN</command>. |
| 1314 |
</para> |
| 1315 |
|
| 1316 |
<para> |
| 1317 |
<programlisting> |
| 1318 |
void |
| 1319 |
ExplainForeignModify(ModifyTableState *mtstate, |
| 1320 |
ResultRelInfo *rinfo, |
| 1321 |
List *fdw_private, |
| 1322 |
int subplan_index, |
| 1323 |
struct ExplainState *es); |
| 1324 |
</programlisting> |
| 1325 |
|
| 1326 |
Print additional <command>EXPLAIN</command> output for a foreign table update. |
| 1327 |
This function can call <function>ExplainPropertyText</function> and |
| 1328 |
related functions to add fields to the <command>EXPLAIN</command> output. |
| 1329 |
The flag fields in <literal>es</literal> can be used to determine what to |
| 1330 |
print, and the state of the <structname>ModifyTableState</structname> node |
| 1331 |
can be inspected to provide run-time statistics in the <command>EXPLAIN |
| 1332 |
ANALYZE</command> case. The first four arguments are the same as for |
| 1333 |
<function>BeginForeignModify</function>. |
| 1334 |
</para> |
| 1335 |
|
| 1336 |
<para> |
| 1337 |
If the <function>ExplainForeignModify</function> pointer is set to |
| 1338 |
<literal>NULL</literal>, no additional information is printed during |
| 1339 |
<command>EXPLAIN</command>. |
| 1340 |
</para> |
| 1341 |
|
| 1342 |
<para> |
| 1343 |
<programlisting> |
| 1344 |
void |
| 1345 |
ExplainDirectModify(ForeignScanState *node, |
| 1346 |
ExplainState *es); |
| 1347 |
</programlisting> |
| 1348 |
|
| 1349 |
Print additional <command>EXPLAIN</command> output for a direct modification |
| 1350 |
on the remote server. |
| 1351 |
This function can call <function>ExplainPropertyText</function> and |
| 1352 |
related functions to add fields to the <command>EXPLAIN</command> output. |
| 1353 |
The flag fields in <literal>es</literal> can be used to determine what to |
| 1354 |
print, and the state of the <structname>ForeignScanState</structname> node |
| 1355 |
can be inspected to provide run-time statistics in the <command>EXPLAIN |
| 1356 |
ANALYZE</command> case. |
| 1357 |
</para> |
| 1358 |
|
| 1359 |
<para> |
| 1360 |
If the <function>ExplainDirectModify</function> pointer is set to |
| 1361 |
<literal>NULL</literal>, no additional information is printed during |
| 1362 |
<command>EXPLAIN</command>. |
| 1363 |
</para> |
| 1364 |
|
| 1365 |
</sect2> |
| 1366 |
|
| 1367 |
<sect2 id="fdw-callbacks-analyze"> |
| 1368 |
<title>FDW Routines for <command>ANALYZE</command></title> |
| 1369 |
|
| 1370 |
<para> |
| 1371 |
<programlisting> |
| 1372 |
bool |
| 1373 |
AnalyzeForeignTable(Relation relation, |
| 1374 |
AcquireSampleRowsFunc *func, |
| 1375 |
BlockNumber *totalpages); |
| 1376 |
</programlisting> |
| 1377 |
|
| 1378 |
This function is called when <xref linkend="sql-analyze"/> is executed on |
| 1379 |
a foreign table. If the FDW can collect statistics for this |
| 1380 |
foreign table, it should return <literal>true</literal>, and provide a pointer |
| 1381 |
to a function that will collect sample rows from the table in |
| 1382 |
<parameter>func</parameter>, plus the estimated size of the table in pages in |
| 1383 |
<parameter>totalpages</parameter>. Otherwise, return <literal>false</literal>. |
| 1384 |
</para> |
| 1385 |
|
| 1386 |
<para> |
| 1387 |
If the FDW does not support collecting statistics for any tables, the |
| 1388 |
<function>AnalyzeForeignTable</function> pointer can be set to <literal>NULL</literal>. |
| 1389 |
</para> |
| 1390 |
|
| 1391 |
<para> |
| 1392 |
If provided, the sample collection function must have the signature |
| 1393 |
<programlisting> |
| 1394 |
int |
| 1395 |
AcquireSampleRowsFunc(Relation relation, |
| 1396 |
int elevel, |
| 1397 |
HeapTuple *rows, |
| 1398 |
int targrows, |
| 1399 |
double *totalrows, |
| 1400 |
double *totaldeadrows); |
| 1401 |
</programlisting> |
| 1402 |
|
| 1403 |
A random sample of up to <parameter>targrows</parameter> rows should be collected |
| 1404 |
from the table and stored into the caller-provided <parameter>rows</parameter> |
| 1405 |
array. The actual number of rows collected must be returned. In |
| 1406 |
addition, store estimates of the total numbers of live and dead rows in |
| 1407 |
the table into the output parameters <parameter>totalrows</parameter> and |
| 1408 |
<parameter>totaldeadrows</parameter>. (Set <parameter>totaldeadrows</parameter> to zero |
| 1409 |
if the FDW does not have any concept of dead rows.) |
| 1410 |
</para> |
| 1411 |
|
| 1412 |
</sect2> |
| 1413 |
|
| 1414 |
<sect2 id="fdw-callbacks-import"> |
| 1415 |
<title>FDW Routines for <command>IMPORT FOREIGN SCHEMA</command></title> |
| 1416 |
|
| 1417 |
<para> |
| 1418 |
<programlisting> |
| 1419 |
List * |
| 1420 |
ImportForeignSchema(ImportForeignSchemaStmt *stmt, Oid serverOid); |
| 1421 |
</programlisting> |
| 1422 |
|
| 1423 |
Obtain a list of foreign table creation commands. This function is |
| 1424 |
called when executing <xref linkend="sql-importforeignschema"/>, and is |
| 1425 |
passed the parse tree for that statement, as well as the OID of the |
| 1426 |
foreign server to use. It should return a list of C strings, each of |
| 1427 |
which must contain a <xref linkend="sql-createforeigntable"/> command. |
| 1428 |
These strings will be parsed and executed by the core server. |
| 1429 |
</para> |
| 1430 |
|
| 1431 |
<para> |
| 1432 |
Within the <structname>ImportForeignSchemaStmt</structname> struct, |
| 1433 |
<structfield>remote_schema</structfield> is the name of the remote schema from |
| 1434 |
which tables are to be imported. |
| 1435 |
<structfield>list_type</structfield> identifies how to filter table names: |
| 1436 |
<literal>FDW_IMPORT_SCHEMA_ALL</literal> means that all tables in the remote |
| 1437 |
schema should be imported (in this case <structfield>table_list</structfield> is |
| 1438 |
empty), <literal>FDW_IMPORT_SCHEMA_LIMIT_TO</literal> means to include only |
