https://bz.apache.org/bugzilla/show_bug.cgi?id=65402
--- Comment #28 from Stefan Eissing <ste...@eissing.org> --- I think what we are looking at is h2 worker starvation in this proxy scenario. When a long reading SSE request is RST_STREAM on the main connection, the h2 worker is in a read and only when that delivers something, it will detect that it has been aborted. That can take up to 30 seconds here. We can test this by raising `H2MaxWorkers` to some very high level and check if the stalls disappear. The real fix would be to have a immediate cancellation of ongoing requests, but I do not know how to do this. In http/1.1, where this problem does not seem to happen, do we not see it because there are just enough event workers or does mpm_event know how to disrupt an ongoing read from the backend? Florian: can you test if a "H2MaxWorkers 512" seemingly makes the problem disappear for a while? -- You are receiving this mail because: You are the assignee for the bug. --------------------------------------------------------------------- To unsubscribe, e-mail: bugs-unsubscr...@httpd.apache.org For additional commands, e-mail: bugs-h...@httpd.apache.org
