model-writtenrun 7737 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Amit Kapila opened the discussion in #1 because parallel INSERT and COPY proposals inspected every target partition for parallel safety, adding planning overhead even when execution ultimately remained serial.
- Ashutosh Bapat proposed partition pruning, delaying checks until enough rows justified parallelism, and assigning partitions to workers in #2; Amit kept pruning outside this work and explained the wasted-path-generation tradeoff of delayed checking in #3.
- After Amit Langote questioned checking target expressions when only the SELECT ran in workers, Amit Kapila explained in #7 that the leader also executes inside parallel mode, and in #11 described why discovering unsafe partitions during execution would require problematic restarts.
- Takayuki Tsunakawa proposed explicit user controls, delayed checks, or cataloged safety information in #5; discussion of conflicting source and target worker counts led Amit Kapila to favor a boolean DML table option while retaining source-side worker settings in #14.
- Hou and Amit Kapila agreed in #16, #17 and #18 that only the directly targeted table's option should govern an INSERT, ignoring child option values when inserting through a partitioned parent.
- Hou posted four patches in #19 implementing a session GUC and table option, both initially disabled, and clarified in #20 that they depended on the parallel INSERT series.
- Greg Nancarrow argued for enabling the table option by default and naming it parallel_dml_enabled in #21; Hou incorporated those changes and extracted preliminary eligibility checks in #22, after which Greg supplied corrected comments and documentation in #23.
- Hou introduced separate storage for partitioned-table options in #25, then incorporated Zhihong Yu's relation-closing cleanup from #24 while rebasing in #26.
- Huang Qiuyan identified a misleading comment about the enabled-by-default table option in #28; Hou corrected it, added partitioned-table tests and rebased onto the evolving underlying implementation in #29.
- The thread ends with Hou requesting further review of the rebased series in #30, without a local commit announcement; the supplied record for thread #42973 identifies the later #471933 as the source of c8f78b6161, which landed INSERT-specific option names and an enabled-by-default GUC.