2 58.5. Row Locking in Foreign Data Wrappers #
4 If an FDW's underlying storage mechanism has a concept of locking
5 individual rows to prevent concurrent updates of those rows, it is
6 usually worthwhile for the FDW to perform row-level locking with as
7 close an approximation as practical to the semantics used in ordinary
8 PostgreSQL tables. There are multiple considerations involved in this.
10 One key decision to be made is whether to perform early locking or late
11 locking. In early locking, a row is locked when it is first retrieved
12 from the underlying store, while in late locking, the row is locked
13 only when it is known that it needs to be locked. (The difference
14 arises because some rows may be discarded by locally-checked
15 restriction or join conditions.) Early locking is much simpler and
16 avoids extra round trips to a remote store, but it can cause locking of
17 rows that need not have been locked, resulting in reduced concurrency
18 or even unexpected deadlocks. Also, late locking is only possible if
19 the row to be locked can be uniquely re-identified later. Preferably
20 the row identifier should identify a specific version of the row, as
23 By default, PostgreSQL ignores locking considerations when interfacing
24 to FDWs, but an FDW can perform early locking without any explicit
25 support from the core code. The API functions described in
26 Section 58.2.6, which were added in PostgreSQL 9.5, allow an FDW to use
27 late locking if it wishes.
29 An additional consideration is that in READ COMMITTED isolation mode,
30 PostgreSQL may need to re-check restriction and join conditions against
31 an updated version of some target tuple. Rechecking join conditions
32 requires re-obtaining copies of the non-target rows that were
33 previously joined to the target tuple. When working with standard
34 PostgreSQL tables, this is done by including the TIDs of the non-target
35 tables in the column list projected through the join, and then
36 re-fetching non-target rows when required. This approach keeps the join
37 data set compact, but it requires inexpensive re-fetch capability, as
38 well as a TID that can uniquely identify the row version to be
39 re-fetched. By default, therefore, the approach used with foreign
40 tables is to include a copy of the entire row fetched from a foreign
41 table in the column list projected through the join. This puts no
42 special demands on the FDW but can result in reduced performance of
43 merge and hash joins. An FDW that is capable of meeting the re-fetch
44 requirements can choose to do it the first way.
46 For an UPDATE or DELETE on a foreign table, it is recommended that the
47 ForeignScan operation on the target table perform early locking on the
48 rows that it fetches, perhaps via the equivalent of SELECT FOR UPDATE.
49 An FDW can detect whether a table is an UPDATE/DELETE target at plan
50 time by comparing its relid to root->parse->resultRelation, or at
51 execution time by using ExecRelationIsTargetRelation(). An alternative
52 possibility is to perform late locking within the ExecForeignUpdate or
53 ExecForeignDelete callback, but no special support is provided for
56 For foreign tables that are specified to be locked by a SELECT FOR
57 UPDATE/SHARE command, the ForeignScan operation can again perform early
58 locking by fetching tuples with the equivalent of SELECT FOR
59 UPDATE/SHARE. To perform late locking instead, provide the callback
60 functions defined in Section 58.2.6. In GetForeignRowMarkType, select
61 rowmark option ROW_MARK_EXCLUSIVE, ROW_MARK_NOKEYEXCLUSIVE,
62 ROW_MARK_SHARE, or ROW_MARK_KEYSHARE depending on the requested lock
63 strength. (The core code will act the same regardless of which of these
64 four options you choose.) Elsewhere, you can detect whether a foreign
65 table was specified to be locked by this type of command by using
66 get_plan_rowmark at plan time, or ExecFindRowMark at execution time;
67 you must check not only whether a non-null rowmark struct is returned,
68 but that its strength field is not LCS_NONE.
70 Lastly, for foreign tables that are used in an UPDATE, DELETE or SELECT
71 FOR UPDATE/SHARE command but are not specified to be row-locked, you
72 can override the default choice to copy entire rows by having
73 GetForeignRowMarkType select option ROW_MARK_REFERENCE when it sees
74 lock strength LCS_NONE. This will cause RefetchForeignRow to be called
75 with that value for markType; it should then re-fetch the row without
76 acquiring any new lock. (If you have a GetForeignRowMarkType function
77 but don't wish to re-fetch unlocked rows, select option ROW_MARK_COPY
80 See src/include/nodes/lockoptions.h, the comments for RowMarkType and
81 PlanRowMark in src/include/nodes/plannodes.h, and the comments for
82 ExecRowMark in src/include/nodes/execnodes.h for additional