Project

General

Profile

Feature #10498

Updated by Greg Farnum about 9 years ago

The ObjectCacher can block write calls if the dirty data limits are exceeded by sleeping on a cond. Unfortunately, we have seen cases where a single write call blocks for hours while others continue to go through.

Josh thinks the way to fix this is by introducing ordered wakeups when we hit this case, which makes sense to me. Strict ordering isn't great but should be sufficient.

Previously:
http://qa-proxy.ceph.com/teuthology/teuthology-2015-01-04_23:04:01-fs-next-testing-basic-multi/686605/

There's nothing apparent there, it starts up and does some writes and then the client log ends, and eventually teuthology times out the run. I'm not sure if this is a real bug or some kind of unrelated hangup, but I'm archiving it for posterity.

Back