summaryrefslogtreecommitdiffstats
path: root/xlators/cluster/afr
diff options
context:
space:
mode:
authorRaghavendra G <rgowdapp@redhat.com>2019-11-20 22:31:48 +0530
committerXavi Hernandez <xhernandez@redhat.com>2019-12-19 12:41:14 +0000
commit3389bc58ea9773a66ed4bc7bd96b93dc1c97e4cc (patch)
treecd0deb6568b8e9ea692cd0f9b3f14ebc2582733e /xlators/cluster/afr
parentd8adc66b79de3f21672884539a65327614e0c6bb (diff)
tests/fuse/bug-965974.t: turn off md-cache, to fix failure of the test
Imagine the following set of operations: 1. touch $M0/file 2. ln $M0/file $M0/file.lnk 3. rm $M1/file 4. wait for md-cache-timeout 5. stat $M0/file.lnk $M0/file stat on $M0/file in step 5 succeeds when md-cache-timeout is non-zero even though it was removed from $M1. The reason is 1. fuse-bridge on $M0 would resolve both file and file.lnk to same inode. This inode i1 is cached in md-cache. 2. After md-cache-timeout lookup on $M0/file.lnk would be sent to backend. This lookup will be successful as file.lnk is present on backend and would refresh the md-cache. 3. The lookup on $M0/file sent within md-cache-timeout after lookup on $M0/file.lnk would be sent with i1. Since i1 was refreshed by lookup on $M0/file.lnk, the cache is deemed valid and md-cache responds lookup on $M0/file as success To fix this failure we can either: 1. remove lookup on $M0/file.lnk, so that lookup on $M0/file is actually sent to backend. 2. turn off md-cache This patch chooses option 2. credits: Csaba Henk <csaba@redhat.com> Change-Id: I352c2acd377fe10c4bdf3b6e53c1de86a4e544c7 Signed-off-by: Raghavendra G <rgowdapp@redhat.com> Updates: bz#1756900
Diffstat (limited to 'xlators/cluster/afr')
0 files changed, 0 insertions, 0 deletions