model-writtenrun 8905 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
May 2014 · 4 entries
- Andres Freund proposed a SQL view exposing shared-memory allocations to diagnose core and extension memory use in #1, then replaced an accidentally attached psql patch with the actual implementation in #3.
- Andres also proposed naming dynamic shared-memory segments in #2; Robert Haas questioned the overhead and usefulness in #5, while Tom Lane argued that adding the parameter early would avoid a later API break in #6.
- Robert argued against adding an unused parameter without working monitoring support in #11, and Andres supplied a two-patch series implementing DSM names and both allocation views, together with missing locking and anonymous-memory accounting, in #17.
- Robert proposed dropping the static prefix and allocated boolean in favor of a named anonymous-memory row in #19; Andres incorporated that design, fixes, and documentation in #27.
Jul - Sep 2014 · 3 entries
- Michael Paquier reviewed DSM validation and the view implementation in #32 and #33, while Marti Raudsepp questioned exposing offsets in #34; Andres defended offsets for pointer diagnosis and free-memory reporting for capacity diagnosis in #35 and #36.
- Security and packaging discussions considered superuser restrictions, an extension, and a debugger-only interface: Tom favored restricted access in #39, Robert withdrew his address-disclosure objection after recognizing that the view reported relative offsets in #49, and Andres defended SQL access for administrators in #51.
- The proposal was returned with feedback after no revision arrived in #53; Andres revived it with a rebase in #54, and Michael explicitly requested revoking public access to both views and functions in #55.
Nov 2019 - Jan 2020 · 6 entries
- Robert revived the work again in #56, retaining main-segment reporting, implementing access restrictions, and deferring DSM support because it needed further design discussion.
- Tom and Andres identified documentation, volatility, and code-cleanup issues in #57 and #58; Robert incorporated those changes in #61 while declining redundant explanatory comments.
- Álvaro Herrera found that alignment padding was incorrectly reported as anonymous memory in #62, and Tom suggested separate overhead accounting in #64; Robert instead added an allocated_size column and explicit padded-size tracking in #65.
- Kyotaro Horiguchi questioned size terminology and anonymous hash allocations in #66; Robert explained why reporting existing allocations should remain separate from reorganizing them in #67, and Kyotaro accepted the patch in #68.
- Robert committed #65 as ed10f32e37 with only a small error-string formatting edit beyond rebasing, announcing completion in #69.
- The thread ended with Andres welcoming the commit in #70; main-segment reporting had landed, while DSM reporting remained deferred under #56 and allocation naming improvements remained follow-up work proposed in #65.