Bug #78405 [PATCH]: FPM with keepalive: Kills worker when no request is seen for $terminate_timeout

Edit report at https://bugs.php.net/bug.php?id=78405&edit=1 ID: 78405 Patch added by: sjon@php.net Reported by: sjon@php.net Summary: FPM with keepalive: Kills worker when no request is seen for $terminate_timeout Status: Verified Type: Bug Package: FPM related Operating System: archLinux PHP Version: 8.1.3 Block user comment: N Private report: N New Comment: The following pull request has been associated: Patch Name: Fix #78405: FPM with keepalive kills workers after $terminate_timeout On GitHub: https://github.com/php/php-src/pull/8163 Patch: https://github.com/php/php-src/pull/8163.patch Previous Comments: ------------------------------------------------------------------------ [2022-02-28 10:46:21] cmb@php.net @sjon, if you find some time, please submit a pull request, since patches are often overlooked. ------------------------------------------------------------------------ [2022-02-25 14:53:55] sjon@php.net The following patch has been added/updated: Patch Name: only-terminate-workers-that-are-actually-running-code Revision: 1645800835 URL: https://bugs.php.net/patch-display.php?bug=78405&patch=only-terminate-workers-that-are-actually-running-code&revision=1645800835 ------------------------------------------------------------------------ [2022-02-25 14:33:32] sjon@php.net I found #69860 which clarified some things. The fact that the timestamps in the logs differ (execution timed out 5.xx sec vs SIGTERM after xxx sec) hint at a different issue: * an fpm worker is spawned, and handles a req, remains running due to keepalive * every time a new request is send over this connection, proc.accepted is reSET * if no request is sent for $terminate_timeout, fpm kills the worker while it is arguable workers also need some form of timeout to prevent exhaustion, I'm not sure the same $terminate_timeout should be applied here. While 5s is a reasonable value to kill a hanging script (for us) it is not reasonable to expect at least one request for every worker at least every $terminate_timeout. Imo a different timeout should apply ------------------------------------------------------------------------ [2022-02-25 09:16:47] sjon@php.net still an issue with latest php, and it's definitely caused by fastcgi_keep_conn being enabled ------------------------------------------------------------------------ [2019-09-24 08:59:43] sjon@php.net This might be "caused" by fastcgi_keep_conn being enabled. Possibly fpm keeps the child running due to keepalive being enabled but the master doesn't realize this and kills the child thinking it is timing out. Should this work or should this become a feature request? See also https://stackoverflow.com/a/51457613/3749523 ------------------------------------------------------------------------ The remainder of the comments for this report are too long. To view the rest of the comments, please view the bug report online at https://bugs.php.net/bug.php?id=78405 -- Edit this bug report at https://bugs.php.net/bug.php?id=78405&edit=1