| 1 |
<!-- doc/src/sgml/arch-dev.sgml --> |
| 2 |
|
| 3 |
<chapter id="overview"> |
| 4 |
<title>Overview of PostgreSQL Internals</title> |
| 5 |
|
| 6 |
<note> |
| 7 |
<title>Author</title> |
| 8 |
<para> |
| 9 |
This chapter originated as part of |
| 10 |
<xref linkend="sim98"/> Stefan Simkovics' |
| 11 |
Master's Thesis prepared at Vienna University of Technology under the direction |
| 12 |
of O.Univ.Prof.Dr. Georg Gottlob and Univ.Ass. Mag. Katrin Seyr. |
| 13 |
</para> |
| 14 |
</note> |
| 15 |
|
| 16 |
<para> |
| 17 |
This chapter gives an overview of the internal structure of the |
| 18 |
backend of <productname>PostgreSQL</productname>. After having |
| 19 |
read the following sections you should have an idea of how a query |
| 20 |
is processed. This chapter is intended to help the reader |
| 21 |
understand the general sequence of operations that occur within the |
| 22 |
backend from the point at which a query is received, to the point |
| 23 |
at which the results are returned to the client. |
| 24 |
</para> |
| 25 |
|
| 26 |
<sect1 id="query-path"> |
| 27 |
<title>The Path of a Query</title> |
| 28 |
|
| 29 |
<para> |
| 30 |
Here we give a short overview of the stages a query has to pass |
| 31 |
to obtain a result. |
| 32 |
</para> |
| 33 |
|
| 34 |
<procedure> |
| 35 |
<step> |
| 36 |
<para> |
| 37 |
A connection from an application program to the <productname>PostgreSQL</productname> |
| 38 |
server has to be established. The application program transmits a |
| 39 |
query to the server and waits to receive the results sent back by the |
| 40 |
server. |
| 41 |
</para> |
| 42 |
</step> |
| 43 |
|
| 44 |
<step> |
| 45 |
<para> |
| 46 |
The <firstterm>parser stage</firstterm> checks the query |
| 47 |
transmitted by the application |
| 48 |
program for correct syntax and creates |
| 49 |
a <firstterm>query tree</firstterm>. |
| 50 |
</para> |
| 51 |
</step> |
| 52 |
|
| 53 |
<step> |
| 54 |
<para> |
| 55 |
The <firstterm>rewrite system</firstterm> takes |
| 56 |
the query tree created by the parser stage and looks for |
| 57 |
any <firstterm>rules</firstterm> (stored in the |
| 58 |
<firstterm>system catalogs</firstterm>) to apply to |
| 59 |
the query tree. It performs the |
| 60 |
transformations given in the <firstterm>rule bodies</firstterm>. |
| 61 |
</para> |
| 62 |
|
| 63 |
<para> |
| 64 |
One application of the rewrite system is in the realization of |
| 65 |
<firstterm>views</firstterm>. |
| 66 |
Whenever a query against a view |
| 67 |
(i.e., a <firstterm>virtual table</firstterm>) is made, |
| 68 |
the rewrite system rewrites the user's query to |
| 69 |
a query that accesses the <firstterm>base tables</firstterm> given in |
| 70 |
the <firstterm>view definition</firstterm> instead. |
| 71 |
</para> |
| 72 |
</step> |
| 73 |
|
| 74 |
<step> |
| 75 |
<para> |
| 76 |
The <firstterm>planner/optimizer</firstterm> takes |
| 77 |
the (rewritten) query tree and creates a |
| 78 |
<firstterm>query plan</firstterm> that will be the input to the |
| 79 |
<firstterm>executor</firstterm>. |
| 80 |
</para> |
| 81 |
|
| 82 |
<para> |
| 83 |
It does so by first creating all possible <firstterm>paths</firstterm> |
| 84 |
leading to the same result. For example if there is an index on a |
| 85 |
relation to be scanned, there are two paths for the |
| 86 |
scan. One possibility is a simple sequential scan and the other |
| 87 |
possibility is to use the index. Next the cost for the execution of |
| 88 |
each path is estimated and the cheapest path is chosen. The cheapest |
| 89 |
path is expanded into a complete plan that the executor can use. |
| 90 |
</para> |
| 91 |
</step> |
| 92 |
|
| 93 |
<step> |
| 94 |
<para> |
| 95 |
The executor recursively steps through |
| 96 |
the <firstterm>plan tree</firstterm> and |
| 97 |
retrieves rows in the way represented by the plan. |
| 98 |
The executor makes use of the |
| 99 |
<firstterm>storage system</firstterm> while scanning |
| 100 |
relations, performs <firstterm>sorts</firstterm> and <firstterm>joins</firstterm>, |