| 1439 |
tables listed in <structfield>table_list</structfield>, |
| 1440 |
and <literal>FDW_IMPORT_SCHEMA_EXCEPT</literal> means to exclude the tables |
| 1441 |
listed in <structfield>table_list</structfield>. |
| 1442 |
<structfield>options</structfield> is a list of options used for the import process. |
| 1443 |
The meanings of the options are up to the FDW. |
| 1444 |
For example, an FDW could use an option to define whether the |
| 1445 |
<literal>NOT NULL</literal> attributes of columns should be imported. |
| 1446 |
These options need not have anything to do with those supported by the |
| 1447 |
FDW as database object options. |
| 1448 |
</para> |
| 1449 |
|
| 1450 |
<para> |
| 1451 |
The FDW may ignore the <structfield>local_schema</structfield> field of |
| 1452 |
the <structname>ImportForeignSchemaStmt</structname>, because the core server |
| 1453 |
will automatically insert that name into the parsed <command>CREATE |
| 1454 |
FOREIGN TABLE</command> commands. |
| 1455 |
</para> |
| 1456 |
|
| 1457 |
<para> |
| 1458 |
The FDW does not have to concern itself with implementing the filtering |
| 1459 |
specified by <structfield>list_type</structfield> and <structfield>table_list</structfield>, |
| 1460 |
either, as the core server will automatically skip any returned commands |
| 1461 |
for tables excluded according to those options. However, it's often |
| 1462 |
useful to avoid the work of creating commands for excluded tables in the |
| 1463 |
first place. The function <function>IsImportableForeignTable()</function> may be |
| 1464 |
useful to test whether a given foreign-table name will pass the filter. |
| 1465 |
</para> |
| 1466 |
|
| 1467 |
<para> |
| 1468 |
If the FDW does not support importing table definitions, the |
| 1469 |
<function>ImportForeignSchema</function> pointer can be set to <literal>NULL</literal>. |
| 1470 |
</para> |
| 1471 |
|
| 1472 |
</sect2> |
| 1473 |
|
| 1474 |
<sect2 id="fdw-callbacks-parallel"> |
| 1475 |
<title>FDW Routines for Parallel Execution</title> |
| 1476 |
<para> |
| 1477 |
A <structname>ForeignScan</structname> node can, optionally, support parallel |
| 1478 |
execution. A parallel <structname>ForeignScan</structname> will be executed |
| 1479 |
in multiple processes and must return each row exactly once across |
| 1480 |
all cooperating processes. To do this, processes can coordinate through |
| 1481 |
fixed-size chunks of dynamic shared memory. This shared memory is not |
| 1482 |
guaranteed to be mapped at the same address in every process, so it |
| 1483 |
must not contain pointers. The following functions are all optional, |
| 1484 |
but most are required if parallel execution is to be supported. |
| 1485 |
</para> |
| 1486 |
|
| 1487 |
<para> |
| 1488 |
<programlisting> |
| 1489 |
bool |
| 1490 |
IsForeignScanParallelSafe(PlannerInfo *root, RelOptInfo *rel, |
| 1491 |
RangeTblEntry *rte); |
| 1492 |
</programlisting> |
| 1493 |
Test whether a scan can be performed within a parallel worker. This |
| 1494 |
function will only be called when the planner believes that a parallel |
| 1495 |
plan might be possible, and should return true if it is safe for that scan |
| 1496 |
to run within a parallel worker. This will generally not be the case if |
| 1497 |
the remote data source has transaction semantics, unless the worker's |
| 1498 |
connection to the data can somehow be made to share the same transaction |
| 1499 |
context as the leader. |
| 1500 |
</para> |
| 1501 |
|
| 1502 |
<para> |
| 1503 |
If this function is not defined, it is assumed that the scan must take |
| 1504 |
place within the parallel leader. Note that returning true does not mean |
| 1505 |
that the scan itself can be done in parallel, only that the scan can be |
| 1506 |
performed within a parallel worker. Therefore, it can be useful to define |
| 1507 |
this method even when parallel execution is not supported. |
| 1508 |
</para> |
| 1509 |
|
| 1510 |
<para> |
| 1511 |
<programlisting> |
| 1512 |
Size |
| 1513 |
EstimateDSMForeignScan(ForeignScanState *node, ParallelContext *pcxt); |
| 1514 |
</programlisting> |
| 1515 |
Estimate the amount of dynamic shared memory that will be required |
| 1516 |
for parallel operation. This may be higher than the amount that will |
| 1517 |
actually be used, but it must not be lower. The return value is in bytes. |
| 1518 |
This function is optional, and can be omitted if not needed; but if it |
| 1519 |
is omitted, the next three functions must be omitted as well, because |
| 1520 |
no shared memory will be allocated for the FDW's use. |
| 1521 |
</para> |
| 1522 |
|
| 1523 |
<para> |
| 1524 |
<programlisting> |
| 1525 |
void |
| 1526 |
InitializeDSMForeignScan(ForeignScanState *node, ParallelContext *pcxt, |
| 1527 |
void *coordinate); |
| 1528 |
</programlisting> |
| 1529 |
Initialize the dynamic shared memory that will be required for parallel |
| 1530 |
operation. <literal>coordinate</literal> points to a shared memory area of |
| 1531 |
size equal to the return value of <function>EstimateDSMForeignScan</function>. |
| 1532 |
This function is optional, and can be omitted if not needed. |
| 1533 |
</para> |
| 1534 |
|
| 1535 |
<para> |
| 1536 |
<programlisting> |
| 1537 |
void |
| 1538 |
ReInitializeDSMForeignScan(ForeignScanState *node, ParallelContext *pcxt, |
| 1539 |
void *coordinate); |
| 1540 |
</programlisting> |
| 1541 |
Re-initialize the dynamic shared memory required for parallel operation |
| 1542 |
when the foreign-scan plan node is about to be re-scanned. |
| 1543 |
This function is optional, and can be omitted if not needed. |
| 1544 |
Recommended practice is that this function reset only shared state, |
| 1545 |
while the <function>ReScanForeignScan</function> function resets only local |
| 1546 |
state. Currently, this function will be called |
| 1547 |
