mirror of
https://github.com/neondatabase/neon.git
synced 2026-08-18 03:58:19 +00:00
84b7588293
Before this patch, we would leave the `timeline-calculate-logical-size-pause` failpoint in `pause` mode at the end of the test. With the switch to a single runtime, somehow we'd end up in a place where the pageserver was half shut down while the failpoint spawn_blocking thread was waiting for the `off` event that never arrived. Failures were reproducible quite well in CI: https://neon-github-public-dev.s3.amazonaws.com/reports/pr-6555/8396322235/index.html#suites/c39429f093f87547b2a3b0943e2522d9/4dacb1efb232b98/ I couldn't repro it locally. I managed to repro it once on an i3en.3xlarge , where I then attached gdb to capture the backtrace. For posterity: https://www.notion.so/neondatabase/debug-test_lazy_attach_activation-teardown-hang-as-part-of-PR-6555-421cb61dc45d4d4e90220c86567f50da?pvs=4