| 101 |
evaluates <firstterm>qualifications</firstterm> and finally hands back the rows derived. |
| 102 |
</para> |
| 103 |
</step> |
| 104 |
</procedure> |
| 105 |
|
| 106 |
<para> |
| 107 |
In the following sections we will cover each of the above listed items |
| 108 |
in more detail to give a better understanding of <productname>PostgreSQL</productname>'s internal |
| 109 |
control and data structures. |
| 110 |
</para> |
| 111 |
</sect1> |
| 112 |
|
| 113 |
<sect1 id="connect-estab"> |
| 114 |
<title>How Connections Are Established</title> |
| 115 |
|
| 116 |
<para> |
| 117 |
<productname>PostgreSQL</productname> implements a |
| 118 |
<quote>process per user</quote> client/server model. |
| 119 |
In this model, every |
| 120 |
<glossterm linkend="glossary-client">client process</glossterm> |
| 121 |
connects to exactly one |
| 122 |
<glossterm linkend="glossary-backend">backend process</glossterm>. |
| 123 |
As we do not know ahead of time how many connections will be made, |
| 124 |
we have to use a <quote>supervisor process</quote> that spawns a new |
| 125 |
backend process every time a connection is requested. This supervisor |
| 126 |
process is called |
| 127 |
<glossterm linkend="glossary-postmaster">postmaster</glossterm> |
| 128 |
and listens at a specified TCP/IP port for incoming connections. |
| 129 |
Whenever it detects a request for a connection, it spawns a new |
| 130 |
backend process. Those backend processes communicate with each |
| 131 |
other and with other processes of the |
| 132 |
<glossterm linkend="glossary-instance">instance</glossterm> |
| 133 |
using <firstterm>semaphores</firstterm> and |
| 134 |
<glossterm linkend="glossary-shared-memory">shared memory</glossterm> |
| 135 |
to ensure data integrity throughout concurrent data access. |
| 136 |
</para> |
| 137 |
|
| 138 |
<para> |
| 139 |
The client process can be any program that understands the |
| 140 |
<productname>PostgreSQL</productname> protocol described in |
| 141 |
<xref linkend="protocol"/>. Many clients are based on the |
| 142 |
C-language library <application>libpq</application>, but several independent |
| 143 |
implementations of the protocol exist, such as the Java |
| 144 |
<application>JDBC</application> driver. |
| 145 |
</para> |
| 146 |
|
| 147 |
<para> |
| 148 |
Once a connection is established, the client process can send a query |
| 149 |
to the backend process it's connected to. The query is transmitted using |
| 150 |
plain text, i.e., there is no parsing done in the client. The backend |
| 151 |
process parses the query, creates an <firstterm>execution plan</firstterm>, |
| 152 |
executes the plan, and returns the retrieved rows to the client |
| 153 |
by transmitting them over the established connection. |
| 154 |
</para> |
| 155 |
</sect1> |
| 156 |
|
| 157 |
<sect1 id="parser-stage"> |
| 158 |
<title>The Parser Stage</title> |
| 159 |
|
| 160 |
<para> |
| 161 |
The <firstterm>parser stage</firstterm> consists of two parts: |
| 162 |
|
| 163 |
<itemizedlist> |
| 164 |
<listitem> |
| 165 |
<para> |
| 166 |
The <firstterm>parser</firstterm> defined in |
| 167 |
<filename>gram.y</filename> and <filename>scan.l</filename> is |
| 168 |
built using the Unix tools <application>bison</application> |
| 169 |
and <application>flex</application>. |
| 170 |
</para> |
| 171 |
</listitem> |
| 172 |
<listitem> |
| 173 |
<para> |
| 174 |
The <firstterm>transformation process</firstterm> does |
| 175 |
modifications and augmentations to the data structures returned by the parser. |
| 176 |
</para> |
| 177 |
</listitem> |
| 178 |
</itemizedlist> |
| 179 |
</para> |
| 180 |
|
| 181 |
<sect2 id="parser-stage-parser"> |
| 182 |
<title>Parser</title> |
| 183 |
|
| 184 |
<para> |
| 185 |
The parser has to check the query string (which arrives as plain |
| 186 |
text) for valid syntax. If the syntax is correct a |
| 187 |
<firstterm>parse tree</firstterm> is built up and handed back; |
| 188 |
otherwise an error is returned. The parser and lexer are |
| 189 |