before <function>ReScanForeignScan</function>, but it's best not to rely on |
| 1548 |
that ordering. |
| 1549 |
</para> |
| 1550 |
|
| 1551 |
<para> |
| 1552 |
<programlisting> |
| 1553 |
void |
| 1554 |
InitializeWorkerForeignScan(ForeignScanState *node, shm_toc *toc, |
| 1555 |
void *coordinate); |
| 1556 |
</programlisting> |
| 1557 |
Initialize a parallel worker's local state based on the shared state |
| 1558 |
set up by the leader during <function>InitializeDSMForeignScan</function>. |
| 1559 |
This function is optional, and can be omitted if not needed. |
| 1560 |
</para> |
| 1561 |
|
| 1562 |
<para> |
| 1563 |
<programlisting> |
| 1564 |
void |
| 1565 |
ShutdownForeignScan(ForeignScanState *node); |
| 1566 |
</programlisting> |
| 1567 |
Release resources when it is anticipated the node will not be executed |
| 1568 |
to completion. This is not called in all cases; sometimes, |
| 1569 |
<literal>EndForeignScan</literal> may be called without this function having |
| 1570 |
been called first. Since the DSM segment used by parallel query is |
| 1571 |
destroyed just after this callback is invoked, foreign data wrappers that |
| 1572 |
wish to take some action before the DSM segment goes away should implement |
| 1573 |
this method. |
| 1574 |
</para> |
| 1575 |
</sect2> |
| 1576 |
|
| 1577 |
<sect2 id="fdw-callbacks-async"> |
| 1578 |
<title>FDW Routines for Asynchronous Execution</title> |
| 1579 |
<para> |
| 1580 |
A <structname>ForeignScan</structname> node can, optionally, support |
| 1581 |
asynchronous execution as described in |
| 1582 |
<filename>src/backend/executor/README</filename>. The following |
| 1583 |
functions are all optional, but are all required if asynchronous |
| 1584 |
execution is to be supported. |
| 1585 |
</para> |
| 1586 |
|
| 1587 |
<para> |
| 1588 |
<programlisting> |
| 1589 |
bool |
| 1590 |
IsForeignPathAsyncCapable(ForeignPath *path); |
| 1591 |
</programlisting> |
| 1592 |
Test whether a given <structname>ForeignPath</structname> path can scan |
| 1593 |
the underlying foreign relation asynchronously. |
| 1594 |
This function will only be called at the end of query planning when the |
| 1595 |
given path is a direct child of an <structname>AppendPath</structname> |
| 1596 |
path and when the planner believes that asynchronous execution improves |
| 1597 |
performance, and should return true if the given path is able to scan the |
| 1598 |
foreign relation asynchronously. |
| 1599 |
</para> |
| 1600 |
|
| 1601 |
<para> |
| 1602 |
If this function is not defined, it is assumed that the given path scans |
| 1603 |
the foreign relation using <function>IterateForeignScan</function>. |
| 1604 |
(This implies that the callback functions described below will never be |
| 1605 |
called, so they need not be provided either.) |
| 1606 |
</para> |
| 1607 |
|
| 1608 |
<para> |
| 1609 |
<programlisting> |
| 1610 |
void |
| 1611 |
ForeignAsyncRequest(AsyncRequest *areq); |
| 1612 |
</programlisting> |
| 1613 |
Produce one tuple asynchronously from the |
| 1614 |
<structname>ForeignScan</structname> node. <literal>areq</literal> is |
| 1615 |
the <structname>AsyncRequest</structname> struct describing the |
| 1616 |
<structname>ForeignScan</structname> node and the parent |
| 1617 |
<structname>Append</structname> node that requested the tuple from it. |
| 1618 |
This function should store the tuple into the slot specified by |
| 1619 |
<literal>areq->result</literal>, and set |
| 1620 |
<literal>areq->request_complete</literal> to <literal>true</literal>; |
| 1621 |
or if it needs to wait on an event external to the core server such as |
| 1622 |
network I/O, and cannot produce any tuple immediately, set the flag to |
| 1623 |
<literal>false</literal>, and set |
| 1624 |
<literal>areq->callback_pending</literal> to <literal>true</literal> |
| 1625 |
for the <structname>ForeignScan</structname> node to get a callback from |
| 1626 |
the callback functions described below. If no more tuples are available, |
| 1627 |
set the slot to NULL or an empty slot, and the |
| 1628 |
<literal>areq->request_complete</literal> flag to |
| 1629 |
<literal>true</literal>. It's recommended to use |
| 1630 |
<function>ExecAsyncRequestDone</function> or |
| 1631 |
<function>ExecAsyncRequestPending</function> to set the output parameters |
| 1632 |
in the <literal>areq</literal>. |
| 1633 |
</para> |
| 1634 |
|
| 1635 |
<para> |
| 1636 |
<programlisting> |
| 1637 |
void |
| 1638 |
ForeignAsyncConfigureWait(AsyncRequest *areq); |
| 1639 |
</programlisting> |
| 1640 |
Configure a file descriptor event for which the |
| 1641 |
<structname>ForeignScan</structname> node wishes to wait. |
| 1642 |
This function will only be called when the |
| 1643 |
<structname>ForeignScan</structname> node has the |
| 1644 |
<literal>areq->callback_pending</literal> flag set, and should add |
| 1645 |
the event to the <structfield>as_eventset</structfield> of the parent |
| 1646 |
<structname>Append</structname> node described by the |
| 1647 |
<literal>areq</literal>. See the comments for |
| 1648 |
<function>ExecAsyncConfigureWait</function> in |
| 1649 |
<filename>src/backend/executor/execAsync.c</filename> for additional |
| 1650 |
information. When the file descriptor event occurs, |
| 1651 |
<function>ForeignAsyncNotify</function> will be called. |
| 1652 |
</para> |
| 1653 |
|
| 1654 |
<para> |
| 1655 |
<programlisting> |
| 1656 |
void |
| 1657 |
ForeignAsyncNotify(AsyncRequest *areq); |
| 1658 |
</programlisting> |
| 1659 |
Process a relevant event that has occurred, then produce one tuple |
| 1660 |
asynchronously from the <structname>ForeignScan</structname> node. |
| 1661 |
This function should set the output parameters in the |
| 1662 |
<literal>areq</literal> in the same way as |
| 1663 |
<function>ForeignAsyncRequest</function>. |
| 1664 |
</para> |
| 1665 |
</sect2> |
| 1666 |
