Mar - Apr 2021

5 commits
Add Result Cache executor node (take 2)
2 Apr 2021David Rowleymaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1Also in: What to call an executor node which lazily caches tuples in a hash table?
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1031 of 4 reviewers saw the final version

The commit matches #103 with formatting, Hou's T_ResultCachePath/optimizer README corrections, and disabled caching in unstable partition_prune tests, confirmed by #105. The other linked thread, thread #43920, concerns naming and has no patchsets.

  • Andy Fansaw an earlier versiontrailerAndy tested correctness and performance, supplied an executor optimization, and reviewed cache handling and planner overhead in #6, #55, #66, #71, and #82, before substantial planner changes.
  • Justin Pryzbysaw an earlier versiontrailerJustin rebased and reviewed assertions, documentation, flags, and costing in #78 and #83, before volatility and hash-operator caching changed the planner.
  • Zhihong Yusaw an earlier versiontrailerZhihong's reviews prompted cache cleanup and EXPLAIN changes acknowledged in #74, followed by code reviews in #99 and #101, before the additional planner caching.
  • Hou Zhijiesaw the final versiontrailerHou identified the incorrect path-node tag and missing optimizer README entry in #104, both incorporated into the second landing.
Add Result Cache executor node
31 Mar 2021David Rowleymaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1Also in: What to call an executor node which lazily caches tuples in a hash table?
LLM analysis, 100% suremodified before commitmoderate changesfrom patchset #1000 of 3 reviewers saw the final version

Relative to #100, the commit adds RestrictInfo.hasheqoperator and check_resultcacheable() in initsplan.c and removes redundant lateral-variable checks, as described in #102. The closer attachment #103 postdates this commit; thread #43920 supplies naming discussion only.

  • Andy Fansaw an earlier versiontrailerAndy tested and reviewed earlier cache implementations in #6, #55, #66, #71, and #82, before the final planner optimizations.
  • Justin Pryzbysaw an earlier versiontrailerJustin reviewed and rebased the series in #78 and examined planner costs in #83, before the later volatility and hash-operator caches.
  • Zhihong Yusaw an earlier versiontrailerZhihong reviewed cache and EXPLAIN code as acknowledged in #74 and directly in #99 and #101, after which David reworked the planner logic.
Allow estimate_num_groups() to pass back further details about the estimation
30 Mar 2021David Rowleymaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #981 of 1 reviewer saw the final version

The diff exactly matches the estimate_num_groups() prerequisite within #98, including EstimationInfo, SELFLAG_USED_DEFAULT, and caller updates; the other three patches are excluded.

  • Justin Pryzbysaw the final versionLLM analysis, 98% sureJustin questioned the flags type in #78, leading to the uint32 change accepted in #79, with only light changes to this prerequisite afterward.
Allow users of simplehash.h to perform direct deletions
30 Mar 2021David Rowleymaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #980 of 1 reviewer saw the final version

The diff exactly matches the simplehash.h prerequisite within #98, adding SH_DELETE_ITEM while retaining SH_DELETE; the rest of the series was committed separately or omitted.

  • Andres Freundsaw an earlier versiontrailerAndres examined pointer invalidation, inlining, and generated-code costs in #17, before #26 changed the implementation to preserve SH_DELETE separately.
Cache if PathTarget and RestrictInfos contain volatile functions
29 Mar 2021David Rowleymaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #970 of 1 reviewer saw the final version

The commit takes the volatility-caching prerequisite from #97, adding explanatory comments and formatting in clauses.c while preserving its lazy tri-state implementation; #98 confirms this prerequisite was pushed.

  • Tom Lanesaw an earlier versionLLM analysis, 100% sureTom reviewed the eager implementation in #94 and requested lazy caching inside contain_volatile_functions(), prompting the substantive rewrite in #96.

Credits across 5 commits

Author
David Rowley5
Reviewer
Andy Fan2
Justin Pryzby2
Zhihong Yu2
Andres Freund1
Hou, Zhijie1
Tested by
Konstantin Knizhnik2
Committer
David Rowley5