There is a particular pattern of NFS requests generated by compilebench that lead nfs translator into a deadlock on an inode lock. It happens on a simple posix configuration while running two compilebench instances from two separate nfs clients. Following message is a sign of such a deadlock where writes do not proceed beyond the fh resolution stage: [2010-06-21 07:55:46] D [nfs3-helpers.c:2274:nfs3_log_rw_call] nfs-nfsv3: XID: ae1a41d0, WRITE: args: FH: hashcount 4, xlid 0, gen 5485193550805074074, ino 36323483, offset: 0, count: 65536, UNSTABLE [2010-06-21 07:55:46] T [nfs3.c:1836:nfs3_write] nfs-nfsv3: FH to Volume: posix [2010-06-21 07:55:46] T [nfs3-helpers.c:2899:nfs3_fh_resolve_inode] nfs-nfsv3: FH needs inode resolution [2010-06-21 07:55:46] T [nfs3-helpers.c:2403:nfs3_fh_resolve_inode_done] nfs-nfsv3: FH inode resolved [2010-06-21 07:55:46] T [nfs3-helpers.c:2113:nfs3_file_open_and_resume] nfs-nfsv3: Opening: /c2/native-0/Documentation/devices.txt ########################################################## ##### Hung for around 3mins, before killed ############### ########################################################## [2010-06-21 07:58:38] W [glusterfsd.c:1009:cleanup_and_exit] glusterfs: shutting down Complete log attached.
PATCH: http://patches.gluster.com/patch/3524 in master (nfs3: Fix race updating op queue on uncached fd open)
Regression test description added. http://test.gluster.com/show_bug.cgi?id=78