implemented using the well-known Unix tools <application>bison</application> |
| 190 |
and <application>flex</application>. |
| 191 |
</para> |
| 192 |
|
| 193 |
<para> |
| 194 |
The <firstterm>lexer</firstterm> is defined in the file |
| 195 |
<filename>scan.l</filename> and is responsible |
| 196 |
for recognizing <firstterm>identifiers</firstterm>, |
| 197 |
the <firstterm>SQL key words</firstterm> etc. For |
| 198 |
every key word or identifier that is found, a <firstterm>token</firstterm> |
| 199 |
is generated and handed to the parser. |
| 200 |
</para> |
| 201 |
|
| 202 |
<para> |
| 203 |
The parser is defined in the file <filename>gram.y</filename> and |
| 204 |
consists of a set of <firstterm>grammar rules</firstterm> and |
| 205 |
<firstterm>actions</firstterm> that are executed whenever a rule |
| 206 |
is fired. The code of the actions (which is actually C code) is |
| 207 |
used to build up the parse tree. |
| 208 |
</para> |
| 209 |
|
| 210 |
<para> |
| 211 |
The file <filename>scan.l</filename> is transformed to the C |
| 212 |
source file <filename>scan.c</filename> using the program |
| 213 |
<application>flex</application> and <filename>gram.y</filename> is |
| 214 |
transformed to <filename>gram.c</filename> using |
| 215 |
<application>bison</application>. After these transformations |
| 216 |
have taken place a normal C compiler can be used to create the |
| 217 |
parser. Never make any changes to the generated C files as they |
| 218 |
will be overwritten the next time <application>flex</application> |
| 219 |
or <application>bison</application> is called. |
| 220 |
|
| 221 |
<note> |
| 222 |
<para> |
| 223 |
The mentioned transformations and compilations are normally done |
| 224 |
automatically using the <firstterm>makefiles</firstterm> |
| 225 |
shipped with the <productname>PostgreSQL</productname> |
| 226 |
source distribution. |
| 227 |
</para> |
| 228 |
</note> |
| 229 |
</para> |
| 230 |
|
| 231 |
<para> |
| 232 |
A detailed description of <application>bison</application> or |
| 233 |
the grammar rules given in <filename>gram.y</filename> would be |
| 234 |
beyond the scope of this manual. There are many books and |
| 235 |
documents dealing with <application>flex</application> and |
| 236 |
<application>bison</application>. You should be familiar with |
| 237 |
<application>bison</application> before you start to study the |
| 238 |
grammar given in <filename>gram.y</filename> otherwise you won't |
| 239 |
understand what happens there. |
| 240 |
</para> |
| 241 |
|
| 242 |
</sect2> |
| 243 |
|
| 244 |
<sect2 id="parser-stage-transformation-process"> |
| 245 |
<title>Transformation Process</title> |
| 246 |
|
| 247 |
<para> |
| 248 |
The parser stage creates a parse tree using only fixed rules about |
| 249 |
the syntactic structure of SQL. It does not make any lookups in the |
| 250 |
system catalogs, so there is no possibility to understand the detailed |
| 251 |
semantics of the requested operations. After the parser completes, |
| 252 |
the <firstterm>transformation process</firstterm> takes the tree handed |
| 253 |
back by the parser as input and does the semantic interpretation needed |
| 254 |
to understand which tables, functions, and operators are referenced by |
| 255 |
the query. The data structure that is built to represent this |
| 256 |
information is called the <firstterm>query tree</firstterm>. |
| 257 |
</para> |
| 258 |
|
| 259 |
<para> |
| 260 |
The reason for separating raw parsing from semantic analysis is that |
| 261 |
system catalog lookups can only be done within a transaction, and we |
| 262 |
do not wish to start a transaction immediately upon receiving a query |
| 263 |
string. The raw parsing stage is sufficient to identify the transaction |
| 264 |
control commands (<command>BEGIN</command>, <command>ROLLBACK</command>, etc.), and |
| 265 |
these can then be correctly executed without any further analysis. |
| 266 |
Once we know that we are dealing with an actual query (such as |
| 267 |
