1 <?xml version="1.0" encoding="UTF-8" standalone="no"?>
2 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"><html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>58.5. Row Locking in Foreign Data Wrappers</title><link rel="stylesheet" type="text/css" href="stylesheet.css" /><link rev="made" href="pgsql-docs@lists.postgresql.org" /><meta name="generator" content="DocBook XSL Stylesheets Vsnapshot" /><link rel="prev" href="fdw-planning.html" title="58.4. Foreign Data Wrapper Query Planning" /><link rel="next" href="tablesample-method.html" title="Chapter 59. Writing a Table Sampling Method" /></head><body id="docContent" class="container-fluid col-10"><div class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="5" align="center">58.5. Row Locking in Foreign Data Wrappers</th></tr><tr><td width="10%" align="left"><a accesskey="p" href="fdw-planning.html" title="58.4. Foreign Data Wrapper Query Planning">Prev</a> </td><td width="10%" align="left"><a accesskey="u" href="fdwhandler.html" title="Chapter 58. Writing a Foreign Data Wrapper">Up</a></td><th width="60%" align="center">Chapter 58. Writing a Foreign Data Wrapper</th><td width="10%" align="right"><a accesskey="h" href="index.html" title="PostgreSQL 18.0 Documentation">Home</a></td><td width="10%" align="right"> <a accesskey="n" href="tablesample-method.html" title="Chapter 59. Writing a Table Sampling Method">Next</a></td></tr></table><hr /></div><div class="sect1" id="FDW-ROW-LOCKING"><div class="titlepage"><div><div><h2 class="title" style="clear: both">58.5. Row Locking in Foreign Data Wrappers <a href="#FDW-ROW-LOCKING" class="id_link">#</a></h2></div></div></div><p>
3 If an FDW's underlying storage mechanism has a concept of locking
4 individual rows to prevent concurrent updates of those rows, it is
5 usually worthwhile for the FDW to perform row-level locking with as
6 close an approximation as practical to the semantics used in
7 ordinary <span class="productname">PostgreSQL</span> tables. There are multiple
8 considerations involved in this.
10 One key decision to be made is whether to perform <em class="firstterm">early
11 locking</em> or <em class="firstterm">late locking</em>. In early locking, a row is
12 locked when it is first retrieved from the underlying store, while in
13 late locking, the row is locked only when it is known that it needs to
14 be locked. (The difference arises because some rows may be discarded by
15 locally-checked restriction or join conditions.) Early locking is much
16 simpler and avoids extra round trips to a remote store, but it can cause
17 locking of rows that need not have been locked, resulting in reduced
18 concurrency or even unexpected deadlocks. Also, late locking is only
19 possible if the row to be locked can be uniquely re-identified later.
20 Preferably the row identifier should identify a specific version of the
21 row, as <span class="productname">PostgreSQL</span> TIDs do.
23 By default, <span class="productname">PostgreSQL</span> ignores locking considerations
24 when interfacing to FDWs, but an FDW can perform early locking without
25 any explicit support from the core code. The API functions described
26 in <a class="xref" href="fdw-callbacks.html#FDW-CALLBACKS-ROW-LOCKING" title="58.2.6. FDW Routines for Row Locking">Section 58.2.6</a>, which were added
27 in <span class="productname">PostgreSQL</span> 9.5, allow an FDW to use late locking if
30 An additional consideration is that in <code class="literal">READ COMMITTED</code>
31 isolation mode, <span class="productname">PostgreSQL</span> may need to re-check
32 restriction and join conditions against an updated version of some
33 target tuple. Rechecking join conditions requires re-obtaining copies
34 of the non-target rows that were previously joined to the target tuple.
35 When working with standard <span class="productname">PostgreSQL</span> tables, this is
36 done by including the TIDs of the non-target tables in the column list
37 projected through the join, and then re-fetching non-target rows when
38 required. This approach keeps the join data set compact, but it
39 requires inexpensive re-fetch capability, as well as a TID that can
40 uniquely identify the row version to be re-fetched. By default,
41 therefore, the approach used with foreign tables is to include a copy of
42 the entire row fetched from a foreign table in the column list projected
43 through the join. This puts no special demands on the FDW but can
44 result in reduced performance of merge and hash joins. An FDW that is
45 capable of meeting the re-fetch requirements can choose to do it the
48 For an <code class="command">UPDATE</code> or <code class="command">DELETE</code> on a foreign table, it
49 is recommended that the <code class="literal">ForeignScan</code> operation on the target
50 table perform early locking on the rows that it fetches, perhaps via the
51 equivalent of <code class="command">SELECT FOR UPDATE</code>. An FDW can detect whether
52 a table is an <code class="command">UPDATE</code>/<code class="command">DELETE</code> target at plan time
53 by comparing its relid to <code class="literal">root->parse->resultRelation</code>,
54 or at execution time by using <code class="function">ExecRelationIsTargetRelation()</code>.
