PHPC-1274: Do not allow forked children to affect parent resources #1059
Conversation
| @@ -0,0 +1,102 @@ | |||
| --TEST-- | |||
| PHPC-1274: Session destruct should not end session from parent process | |||
There was a problem hiding this comment.
This is currently not functioning as a regression test, as I'm having trouble getting the child process to actually kill the session when it exits if I disable the call to mongoc_client_reset() in the Session's free_object handler.
I think this may be because the session pool is really only destroyed during the call to mongoc_client_destroy(), and PHPC purposefully avoids calling that due to 600db56 (PHPC-912). If that's the case, perhaps we can't implement a regression test for this at all and will have to make do with the "does not abort" test.
There was a problem hiding this comment.
I'd like to leave this test in place despite my last comment, as it could become more relevant when we get to PHPC-1522. And even if we can't implement that (i.e. allow children to call mongoc_client_destroy() on clients created by parents), this may still be useful as our only test to demonstrate that child processes don't kill active sessions of a parent process.
jmikola
changed the title
jmikola
changed the title
jmikola
changed the title
jmikola
force-pushed
the
phpc-1274
branch
4 times, most recently
from
6d05ac6 to
58fc57c
Compare
|
Reset logic looks sound to me. LGTM once |
Check PID before creating/freeing a session or cursor to avoid interfering with resources owned by a parent process. Additionally, prohibit resetting a client multiple times to avoid cases where child resources may not be cleaned up.
https://jira.mongodb.org/browse/PHPC-1274
This PR will supersede both #941 and #1027