<command>SELECT</command> or <command>UPDATE</command>), it is okay to |
| 268 |
start a transaction if we're not already in one. Only then can the |
| 269 |
transformation process be invoked. |
| 270 |
</para> |
| 271 |
|
| 272 |
<para> |
| 273 |
The query tree created by the transformation process is structurally |
| 274 |
similar to the raw parse tree in most places, but it has many differences |
| 275 |
in detail. For example, a <structname>FuncCall</structname> node in the |
| 276 |
parse tree represents something that looks syntactically like a function |
| 277 |
call. This might be transformed to either a <structname>FuncExpr</structname> |
| 278 |
or <structname>Aggref</structname> node depending on whether the referenced |
| 279 |
name turns out to be an ordinary function or an aggregate function. |
| 280 |
Also, information about the actual data types of columns and expression |
| 281 |
results is added to the query tree. |
| 282 |
</para> |
| 283 |
</sect2> |
| 284 |
</sect1> |
| 285 |
|
| 286 |
<sect1 id="rule-system"> |
| 287 |
<title>The <productname>PostgreSQL</productname> Rule System</title> |
| 288 |
|
| 289 |
<para> |
| 290 |
<productname>PostgreSQL</productname> supports a powerful |
| 291 |
<firstterm>rule system</firstterm> for the specification |
| 292 |
of <firstterm>views</firstterm> and ambiguous <firstterm>view updates</firstterm>. |
| 293 |
Originally the <productname>PostgreSQL</productname> |
| 294 |
rule system consisted of two implementations: |
| 295 |
|
| 296 |
<itemizedlist> |
| 297 |
<listitem> |
| 298 |
<para> |
| 299 |
The first one worked using <firstterm>row level</firstterm> processing and was |
| 300 |
implemented deep in the <firstterm>executor</firstterm>. The rule system was |
| 301 |
called whenever an individual row had been accessed. This |
| 302 |
implementation was removed in 1995 when the last official release |
| 303 |
of the <productname>Berkeley Postgres</productname> project was |
| 304 |
transformed into <productname>Postgres95</productname>. |
| 305 |
</para> |
| 306 |
</listitem> |
| 307 |
|
| 308 |
<listitem> |
| 309 |
<para> |
| 310 |
The second implementation of the rule system is a technique |
| 311 |
called <firstterm>query rewriting</firstterm>. |
| 312 |
The <firstterm>rewrite system</firstterm> is a module |
| 313 |
that exists between the <firstterm>parser stage</firstterm> and the |
| 314 |
<firstterm>planner/optimizer</firstterm>. This technique is still implemented. |
| 315 |
</para> |
| 316 |
</listitem> |
| 317 |
</itemizedlist> |
| 318 |
</para> |
| 319 |
|
| 320 |
<para> |
| 321 |
The query rewriter is discussed in some detail in |
| 322 |
<xref linkend="rules"/>, so there is no need to cover it here. |
| 323 |
We will only point out that both the input and the output of the |
| 324 |
rewriter are query trees, that is, there is no change in the |
| 325 |
representation or level of semantic detail in the trees. Rewriting |
| 326 |
can be thought of as a form of macro expansion. |
| 327 |
</para> |
| 328 |
|
| 329 |
</sect1> |
| 330 |
|
| 331 |
<sect1 id="planner-optimizer"> |
| 332 |
<title>Planner/Optimizer</title> |
| 333 |
|
| 334 |
<para> |
| 335 |
The task of the <firstterm>planner/optimizer</firstterm> is to |
| 336 |
create an optimal execution plan. A given SQL query (and hence, a |
| 337 |
query tree) can be actually executed in a wide variety of |
| 338 |
different ways, each of which will produce the same set of |
| 339 |
results. If it is computationally feasible, the query optimizer |
| 340 |
will examine each of these possible execution plans, ultimately |
| 341 |
selecting the execution plan that is expected to run the fastest. |
| 342 |
</para> |
| 343 |
|
| 344 |
<note> |
| 345 |
<para> |
| 346 |
In some situations, examining each possible way in which a query |
| 347 |
can be executed would take an excessive amount of time and memory. |
| 348 |
In particular, this occurs when executing queries |
| 349 |
involving large numbers of join operations. In order to determine |
| 350 |