|
| 1667 |
<sect2 id="fdw-callbacks-reparameterize-paths"> |
| 1668 |
<title>FDW Routines for Reparameterization of Paths</title> |
| 1669 |
|
| 1670 |
<para> |
| 1671 |
<programlisting> |
| 1672 |
List * |
| 1673 |
ReparameterizeForeignPathByChild(PlannerInfo *root, List *fdw_private, |
| 1674 |
RelOptInfo *child_rel); |
| 1675 |
</programlisting> |
| 1676 |
This function is called while converting a path parameterized by the |
| 1677 |
top-most parent of the given child relation <literal>child_rel</literal> to be |
| 1678 |
parameterized by the child relation. The function is used to reparameterize |
| 1679 |
any paths or translate any expression nodes saved in the given |
| 1680 |
<literal>fdw_private</literal> member of a <structname>ForeignPath</structname>. The |
| 1681 |
callback may use <literal>reparameterize_path_by_child</literal>, |
| 1682 |
<literal>adjust_appendrel_attrs</literal> or |
| 1683 |
<literal>adjust_appendrel_attrs_multilevel</literal> as required. |
| 1684 |
</para> |
| 1685 |
</sect2> |
| 1686 |
|
| 1687 |
</sect1> |
| 1688 |
|
| 1689 |
<sect1 id="fdw-helpers"> |
| 1690 |
<title>Foreign Data Wrapper Helper Functions</title> |
| 1691 |
|
| 1692 |
<para> |
| 1693 |
Several helper functions are exported from the core server so that |
| 1694 |
authors of foreign data wrappers can get easy access to attributes of |
| 1695 |
FDW-related objects, such as FDW options. |
| 1696 |
To use any of these functions, you need to include the header file |
| 1697 |
<filename>foreign/foreign.h</filename> in your source file. |
| 1698 |
That header also defines the struct types that are returned by |
| 1699 |
these functions. |
| 1700 |
</para> |
| 1701 |
|
| 1702 |
<para> |
| 1703 |
<programlisting> |
| 1704 |
ForeignDataWrapper * |
| 1705 |
GetForeignDataWrapperExtended(Oid fdwid, bits16 flags); |
| 1706 |
</programlisting> |
| 1707 |
|
| 1708 |
This function returns a <structname>ForeignDataWrapper</structname> |
| 1709 |
object for the foreign-data wrapper with the given OID. A |
| 1710 |
<structname>ForeignDataWrapper</structname> object contains properties |
| 1711 |
of the FDW (see <filename>foreign/foreign.h</filename> for details). |
| 1712 |
<structfield>flags</structfield> is a bitwise-or'd bit mask indicating |
| 1713 |
an extra set of options. It can take the value |
| 1714 |
<literal>FDW_MISSING_OK</literal>, in which case a <literal>NULL</literal> |
| 1715 |
result is returned to the caller instead of an error for an undefined |
| 1716 |
object. |
| 1717 |
</para> |
| 1718 |
|
| 1719 |
<para> |
| 1720 |
<programlisting> |
| 1721 |
ForeignDataWrapper * |
| 1722 |
GetForeignDataWrapper(Oid fdwid); |
| 1723 |
</programlisting> |
| 1724 |
|
| 1725 |
This function returns a <structname>ForeignDataWrapper</structname> |
| 1726 |
object for the foreign-data wrapper with the given OID. A |
| 1727 |
<structname>ForeignDataWrapper</structname> object contains properties |
| 1728 |
of the FDW (see <filename>foreign/foreign.h</filename> for details). |
| 1729 |
</para> |
| 1730 |
|
| 1731 |
<para> |
| 1732 |
<programlisting> |
| 1733 |
ForeignServer * |
| 1734 |
GetForeignServerExtended(Oid serverid, bits16 flags); |
| 1735 |
</programlisting> |
| 1736 |
|
| 1737 |
This function returns a <structname>ForeignServer</structname> object |
| 1738 |
for the foreign server with the given OID. A |
| 1739 |
<structname>ForeignServer</structname> object contains properties |
| 1740 |
of the server (see <filename>foreign/foreign.h</filename> for details). |
| 1741 |
<structfield>flags</structfield> is a bitwise-or'd bit mask indicating |
| 1742 |
an extra set of options. It can take the value |
| 1743 |
<literal>FSV_MISSING_OK</literal>, in which case a <literal>NULL</literal> |
| 1744 |
result is returned to the caller instead of an error for an undefined |
| 1745 |
object. |
| 1746 |
</para> |
| 1747 |
|
| 1748 |
<para> |
| 1749 |
<programlisting> |
| 1750 |
ForeignServer * |
| 1751 |
GetForeignServer(Oid serverid); |
| 1752 |
</programlisting> |
| 1753 |
|
| 1754 |
This function returns a <structname>ForeignServer</structname> object |
| 1755 |
for the foreign server with the given OID. A |
| 1756 |
<structname>ForeignServer</structname> object contains properties |
| 1757 |
of the server (see <filename>foreign/foreign.h</filename> for details). |
| 1758 |
</para> |
| 1759 |
|
| 1760 |
<para> |
| 1761 |
<programlisting> |
| 1762 |
UserMapping * |
| 1763 |
GetUserMapping(Oid userid, Oid serverid); |
| 1764 |
</programlisting> |
| 1765 |
|
| 1766 |
This function returns a <structname>UserMapping</structname> object for |
| 1767 |
the user mapping of the given role on the given server. (If there is no |
| 1768 |
mapping for the specific user, it will return the mapping for |
| 1769 |
<literal>PUBLIC</literal>, or throw error if there is none.) A |
| 1770 |
<structname>UserMapping</structname> object contains properties of the |
| 1771 |
user mapping (see <filename>foreign/foreign.h</filename> for details). |
| 1772 |
</para> |
| 1773 |
|
| 1774 |
<para> |
| 1775 |
<programlisting> |
| 1776 |
ForeignTable * |
| 1777 |
GetForeignTable(Oid relid); |
| 1778 |
</programlisting> |
| 1779 |
|
| 1780 |
This function returns a <structname>ForeignTable</structname> object for |
| 1781 |
the foreign table with the given OID. A |
| 1782 |
<structname>ForeignTable</structname> object contains properties of the |
| 1783 |
foreign table (see <filename>foreign/foreign.h</filename> for details). |
| 1784 |
</para> |
| 1785 |
|
| 1786 |
<para> |
| 1787 |
<programlisting> |
| 1788 |
List * |
| 1789 |
GetForeignColumnOptions(Oid relid, AttrNumber attnum); |
| 1790 |
</programlisting> |
| 1791 |
|
| 1792 |
This function returns the per-column FDW options for the column with the |
| 1793 |
given foreign table OID and attribute number, in the form of a list of |
| 1794 |
