Description of problem: 2x2 distributed replicate volume with quota and profile enabled. 1 fuse client executing a multithreaded program. 1 nfs client executing fs-perf-test. While tests were going had done replace brick twice. The multithreaded application i(which spawns threads, with each thread opening the same file, and performing different fops on their respective fds such as read, write, fcntl locking etc) is hung. No server was brought down. Statedumps indicate that blocked locks are present. Version-Release number of selected component (if applicable): How reproducible: Steps to Reproduce: 1. 2. 3. Actual results: inodelk.inodelk[0](ACTIVE)=type=WRITE, start=9223372036854775806, len=0, pid=18446744072315678016, lk-owner=140250763177280, transport=0x23434 60, blocked at Wed Jan 4 03:26:26 2012 , granted at Wed Jan 4 03:26:26 2012 inodelk.inodelk[1](ACTIVE)=type=WRITE, start=0, len=131072, pid=18446744073709551615, lk-owner=140250764101896, transport=0x2343460, granted a t Wed Jan 4 03:26:26 2012 inodelk.inodelk[2](BLOCKED)=type=WRITE, start=0, len=0, pid=18446744073709551615, lk-owner=140250764101896, transport=0x2343460, blocked at Wed Jan 4 03:26:26 2012 Expected results: Additional info:
Johnny, Do you have the logs and statedumps of clients, bricks somewhere we can check?. Are you observing this issue on every run of this testcase?. Pranith
please update these bugs w.r.to 3.3.0qa27, need to work on it as per target milestone set.
CHANGE: http://review.gluster.com/3370 (protocol/server: del_locker should delete one locker per unlock) merged in master by Anand Avati (avati)
Bug has been reproduced and patch has been submitted for this.
Checked with glusterfs-3.3.0qa43. And did not find the blocked inodelks.