a reasonable (not necessarily optimal) query plan in a reasonable amount |
| 351 |
of time, <productname>PostgreSQL</productname> uses a <firstterm>Genetic |
| 352 |
Query Optimizer</firstterm> (see <xref linkend="geqo"/>) when the number of joins |
| 353 |
exceeds a threshold (see <xref linkend="guc-geqo-threshold"/>). |
| 354 |
</para> |
| 355 |
</note> |
| 356 |
|
| 357 |
<para> |
| 358 |
The planner's search procedure actually works with data structures |
| 359 |
called <firstterm>paths</firstterm>, which are simply cut-down representations of |
| 360 |
plans containing only as much information as the planner needs to make |
| 361 |
its decisions. After the cheapest path is determined, a full-fledged |
| 362 |
<firstterm>plan tree</firstterm> is built to pass to the executor. This represents |
| 363 |
the desired execution plan in sufficient detail for the executor to run it. |
| 364 |
In the rest of this section we'll ignore the distinction between paths |
| 365 |
and plans. |
| 366 |
</para> |
| 367 |
|
| 368 |
<sect2 id="planner-optimizer-generating-possible-plans"> |
| 369 |
<title>Generating Possible Plans</title> |
| 370 |
|
| 371 |
<para> |
| 372 |
The planner/optimizer starts by generating plans for scanning each |
| 373 |
individual relation (table) used in the query. The possible plans |
| 374 |
are determined by the available indexes on each relation. |
| 375 |
There is always the possibility of performing a |
| 376 |
sequential scan on a relation, so a sequential scan plan is always |
| 377 |
created. Assume an index is defined on a |
| 378 |
relation (for example a B-tree index) and a query contains the |
| 379 |
restriction |
| 380 |
<literal>relation.attribute OPR constant</literal>. If |
| 381 |
<literal>relation.attribute</literal> happens to match the key of the B-tree |
| 382 |
index and <literal>OPR</literal> is one of the operators listed in |
| 383 |
the index's <firstterm>operator class</firstterm>, another plan is created using |
| 384 |
the B-tree index to scan the relation. If there are further indexes |
| 385 |
present and the restrictions in the query happen to match a key of an |
| 386 |
index, further plans will be considered. Index scan plans are also |
| 387 |
generated for indexes that have a sort ordering that can match the |
| 388 |
query's <literal>ORDER BY</literal> clause (if any), or a sort ordering that |
| 389 |
might be useful for merge joining (see below). |
| 390 |
</para> |
| 391 |
|
| 392 |
<para> |
| 393 |
If the query requires joining two or more relations, |
| 394 |
plans for joining relations are considered |
| 395 |
after all feasible plans have been found for scanning single relations. |
| 396 |
The three available join strategies are: |
| 397 |
|
| 398 |
<itemizedlist> |
| 399 |
<listitem> |
| 400 |
<para> |
| 401 |
<firstterm>nested loop join</firstterm>: The right relation is scanned |
| 402 |
once for every row found in the left relation. This strategy |
| 403 |
is easy to implement but can be very time consuming. (However, |
| 404 |
if the right relation can be scanned with an index scan, this can |
| 405 |
be a good strategy. It is possible to use values from the current |
| 406 |
row of the left relation as keys for the index scan of the right.) |
| 407 |
</para> |
| 408 |
</listitem> |
| 409 |
|
| 410 |
<listitem> |
| 411 |
<para> |
| 412 |
<firstterm>merge join</firstterm>: Each relation is sorted on the join |
| 413 |
attributes before the join starts. Then the two relations are |
| 414 |
scanned in parallel, and matching rows are combined to form |
| 415 |
join rows. This kind of join is |
| 416 |
attractive because each relation has to be scanned only once. |
| 417 |
The required sorting might be achieved either by an explicit sort |
| 418 |
step, or by scanning the relation in the proper order using an |
| 419 |
index on the join key. |
| 420 |
</para> |
| 421 |
</listitem> |
| 422 |
|
| 423 |
<listitem> |
| 424 |
<para> |
| 425 |
<firstterm>hash join</firstterm>: the right relation is first scanned |
| 426 |