<structname>DefElem</structname>. NIL is returned if the column has no |
| 1795 |
options. |
| 1796 |
</para> |
| 1797 |
|
| 1798 |
<para> |
| 1799 |
Some object types have name-based lookup functions in addition to the |
| 1800 |
OID-based ones: |
| 1801 |
</para> |
| 1802 |
|
| 1803 |
<para> |
| 1804 |
<programlisting> |
| 1805 |
ForeignDataWrapper * |
| 1806 |
GetForeignDataWrapperByName(const char *name, bool missing_ok); |
| 1807 |
</programlisting> |
| 1808 |
|
| 1809 |
This function returns a <structname>ForeignDataWrapper</structname> |
| 1810 |
object for the foreign-data wrapper with the given name. If the wrapper |
| 1811 |
is not found, return NULL if missing_ok is true, otherwise raise an |
| 1812 |
error. |
| 1813 |
</para> |
| 1814 |
|
| 1815 |
<para> |
| 1816 |
<programlisting> |
| 1817 |
ForeignServer * |
| 1818 |
GetForeignServerByName(const char *name, bool missing_ok); |
| 1819 |
</programlisting> |
| 1820 |
|
| 1821 |
This function returns a <structname>ForeignServer</structname> object |
| 1822 |
for the foreign server with the given name. If the server is not found, |
| 1823 |
return NULL if missing_ok is true, otherwise raise an error. |
| 1824 |
</para> |
| 1825 |
|
| 1826 |
</sect1> |
| 1827 |
|
| 1828 |
<sect1 id="fdw-planning"> |
| 1829 |
<title>Foreign Data Wrapper Query Planning</title> |
| 1830 |
|
| 1831 |
<para> |
| 1832 |
The FDW callback functions <function>GetForeignRelSize</function>, |
| 1833 |
<function>GetForeignPaths</function>, <function>GetForeignPlan</function>, |
| 1834 |
<function>PlanForeignModify</function>, <function>GetForeignJoinPaths</function>, |
| 1835 |
<function>GetForeignUpperPaths</function>, and <function>PlanDirectModify</function> |
| 1836 |
must fit into the workings of the <productname>PostgreSQL</productname> planner. |
| 1837 |
Here are some notes about what they must do. |
| 1838 |
</para> |
| 1839 |
|
| 1840 |
<para> |
| 1841 |
The information in <literal>root</literal> and <literal>baserel</literal> can be used |
| 1842 |
to reduce the amount of information that has to be fetched from the |
| 1843 |
foreign table (and therefore reduce the cost). |
| 1844 |
<literal>baserel->baserestrictinfo</literal> is particularly interesting, as |
| 1845 |
it contains restriction quals (<literal>WHERE</literal> clauses) that should be |
| 1846 |
used to filter the rows to be fetched. (The FDW itself is not required |
| 1847 |
to enforce these quals, as the core executor can check them instead.) |
| 1848 |
<literal>baserel->reltarget->exprs</literal> can be used to determine which |
| 1849 |
columns need to be fetched; but note that it only lists columns that |
| 1850 |
have to be emitted by the <structname>ForeignScan</structname> plan node, not |
| 1851 |
columns that are used in qual evaluation but not output by the query. |
| 1852 |
</para> |
| 1853 |
|
| 1854 |
<para> |
| 1855 |
Various private fields are available for the FDW planning functions to |
| 1856 |
keep information in. Generally, whatever you store in FDW private fields |
| 1857 |
should be palloc'd, so that it will be reclaimed at the end of planning. |
| 1858 |
</para> |
| 1859 |
|
| 1860 |
<para> |
| 1861 |
<literal>baserel->fdw_private</literal> is a <type>void</type> pointer that is |
| 1862 |
available for FDW planning functions to store information relevant to |
| 1863 |
the particular foreign table. The core planner does not touch it except |
| 1864 |
to initialize it to NULL when the <literal>RelOptInfo</literal> node is created. |
| 1865 |
It is useful for passing information forward from |
| 1866 |
<function>GetForeignRelSize</function> to <function>GetForeignPaths</function> and/or |
| 1867 |
<function>GetForeignPaths</function> to <function>GetForeignPlan</function>, thereby |
| 1868 |
avoiding recalculation. |
| 1869 |
</para> |
| 1870 |
|
| 1871 |
<para> |
| 1872 |
<function>GetForeignPaths</function> can identify the meaning of different |
| 1873 |
access paths by storing private information in the |
| 1874 |
<structfield>fdw_private</structfield> field of <structname>ForeignPath</structname> nodes. |
| 1875 |
<structfield>fdw_private</structfield> is declared as a <type>List</type> pointer, but |
| 1876 |
could actually contain anything since the core planner does not touch |
| 1877 |
it. However, best practice is to use a representation that's dumpable |
| 1878 |
by <function>nodeToString</function>, for use with debugging support available |
| 1879 |
in the backend. |
| 1880 |
</para> |
| 1881 |
|
| 1882 |
<para> |
| 1883 |
<function>GetForeignPlan</function> can examine the <structfield>fdw_private</structfield> |
| 1884 |
field of the selected <structname>ForeignPath</structname> node, and can generate |
| 1885 |
<structfield>fdw_exprs</structfield> and <structfield>fdw_private</structfield> lists to be |
| 1886 |
placed in the <structname>ForeignScan</structname> plan node, where they will be |
| 1887 |
available at execution time. Both of these lists must be |
| 1888 |
represented in a form that <function>copyObject</function> knows how to copy. |
| 1889 |
The <structfield>fdw_private</structfield> list has no other restrictions and is |
| 1890 |
not interpreted by the core backend in any way. The |
| 1891 |
<structfield>fdw_exprs</structfield> list, if not NIL, is expected to contain |
| 1892 |
expression trees that are intended to be executed at run time. These |
| 1893 |
trees will undergo post-processing by the planner to make them fully |
| 1894 |
executable. |
| 1895 |
</para> |
| 1896 |
|
| 1897 |
<para> |
| 1898 |
In <function>GetForeignPlan</function>, generally the passed-in target list can |
| 1899 |
be copied into the plan node as-is. The passed <literal>scan_clauses</literal> list |
| 1900 |
contains the same clauses as <literal>baserel->baserestrictinfo</literal>, |
| 1901 |
