model-writtenrun 9516 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Apr - Jul 2018 · 16 entries
- David Rowley opened the thread because PostgreSQL 11 lacked a way to disable SELECT partition pruning, asking whether planning and execution pruning needed separate controls and how prepared plans should behave #1.
- Justin Pryzby proposed extending constraint_exclusion, while David Rowley opposed overloading a setting named for CHECK constraints and Ashutosh Bapat supported a single switch for debugging and bug workarounds #2 #4 #6.
- Amit Langote argued that the new switch should control declarative partition pruning for every command, including UPDATE and DELETE, while preserving constraint_exclusion for traditional inheritance #9 #11 #13.
- David's first implementation disabled both planning-time pruning and generation of execution-time pruning information, but Amit's testing showed that constraint_exclusion still incorrectly controlled UPDATE pruning #15 #18.
- David introduced InheritanceKind to distinguish declarative partitioning from inheritance, added regression tests and configuration coverage, and received Amit's review of the revised implementation #22 #23.
- Álvaro Herrera committed the revised GUC patch as 055fb8d33d, with equivalent control-flow cleanup, extra tests, and documentation edits, leaving the broader partitioning chapter for follow-up #24.
- David then drafted separate documentation explaining pruning, its execution phases, and its relationship to constraint exclusion; documentation of other partitioning optimizations was left outside this patch #30 #31.
- Robert Haas questioned whether the GUC justified its permanent cost, David measured roughly a 1.1 percent improvement in a workload that could not benefit from pruning, and Álvaro emphasized its value as a workaround for possible bugs #32 #34 #37.
- Reviews from Amit Langote, David G. Johnston, and Justin Pryzby refined the pruning prose across several revisions, while David retained descriptions of initialization and unsupported execution cases because users needed to understand those limitations #36 #38 #45 #47 #49.
- Amit removed an obsolete constraint-exclusion caveat, but Álvaro's subsequent restructuring delta was based on David's preceding version and moved pruning ahead of constraint exclusion #51 #52.
- Robert explained that constraint exclusion remained useful even for declaratively partitioned and unpartitioned tables, persuading Álvaro to retain it as a separate section rather than a subsection of pruning #55 #56.
- The discussion retained the established partition-pruning name, and Álvaro continued to favor keeping the GUC until experience demonstrated that it was unnecessary #58 #59 #62.
- After Amit reviewed the rendered rewrite, Álvaro committed bebc46931a from the documentation series and his restructuring delta, incorporating corrections and dropping incorrect execution-time examples #68 #70.
- Robert, David, and Amit explored clearer EXPLAIN annotations for pruning phases and the partitioned relation, but that discussion produced no implementation in the supplied patchsets #74 #75 #76.
- Justin supplied another round of documentation corrections as an inline patch, Amit reviewed its wording and terminology, and Álvaro committed the refined result as 12b9affb32 #80 #83 #84 #88.
- Justin's subsequent terminology cleanup was reported committed by Peter Eisentraut and then backpatched to PostgreSQL 11 by Álvaro #90 #91 #95.
Mar - Apr 2019 · 5 entries
- David revived the thread in March 2019 after a report of excessive DELETE planning memory with 4,000 partitions, noting that the documentation rewrite had narrowed the large-hierarchy warning to inheritance #96.
- The warning discussion distinguished pruning costs from quadratic planner memory growth, and David attached a new warning using wording developed with Justin, Amit, and Robert #99 #103 #105 #106.
- After development-branch planner improvements, David revised the warning to distinguish inheritance from declarative partitioning and retained the original version for older branches; related UPDATE/DELETE documentation changes were deliberately kept in another thread #110 #112 #113.
- Justin revised the warning and added a second patch about features supported by declarative partitioning, but David questioned removing promises of future inheritance improvements and requested care with older-version wording #114 #115.
- The thread ends with the warning series still under discussion, including how to describe UPDATE/DELETE memory consumption, and with no recorded landing for either final patch #116.