Closed erikdsjostrom closed 3 years ago
de-registering should probably be under a lock too right? I.e. if multiple pollers de-register at the same time.
de-registering should probably be under a lock too right? I.e. if multiple pollers de-register at the same time.
Yes, the only reason it's not under a lock is that I forgot to add it.
Rebased and added locking for deregistering.
Aren't we still missing something like fcntl.flock(file_descriptor, fcntl.LOCK_EX | fcntl.LOCK_NB)
when de-registering?
the new script is missing in Makefile.am
, so it's not included in the builds.
Correct again, the heat is getting to me.
For registering I think it's fine that we don't retry getting the lock. However for de-registering that doesn't work as the container will just exit/be deleted without having de-registered. auto_delete
will take care of removing the dead node eventually, but that is not ideal.
For registering I think it's fine that we don't retry getting the lock. However for de-registering that doesn't work as the container will just exit/be deleted without having de-registered.
auto_delete
will take care of removing the dead node eventually, but that is not ideal.
I agree, that would be an issue. But I think it's best to handle that on the slim-poller side, I'll start brewing a new commit.
Changes needed for slim-poller refactor.