but may be re-ordered for better execution efficiency. In simple cases |
| 1902 |
the FDW can just strip <structname>RestrictInfo</structname> nodes from the |
| 1903 |
<literal>scan_clauses</literal> list (using <function>extract_actual_clauses</function>) and put |
| 1904 |
all the clauses into the plan node's qual list, which means that all the |
| 1905 |
clauses will be checked by the executor at run time. More complex FDWs |
| 1906 |
may be able to check some of the clauses internally, in which case those |
| 1907 |
clauses can be removed from the plan node's qual list so that the |
| 1908 |
executor doesn't waste time rechecking them. |
| 1909 |
</para> |
| 1910 |
|
| 1911 |
<para> |
| 1912 |
As an example, the FDW might identify some restriction clauses of the |
| 1913 |
form <replaceable>foreign_variable</replaceable> <literal>=</literal> |
| 1914 |
<replaceable>sub_expression</replaceable>, which it determines can be executed on |
| 1915 |
the remote server given the locally-evaluated value of the |
| 1916 |
<replaceable>sub_expression</replaceable>. The actual identification of such a |
| 1917 |
clause should happen during <function>GetForeignPaths</function>, since it would |
| 1918 |
affect the cost estimate for the path. The path's |
| 1919 |
<structfield>fdw_private</structfield> field would probably include a pointer to |
| 1920 |
the identified clause's <structname>RestrictInfo</structname> node. Then |
| 1921 |
<function>GetForeignPlan</function> would remove that clause from <literal>scan_clauses</literal>, |
| 1922 |
but add the <replaceable>sub_expression</replaceable> to <structfield>fdw_exprs</structfield> |
| 1923 |
to ensure that it gets massaged into executable form. It would probably |
| 1924 |
also put control information into the plan node's |
| 1925 |
<structfield>fdw_private</structfield> field to tell the execution functions what |
| 1926 |
to do at run time. The query transmitted to the remote server would |
| 1927 |
involve something like <literal>WHERE <replaceable>foreign_variable</replaceable> = |
| 1928 |
$1</literal>, with the parameter value obtained at run time from |
| 1929 |
evaluation of the <structfield>fdw_exprs</structfield> expression tree. |
| 1930 |
</para> |
| 1931 |
|
| 1932 |
<para> |
| 1933 |
Any clauses removed from the plan node's qual list must instead be added |
| 1934 |
to <literal>fdw_recheck_quals</literal> or rechecked by |
| 1935 |
<literal>RecheckForeignScan</literal> in order to ensure correct behavior |
| 1936 |
at the <literal>READ COMMITTED</literal> isolation level. When a concurrent |
| 1937 |
update occurs for some other table involved in the query, the executor |
| 1938 |
may need to verify that all of the original quals are still satisfied for |
| 1939 |
the tuple, possibly against a different set of parameter values. Using |
| 1940 |
<literal>fdw_recheck_quals</literal> is typically easier than implementing checks |
| 1941 |
inside <literal>RecheckForeignScan</literal>, but this method will be |
| 1942 |
insufficient when outer joins have been pushed down, since the join tuples |
| 1943 |
in that case might have some fields go to NULL without rejecting the |
| 1944 |
tuple entirely. |
| 1945 |
</para> |
| 1946 |
|
| 1947 |
<para> |
| 1948 |
Another <structname>ForeignScan</structname> field that can be filled by FDWs |
| 1949 |
is <structfield>fdw_scan_tlist</structfield>, which describes the tuples returned by |
| 1950 |
the FDW for this plan node. For simple foreign table scans this can be |
| 1951 |
set to <literal>NIL</literal>, implying that the returned tuples have the |
| 1952 |
row type declared for the foreign table. A non-<symbol>NIL</symbol> value must be a |
| 1953 |
target list (list of <structname>TargetEntry</structname>s) containing Vars and/or |
| 1954 |
expressions representing the returned columns. This might be used, for |
| 1955 |
example, to show that the FDW has omitted some columns that it noticed |
| 1956 |
won't be needed for the query. Also, if the FDW can compute expressions |
| 1957 |
used by the query more cheaply than can be done locally, it could add |
| 1958 |
those expressions to <structfield>fdw_scan_tlist</structfield>. Note that join |
| 1959 |
plans (created from paths made by <function>GetForeignJoinPaths</function>) must |
| 1960 |
always supply <structfield>fdw_scan_tlist</structfield> to describe the set of |
| 1961 |
columns they will return. |
| 1962 |
</para> |
| 1963 |
|
| 1964 |
<para> |
| 1965 |
The FDW should always construct at least one path that depends only on |
| 1966 |
the table's restriction clauses. In join queries, it might also choose |
| 1967 |
to construct path(s) that depend on join clauses, for example |
| 1968 |
<replaceable>foreign_variable</replaceable> <literal>=</literal> |
| 1969 |
<replaceable>local_variable</replaceable>. Such clauses will not be found in |
| 1970 |
<literal>baserel->baserestrictinfo</literal> but must be sought in the |
| 1971 |
relation's join lists. A path using such a clause is called a |
| 1972 |
<quote>parameterized path</quote>. It must identify the other relations |
| 1973 |
used in the selected join clause(s) with a suitable value of |
| 1974 |
<literal>param_info</literal>; use <function>get_baserel_parampathinfo</function> |
| 1975 |
to compute that value. In <function>GetForeignPlan</function>, the |
| 1976 |
<replaceable>local_variable</replaceable> portion of the join clause would be added |
| 1977 |
to <structfield>fdw_exprs</structfield>, and then at run time the case works the |
| 1978 |
same as for an ordinary restriction clause. |
| 1979 |
</para> |
| 1980 |
|
| 1981 |
<para> |
| 1982 |
If an FDW supports remote joins, <function>GetForeignJoinPaths</function> should |
| 1983 |
produce <structname>ForeignPath</structname>s for potential remote joins in much |
| 1984 |
