summaryrefslogtreecommitdiffstats
path: root/xlators/cluster
diff options
context:
space:
mode:
authorShehjar Tikoo <shehjart@gluster.com>2010-07-27 06:16:53 +0000
committerAnand V. Avati <avati@dev.gluster.com>2010-07-28 02:24:49 -0700
commit7ced5bf710bbbee3290845b7ee65b5c895c06e7d (patch)
treed72b5b844262838bb2190aa91efc4b96b22502d3 /xlators/cluster
parent045efc1f5d59d98fff28ea29039ead4015fe1e6b (diff)
replicate: Set setxattr wait-count to child-count to avoid race
A race is seen during a compilebench load against NFS where a NFSv3 MKDIR is immediately followed by a LOOKUP on the new directory. By the time this lookup comes in, the setxattr issued by distribute for the mkdir has not returned yet from all the replicas. Since replicate returns a reply for setxattr after receiving reply only from one subvol, the MKDIR is unwound from distribute. When the LOOKUP is received at distribute, the setxattr has still not completed at the second replica, whereas that is where the lookup is sent by replicate. Finding that the xattr is missing from the dir, distribute returns ESTALE to NFS, which return NFS3ERR_STALE to nfs client. This patch changes replicate setxattr behaviour by forcing replicate to wait for all children rather than one to call back before unwinding the stack. Signed-off-by: Shehjar Tikoo <shehjart@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 937 (compilebench fails on NFS+dist+repl+all perf xlators) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=937
Diffstat (limited to 'xlators/cluster')
-rw-r--r--xlators/cluster/afr/src/afr-inode-write.c2
1 files changed, 1 insertions, 1 deletions
diff --git a/xlators/cluster/afr/src/afr-inode-write.c b/xlators/cluster/afr/src/afr-inode-write.c
index 37909181e..eff01da86 100644
--- a/xlators/cluster/afr/src/afr-inode-write.c
+++ b/xlators/cluster/afr/src/afr-inode-write.c
@@ -1296,7 +1296,7 @@ afr_setxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
}
local->success_count++;
- if (local->success_count == priv->wait_count) {
+ if (local->success_count == priv->child_count) {
need_unwind = 1;
}
}