and loaded into a hash table, using its join attributes as hash keys. |
| 427 |
Next the left relation is scanned and the |
| 428 |
appropriate values of every row found are used as hash keys to |
| 429 |
locate the matching rows in the table. |
| 430 |
</para> |
| 431 |
</listitem> |
| 432 |
</itemizedlist> |
| 433 |
</para> |
| 434 |
|
| 435 |
<para> |
| 436 |
When the query involves more than two relations, the final result |
| 437 |
must be built up by a tree of join steps, each with two inputs. |
| 438 |
The planner examines different possible join sequences to find the |
| 439 |
cheapest one. |
| 440 |
</para> |
| 441 |
|
| 442 |
<para> |
| 443 |
If the query uses fewer than <xref linkend="guc-geqo-threshold"/> |
| 444 |
relations, a near-exhaustive search is conducted to find the best |
| 445 |
join sequence. The planner preferentially considers joins between any |
| 446 |
two relations for which there exists a corresponding join clause in the |
| 447 |
<literal>WHERE</literal> qualification (i.e., for |
| 448 |
which a restriction like <literal>where rel1.attr1=rel2.attr2</literal> |
| 449 |
exists). Join pairs with no join clause are considered only when there |
| 450 |
is no other choice, that is, a particular relation has no available |
| 451 |
join clauses to any other relation. All possible plans are generated for |
| 452 |
every join pair considered by the planner, and the one that is |
| 453 |
(estimated to be) the cheapest is chosen. |
| 454 |
</para> |
| 455 |
|
| 456 |
<para> |
| 457 |
When <varname>geqo_threshold</varname> is exceeded, the join |
| 458 |
sequences considered are determined by heuristics, as described |
| 459 |
in <xref linkend="geqo"/>. Otherwise the process is the same. |
| 460 |
</para> |
| 461 |
|
| 462 |
<para> |
| 463 |
The finished plan tree consists of sequential or index scans of |
| 464 |
the base relations, plus nested-loop, merge, or hash join nodes as |
| 465 |
needed, plus any auxiliary steps needed, such as sort nodes or |
| 466 |
aggregate-function calculation nodes. Most of these plan node |
| 467 |
types have the additional ability to do <firstterm>selection</firstterm> |
| 468 |
(discarding rows that do not meet a specified Boolean condition) |
| 469 |
and <firstterm>projection</firstterm> (computation of a derived column set |
| 470 |
based on given column values, that is, evaluation of scalar |
| 471 |
expressions where needed). One of the responsibilities of the |
| 472 |
planner is to attach selection conditions from the |
| 473 |
<literal>WHERE</literal> clause and computation of required |
| 474 |
output expressions to the most appropriate nodes of the plan |
| 475 |
tree. |
| 476 |
</para> |
| 477 |
</sect2> |
| 478 |
</sect1> |
| 479 |
|
| 480 |
<sect1 id="executor"> |
| 481 |
<title>Executor</title> |
| 482 |
|
| 483 |
<para> |
| 484 |
The <firstterm>executor</firstterm> takes the plan created by the |
| 485 |
planner/optimizer and recursively processes it to extract the required set |
| 486 |
of rows. This is essentially a demand-pull pipeline mechanism. |
| 487 |
Each time a plan node is called, it must deliver one more row, or |
| 488 |
report that it is done delivering rows. |
| 489 |
</para> |
| 490 |
|
| 491 |
<para> |
| 492 |
To provide a concrete example, assume that the top |
| 493 |
node is a <literal>MergeJoin</literal> node. |
| 494 |
Before any merge can be done two rows have to be fetched (one from |
| 495 |
each subplan). So the executor recursively calls itself to |
| 496 |
process the subplans (it starts with the subplan attached to |
| 497 |
<literal>lefttree</literal>). The new top node (the top node of the left |
| 498 |
subplan) is, let's say, a |
| 499 |
<literal>Sort</literal> node and again recursion is needed to obtain |
| 500 |
an input row. The child node of the <literal>Sort</literal> might |
| 501 |
be a <literal>SeqScan</literal> node, representing actual reading of a table. |
| 502 |
Execution of this node causes the executor to fetch a row from the |
| 503 |
table and return it up to the calling node. The <literal>Sort</literal> |