the same way as <function>GetForeignPaths</function> works for base tables. |
| 1985 |
Information about the intended join can be passed forward |
| 1986 |
to <function>GetForeignPlan</function> in the same ways described above. |
| 1987 |
However, <structfield>baserestrictinfo</structfield> is not relevant for join |
| 1988 |
relations; instead, the relevant join clauses for a particular join are |
| 1989 |
passed to <function>GetForeignJoinPaths</function> as a separate parameter |
| 1990 |
(<literal>extra->restrictlist</literal>). |
| 1991 |
</para> |
| 1992 |
|
| 1993 |
<para> |
| 1994 |
An FDW might additionally support direct execution of some plan actions |
| 1995 |
that are above the level of scans and joins, such as grouping or |
| 1996 |
aggregation. To offer such options, the FDW should generate paths and |
| 1997 |
insert them into the appropriate <firstterm>upper relation</firstterm>. For |
| 1998 |
example, a path representing remote aggregation should be inserted into |
| 1999 |
the <literal>UPPERREL_GROUP_AGG</literal> relation, using <function>add_path</function>. |
| 2000 |
This path will be compared on a cost basis with local aggregation |
| 2001 |
performed by reading a simple scan path for the foreign relation (note |
| 2002 |
that such a path must also be supplied, else there will be an error at |
| 2003 |
plan time). If the remote-aggregation path wins, which it usually would, |
| 2004 |
it will be converted into a plan in the usual way, by |
| 2005 |
calling <function>GetForeignPlan</function>. The recommended place to generate |
| 2006 |
such paths is in the <function>GetForeignUpperPaths</function> |
| 2007 |
callback function, which is called for each upper relation (i.e., each |
| 2008 |
post-scan/join processing step), if all the base relations of the query |
| 2009 |
come from the same FDW. |
| 2010 |
</para> |
| 2011 |
|
| 2012 |
<para> |
| 2013 |
<function>PlanForeignModify</function> and the other callbacks described in |
| 2014 |
<xref linkend="fdw-callbacks-update"/> are designed around the assumption |
| 2015 |
that the foreign relation will be scanned in the usual way and then |
| 2016 |
individual row updates will be driven by a local <literal>ModifyTable</literal> |
| 2017 |
plan node. This approach is necessary for the general case where an |
| 2018 |
update requires reading local tables as well as foreign tables. |
| 2019 |
However, if the operation could be executed entirely by the foreign |
| 2020 |
server, the FDW could generate a path representing that and insert it |
| 2021 |
into the <literal>UPPERREL_FINAL</literal> upper relation, where it would |
| 2022 |
compete against the <literal>ModifyTable</literal> approach. This approach |
| 2023 |
could also be used to implement remote <literal>SELECT FOR UPDATE</literal>, |
| 2024 |
rather than using the row locking callbacks described in |
| 2025 |
<xref linkend="fdw-callbacks-row-locking"/>. Keep in mind that a path |
| 2026 |
inserted into <literal>UPPERREL_FINAL</literal> is responsible for |
| 2027 |
implementing <emphasis>all</emphasis> behavior of the query. |
| 2028 |
</para> |
| 2029 |
|
| 2030 |
<para> |
| 2031 |
When planning an <command>UPDATE</command> or <command>DELETE</command>, |
| 2032 |
<function>PlanForeignModify</function> and <function>PlanDirectModify</function> |
| 2033 |
can look up the <structname>RelOptInfo</structname> |
| 2034 |
struct for the foreign table and make use of the |
| 2035 |
<literal>baserel->fdw_private</literal> data previously created by the |
| 2036 |
scan-planning functions. However, in <command>INSERT</command> the target |
| 2037 |
table is not scanned so there is no <structname>RelOptInfo</structname> for it. |
| 2038 |
The <structname>List</structname> returned by <function>PlanForeignModify</function> has |
| 2039 |
the same restrictions as the <structfield>fdw_private</structfield> list of a |
| 2040 |
<structname>ForeignScan</structname> plan node, that is it must contain only |
| 2041 |
structures that <function>copyObject</function> knows how to copy. |
| 2042 |
</para> |
| 2043 |
|
| 2044 |
<para> |
| 2045 |
<command>INSERT</command> with an <literal>ON CONFLICT</literal> clause does not |
| 2046 |
support specifying the conflict target, as unique constraints or |
| 2047 |
exclusion constraints on remote tables are not locally known. This |
| 2048 |
in turn implies that <literal>ON CONFLICT DO UPDATE</literal> is not supported, |
| 2049 |
since the specification is mandatory there. |
| 2050 |
</para> |
| 2051 |
|
| 2052 |
</sect1> |
| 2053 |
|
| 2054 |
<sect1 id="fdw-row-locking"> |
| 2055 |
<title>Row Locking in Foreign Data Wrappers</title> |
| 2056 |
|
| 2057 |
<para> |
| 2058 |
If an FDW's underlying storage mechanism has a concept of locking |
| 2059 |
individual rows to prevent concurrent updates of those rows, it is |
| 2060 |
usually worthwhile for the FDW to perform row-level locking with as |
| 2061 |
close an approximation as practical to the semantics used in |
| 2062 |
ordinary <productname>PostgreSQL</productname> tables. There are multiple |
| 2063 |
considerations involved in this. |
| 2064 |
</para> |
| 2065 |
|
| 2066 |
<para> |
| 2067 |
One key decision to be made is whether to perform <firstterm>early |
| 2068 |
locking</firstterm> or <firstterm>late locking</firstterm>. In early locking, a row is |
| 2069 |
locked when it is first retrieved from the underlying store, while in |
| 2070 |
late locking, the row is locked only when it is known that it needs to |
| 2071 |
be locked. (The difference arises because some rows may be discarded by |
| 2072 |
locally-checked restriction or join conditions.) Early locking is much |
| 2073 |
simpler and avoids extra round trips to a remote store, but it can cause |
| 2074 |
locking of rows that need not have been locked, resulting in reduced |
| 2075 |
