Hi Seongjun, On Wed, Oct 28, 2026 at 10:00 PM 신성준 <[email protected]> wrote: > v9 keeps an outer wait event across a log write by using > pgstat_report_wait_start_nested() / pgstat_report_wait_end_nested() > at the new logging sites. That was the open point from Nik's last > review. > > What would help most now: > - Nik, a rerun of your harness against v9 when you have a free > moment. No hurry.
My AI harness tested v9 on master at b6178c942ea7, cassert build. The outer event survives the log write now. Sampled after ereport(LOG) returned, inside the extension's region, v9 shows active | Extension | HarnessOuterWait for stderr with logging_collector off and on, csvlog, jsonlog, all three together, and syslog. v8 shows an empty wait_event_type and wait_event in the same runs, so the check can tell them apart. The core regression suite passes with v9. I also read the wrapped sites. They wrap only leaf OS calls (write(), openlog(), syslog(), and the Windows calls in 0002), and nothing in between can ereport, so I see no path that leaves start without end. No new findings. I did not test Windows or the macro alternative Henson described. >From the tester's side I consider v9 ready. Thanks, Nik