| 504 |
node will repeatedly call its child to obtain all the rows to be sorted. |
| 505 |
When the input is exhausted (as indicated by the child node returning |
| 506 |
a NULL instead of a row), the <literal>Sort</literal> code performs |
| 507 |
the sort, and finally is able to return its first output row, namely |
| 508 |
the first one in sorted order. It keeps the remaining rows stored so |
| 509 |
that it can deliver them in sorted order in response to later demands. |
| 510 |
</para> |
| 511 |
|
| 512 |
<para> |
| 513 |
The <literal>MergeJoin</literal> node similarly demands the first row |
| 514 |
from its right subplan. Then it compares the two rows to see if they |
| 515 |
can be joined; if so, it returns a join row to its caller. On the next |
| 516 |
call, or immediately if it cannot join the current pair of inputs, |
| 517 |
it advances to the next row of one table |
| 518 |
or the other (depending on how the comparison came out), and again |
| 519 |
checks for a match. Eventually, one subplan or the other is exhausted, |
| 520 |
and the <literal>MergeJoin</literal> node returns NULL to indicate that |
| 521 |
no more join rows can be formed. |
| 522 |
</para> |
| 523 |
|
| 524 |
<para> |
| 525 |
Complex queries can involve many levels of plan nodes, but the general |
| 526 |
approach is the same: each node computes and returns its next output |
| 527 |
row each time it is called. Each node is also responsible for applying |
| 528 |
any selection or projection expressions that were assigned to it by |
| 529 |
the planner. |
| 530 |
</para> |
| 531 |
|
| 532 |
<para> |
| 533 |
The executor mechanism is used to evaluate all five basic SQL query |
| 534 |
types: <command>SELECT</command>, <command>INSERT</command>, |
| 535 |
<command>UPDATE</command>, <command>DELETE</command>, and |
| 536 |
<command>MERGE</command>. |
| 537 |
For <command>SELECT</command>, the top-level executor code |
| 538 |
only needs to send each row returned by the query plan tree |
| 539 |
off to the client. <command>INSERT ... SELECT</command>, |
| 540 |
<command>UPDATE</command>, <command>DELETE</command>, and |
| 541 |
<command>MERGE</command> |
| 542 |
are effectively <command>SELECT</command>s under a special |
| 543 |
top-level plan node called <literal>ModifyTable</literal>. |
| 544 |
</para> |
| 545 |
|
| 546 |
<para> |
| 547 |
<command>INSERT ... SELECT</command> feeds the rows up |
| 548 |
to <literal>ModifyTable</literal> for insertion. For |
| 549 |
<command>UPDATE</command>, the planner arranges that each |
| 550 |
computed row includes all the updated column values, plus the |
| 551 |
<firstterm>TID</firstterm> (tuple ID, or row ID) of the original |
| 552 |
target row; this data is fed up to the <literal>ModifyTable</literal> |
| 553 |
node, which uses the information to create a new updated row and |
| 554 |
mark the old row deleted. For <command>DELETE</command>, the only |
| 555 |
column that is actually returned by the plan is the TID, and the |
| 556 |
<literal>ModifyTable</literal> node simply uses the TID to visit each |
| 557 |
target row and mark it deleted. For <command>MERGE</command>, the |
| 558 |
planner joins the source and target relations, and includes all |
| 559 |
column values required by any of the <literal>WHEN</literal> clauses, |
| 560 |
plus the TID of the target row; this data is fed up to the |
| 561 |
<literal>ModifyTable</literal> node, which uses the information to |
| 562 |
work out which <literal>WHEN</literal> clause to execute, and then |
| 563 |
inserts, updates or deletes the target row, as required. |
| 564 |
</para> |
| 565 |
|
| 566 |
<para> |
| 567 |
A simple <command>INSERT ... VALUES</command> command creates a |
| 568 |
trivial plan tree consisting of a single <literal>Result</literal> |
| 569 |
node, which computes just one result row, feeding that up |
| 570 |
to <literal>ModifyTable</literal> to perform the insertion. |
| 571 |
</para> |
| 572 |
|
| 573 |
</sect1> |
| 574 |
|
| 575 |
</chapter> |