concurrency or even unexpected deadlocks. Also, late locking is only |
| 2076 |
possible if the row to be locked can be uniquely re-identified later. |
| 2077 |
Preferably the row identifier should identify a specific version of the |
| 2078 |
row, as <productname>PostgreSQL</productname> TIDs do. |
| 2079 |
</para> |
| 2080 |
|
| 2081 |
<para> |
| 2082 |
By default, <productname>PostgreSQL</productname> ignores locking considerations |
| 2083 |
when interfacing to FDWs, but an FDW can perform early locking without |
| 2084 |
any explicit support from the core code. The API functions described |
| 2085 |
in <xref linkend="fdw-callbacks-row-locking"/>, which were added |
| 2086 |
in <productname>PostgreSQL</productname> 9.5, allow an FDW to use late locking if |
| 2087 |
it wishes. |
| 2088 |
</para> |
| 2089 |
|
| 2090 |
<para> |
| 2091 |
An additional consideration is that in <literal>READ COMMITTED</literal> |
| 2092 |
isolation mode, <productname>PostgreSQL</productname> may need to re-check |
| 2093 |
restriction and join conditions against an updated version of some |
| 2094 |
target tuple. Rechecking join conditions requires re-obtaining copies |
| 2095 |
of the non-target rows that were previously joined to the target tuple. |
| 2096 |
When working with standard <productname>PostgreSQL</productname> tables, this is |
| 2097 |
done by including the TIDs of the non-target tables in the column list |
| 2098 |
projected through the join, and then re-fetching non-target rows when |
| 2099 |
required. This approach keeps the join data set compact, but it |
| 2100 |
requires inexpensive re-fetch capability, as well as a TID that can |
| 2101 |
uniquely identify the row version to be re-fetched. By default, |
| 2102 |
therefore, the approach used with foreign tables is to include a copy of |
| 2103 |
the entire row fetched from a foreign table in the column list projected |
| 2104 |
through the join. This puts no special demands on the FDW but can |
| 2105 |
result in reduced performance of merge and hash joins. An FDW that is |
| 2106 |
capable of meeting the re-fetch requirements can choose to do it the |
| 2107 |
first way. |
| 2108 |
</para> |
| 2109 |
|
| 2110 |
<para> |
| 2111 |
For an <command>UPDATE</command> or <command>DELETE</command> on a foreign table, it |
| 2112 |
is recommended that the <literal>ForeignScan</literal> operation on the target |
| 2113 |
table perform early locking on the rows that it fetches, perhaps via the |
| 2114 |
equivalent of <command>SELECT FOR UPDATE</command>. An FDW can detect whether |
| 2115 |
a table is an <command>UPDATE</command>/<command>DELETE</command> target at plan time |
| 2116 |
by comparing its relid to <literal>root->parse->resultRelation</literal>, |
| 2117 |
or at execution time by using <function>ExecRelationIsTargetRelation()</function>. |
| 2118 |
An alternative possibility is to perform late locking within the |
| 2119 |
<function>ExecForeignUpdate</function> or <function>ExecForeignDelete</function> |
| 2120 |
callback, but no special support is provided for this. |
| 2121 |
</para> |
| 2122 |
|
| 2123 |
<para> |
| 2124 |
For foreign tables that are specified to be locked by a <command>SELECT |
| 2125 |
FOR UPDATE/SHARE</command> command, the <literal>ForeignScan</literal> operation can |
| 2126 |
again perform early locking by fetching tuples with the equivalent |
| 2127 |
of <command>SELECT FOR UPDATE/SHARE</command>. To perform late locking |
| 2128 |
instead, provide the callback functions defined |
| 2129 |
in <xref linkend="fdw-callbacks-row-locking"/>. |
| 2130 |
In <function>GetForeignRowMarkType</function>, select rowmark option |
| 2131 |
<literal>ROW_MARK_EXCLUSIVE</literal>, <literal>ROW_MARK_NOKEYEXCLUSIVE</literal>, |
| 2132 |
<literal>ROW_MARK_SHARE</literal>, or <literal>ROW_MARK_KEYSHARE</literal> depending |
| 2133 |
on the requested lock strength. (The core code will act the same |
| 2134 |
regardless of which of these four options you choose.) |
| 2135 |
Elsewhere, you can detect whether a foreign table was specified to be |
| 2136 |
locked by this type of command by using <function>get_plan_rowmark</function> at |
| 2137 |
plan time, or <function>ExecFindRowMark</function> at execution time; you must |
| 2138 |
check not only whether a non-null rowmark struct is returned, but that |
| 2139 |
its <structfield>strength</structfield> field is not <literal>LCS_NONE</literal>. |
| 2140 |
</para> |
| 2141 |
|
| 2142 |
<para> |
| 2143 |
Lastly, for foreign tables that are used in an <command>UPDATE</command>, |
| 2144 |
<command>DELETE</command> or <command>SELECT FOR UPDATE/SHARE</command> command but |
| 2145 |
are not specified to be row-locked, you can override the default choice |
| 2146 |
to copy entire rows by having <function>GetForeignRowMarkType</function> select |
| 2147 |
option <literal>ROW_MARK_REFERENCE</literal> when it sees lock strength |
| 2148 |
<literal>LCS_NONE</literal>. This will cause <function>RefetchForeignRow</function> to |
| 2149 |
be called with that value for <structfield>markType</structfield>; it should then |
| 2150 |
re-fetch the row without acquiring any new lock. (If you have |
| 2151 |
a <function>GetForeignRowMarkType</function> function but don't wish to re-fetch |
| 2152 |
unlocked rows, select option <literal>ROW_MARK_COPY</literal> |
| 2153 |
for <literal>LCS_NONE</literal>.) |
| 2154 |
</para> |
| 2155 |
|
| 2156 |
<para> |
| 2157 |
See <filename>src/include/nodes/lockoptions.h</filename>, the comments |
| 2158 |
for <type>RowMarkType</type> and <type>PlanRowMark</type> |
| 2159 |
in <filename>src/include/nodes/plannodes.h</filename>, and the comments for |
| 2160 |
<type>ExecRowMark</type> in <filename>src/include/nodes/execnodes.h</filename> for |
| 2161 |
additional information. |
| 2162 |
</para> |
| 2163 |
|
| 2164 |
</sect1> |
| 2165 |
|
| 2166 |
</chapter> |