55 An alternative possibility is to perform late locking within the
56 <code class="function">ExecForeignUpdate</code> or <code class="function">ExecForeignDelete</code>
57 callback, but no special support is provided for this.
59 For foreign tables that are specified to be locked by a <code class="command">SELECT
60 FOR UPDATE/SHARE</code> command, the <code class="literal">ForeignScan</code> operation can
61 again perform early locking by fetching tuples with the equivalent
62 of <code class="command">SELECT FOR UPDATE/SHARE</code>. To perform late locking
63 instead, provide the callback functions defined
64 in <a class="xref" href="fdw-callbacks.html#FDW-CALLBACKS-ROW-LOCKING" title="58.2.6. FDW Routines for Row Locking">Section 58.2.6</a>.
65 In <code class="function">GetForeignRowMarkType</code>, select rowmark option
66 <code class="literal">ROW_MARK_EXCLUSIVE</code>, <code class="literal">ROW_MARK_NOKEYEXCLUSIVE</code>,
67 <code class="literal">ROW_MARK_SHARE</code>, or <code class="literal">ROW_MARK_KEYSHARE</code> depending
68 on the requested lock strength. (The core code will act the same
69 regardless of which of these four options you choose.)
70 Elsewhere, you can detect whether a foreign table was specified to be
71 locked by this type of command by using <code class="function">get_plan_rowmark</code> at
72 plan time, or <code class="function">ExecFindRowMark</code> at execution time; you must
73 check not only whether a non-null rowmark struct is returned, but that
74 its <code class="structfield">strength</code> field is not <code class="literal">LCS_NONE</code>.
76 Lastly, for foreign tables that are used in an <code class="command">UPDATE</code>,
77 <code class="command">DELETE</code> or <code class="command">SELECT FOR UPDATE/SHARE</code> command but
78 are not specified to be row-locked, you can override the default choice
79 to copy entire rows by having <code class="function">GetForeignRowMarkType</code> select
80 option <code class="literal">ROW_MARK_REFERENCE</code> when it sees lock strength
81 <code class="literal">LCS_NONE</code>. This will cause <code class="function">RefetchForeignRow</code> to
82 be called with that value for <code class="structfield">markType</code>; it should then
83 re-fetch the row without acquiring any new lock. (If you have
84 a <code class="function">GetForeignRowMarkType</code> function but don't wish to re-fetch
85 unlocked rows, select option <code class="literal">ROW_MARK_COPY</code>
86 for <code class="literal">LCS_NONE</code>.)
88 See <code class="filename">src/include/nodes/lockoptions.h</code>, the comments
89 for <code class="type">RowMarkType</code> and <code class="type">PlanRowMark</code>
90 in <code class="filename">src/include/nodes/plannodes.h</code>, and the comments for
91 <code class="type">ExecRowMark</code> in <code class="filename">src/include/nodes/execnodes.h</code> for
92 additional information.
93 </p></div><div class="navfooter"><hr /><table width="100%" summary="Navigation footer"><tr><td width="40%" align="left"><a accesskey="p" href="fdw-planning.html" title="58.4. Foreign Data Wrapper Query Planning">Prev</a> </td><td width="20%" align="center"><a accesskey="u" href="fdwhandler.html" title="Chapter 58. Writing a Foreign Data Wrapper">Up</a></td><td width="40%" align="right"> <a accesskey="n" href="tablesample-method.html" title="Chapter 59. Writing a Table Sampling Method">Next</a></td></tr><tr><td width="40%" align="left" valign="top">58.4. Foreign Data Wrapper Query Planning </td><td width="20%" align="center"><a accesskey="h" href="index.html" title="PostgreSQL 18.0 Documentation">Home</a></td><td width="40%" align="right" valign="top"> Chapter 59. Writing a Table Sampling Method</td></tr></table></div></body></html>