model-writtenrun 4274 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- The thread began with Andres Freund's announcement of smgrzeroextend(), FileZero(), and FileFallocate(), using allocation for larger extensions and explicit zeroing for small extensions or unsupported platforms #1.
- Christoph Berg reported intermittent PostgreSQL 16 pgbench failures caused by FileFallocate() returning an interrupted-system-call error on overlayfs backed by tmpfs and ext4 #2.
- Melanie Plageman identified the missing EINTR retry, and Andres broadened the fix to other susceptible fd.c operations, posting a prototype that used pg_* wrappers where available and introduced pg_ftruncate() #3 #4 #5.
- Christoph's builds passed with the prototype, while Thomas Munro judged simple retries appropriate for bounded file extensions despite starvation problems previously encountered with much larger shared-memory allocations #6 #7.
- Kyotaro Horiguchi advocated blocking signals to prevent repeated interruptions, but Andres defended retries because these extensions are bounded; signal masking was not adopted #10 #11 #12.
- Alvaro Herrera examined the consequences of leaving some retries outside pg_* wrappers, Christoph confirmed continued successful testing, and Andres chose to retain the minimal prototype approach #13 #14 #15.
- Andres posted the final patch with only comment and whitespace changes to the implementation and reported successful CI runs; Tom Lane confirmed there was still time to land it before beta2 #16 #17.
- Andres committed #16 unchanged as 0d369ac650, retaining retries rather than signal masking and declining backpatches because the reported allocation problem was new to PostgreSQL 16 and older paths lacked comparable reports.
- After Amit Kapila asked whether the committed fix resolved the open item, Andres confirmed that he had closed it, leaving no outstanding patch work identified in the thread #19 #20.