Mar 2019

6 commits
Add "split after new tuple" nbtree optimization.
25 Mar 2019Peter Geogheganmaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 more
LLM analysis, 99% suremodified before commitlight changesfrom patchset #1141 of 1 reviewer saw the final version

Matches the complete split-after-new-tuple component of #114, with rewritten explanatory comments and equivalent usemult assignment cleanup in _bt_afternewitemoff(). The experimental removal of restrictions in #124 was posted after this commit.

  • Heikki Linnakangassaw the final versiontrailerReviewed the simplified split-after-new-tuple implementation in #59 and examined its fillfactor/contention tradeoff in #88, after which this component received only light cleanup.
Add nbtree high key "continuescan" optimization.
23 Mar 2019Peter Geogheganmaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 more
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1141 of 1 reviewer saw the final version

Matches the complete continuescan component of #114, including Heikki's _bt_readpage() restructuring from #101. Remaining differences are comment trimming and equivalent placement of the assertion and continuescan initialization.

  • Heikki Linnakangassaw the final versiontrailerReviewed #86 and supplied the replacement #101 that consolidates dead-tuple and boundary checks in _bt_readpage(), followed by light parameter-passing and editorial adjustments.
Allow amcheck to re-find tuples using new search.
20 Mar 2019Peter Geogheganmaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 more
LLM analysis, 100% surelanded as postedfrom patchset #1141 of 1 reviewer saw the final version

The complete amcheck rootdescend component of #114 has the same stable git patch-id as the commit, covering verify_nbtree.c, extension SQL/versioning, tests, and documentation.

  • Heikki Linnakangassaw the final versiontrailerExamined the exhaustive root-search check and its corruption coverage in #104, #106, and #110, with only naming and documentation adjustments before #114.
Consider secondary factors during nbtree splits.
20 Mar 2019Peter Geogheganmaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 more
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1140 of 1 reviewer saw the final version

Matches the complete secondary-split-factors component of #114; the remaining changes correct README wording and clarify a comment about the rightmost optimization in nbtsplitloc.c.

  • Heikki Linnakangassaw an earlier versiontrailerReviewed split selection in #45, supplied a prototype in #53, and reviewed #58 in #59, before the strategy and interval-edge logic changes in #61 and #86.
Make heap TID a tiebreaker nbtree index column.
20 Mar 2019Peter Geogheganmaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 more
LLM analysis, 100% suremodified before commitmoderate changesfrom patchset #1151 of 2 reviewers saw the final version

Uses the second replacement patch in #115, retaining Peter's authorship and Heikki's separate insertion loops. The committed nbtinsert.c corrects the heapkeyspace high-key comparison from != 0 to <= 0, as diagnosed in #116, alongside documentation cleanup and the agreement reported in #121.

  • Heikki Linnakangassaw the final versiontrailerReviewed tuple representation, compatibility, and pivot-search semantics, supplied the final insertion-loop revision #115, and reached agreement on the remaining corrections reported in #121.
  • Alexander Korotkovsaw an earlier versiontrailerQuestioned legacy duplicate handling and maximum pivot-tuple size in #49, before subsequent comparison, locking, and insertion-state revisions.
Refactor nbtree insertion scankeys.
20 Mar 2019Peter Geogheganmaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 more
LLM analysis, 100% suremodified before commitmoderate changesfrom patchset #1151 of 2 reviewers saw the final version

Uses the first replacement patch in #115, whose From header identifies Peter, retaining BTInsertStateData, _bt_binsrch_insert(), and _bt_stepright(). The commit restores the nbuf guard around cached-bound termination in _bt_check_unique(), adds an assertion, and revises comments and the commit message.

  • Heikki Linnakangassaw the final versiontrailerRequested the standalone refactoring, supplied the state-separation implementation #67 and final replacement #115, and settled remaining changes before commit as reported in #121.
  • Andrey Lepikhovsaw an earlier versiontrailerReported testing and interface concerns in #17, then reviewed BTScanInsert and proposed separating cached bounds and binary-search logic in #32, well before the final structural revisions.

Credits across 6 commits

Author
Peter Geoghegan6
Heikki Linnakangas2
Reviewer
Heikki Linnakangas6
Alexander Korotkov1
Andrei Lepikhov1
Committer
Peter Geoghegan6