Back to bug 1006809

Who When What Removed Added
Krutika Dhananjay 2014-01-16 09:54:16 UTC Status NEW ASSIGNED
CC kdhananj
Assignee rhs-bugs kdhananj
Krutika Dhananjay 2014-02-05 08:44:00 UTC Status ASSIGNED POST
Jay Turner 2014-02-07 12:16:22 UTC Target Milestone --- RHS 3.0
Nagaprasad Sathyanarayana 2014-05-06 10:56:30 UTC Priority unspecified high
Nagaprasad Sathyanarayana 2014-05-07 17:14:19 UTC CC nsathyan
Krutika Dhananjay 2014-05-23 09:06:10 UTC Status POST ON_QA
Fixed In Version glusterfs-3.6.0.3-1.el6rhs
shylesh 2014-06-20 06:15:06 UTC CC ssamanta
QA Contact sdharane shmohan
Status ON_QA ASSIGNED
Raghavendra G 2014-06-23 10:32:05 UTC CC rgowdapp
Raghavendra G 2014-06-23 12:57:51 UTC Status ASSIGNED MODIFIED
Vivek Agarwal 2014-06-24 06:42:18 UTC Status MODIFIED ON_QA
CC vagarwal
Fixed In Version glusterfs-3.6.0.3-1.el6rhs glusterfs-3.6.0.22-1
Krutika Dhananjay 2014-06-24 11:07:38 UTC Doc Text Earlier, the cases of a) file/directory entry not being present, and b) parent directory not being present (say on a newly added brick) used to both fail with ENOENT error. This was causing the newly added brick to be considered for layout assignments. And subsequently when a file that is about to be created hashes to the new brick, it would fail with ENOENT at the application level.

The fix involves treating parent being absent on a sub-volume (in this case because the directory hierarchy is yet to be constructed on the newly added brick) as ESTALE error (as opposed to ENOENT). As a result, the newly added brick is not considered for layout assignment, thereby fixing 'No such file or directory' failures on the mount point on subsequent directory entry creations.
Raghavendra G 2014-06-25 16:39:59 UTC Doc Text Earlier, the cases of a) file/directory entry not being present, and b) parent directory not being present (say on a newly added brick) used to both fail with ENOENT error. This was causing the newly added brick to be considered for layout assignments. And subsequently when a file that is about to be created hashes to the new brick, it would fail with ENOENT at the application level.

The fix involves treating parent being absent on a sub-volume (in this case because the directory hierarchy is yet to be constructed on the newly added brick) as ESTALE error (as opposed to ENOENT). As a result, the newly added brick is not considered for layout assignment, thereby fixing 'No such file or directory' failures on the mount point on subsequent directory entry creations. 		Earlier, mkdir failures because of parents not being present returned ENOENT. DHT-selfheal considers a brick which returned ENOENT on lookup as part of layout assuming that the lookup might be racing with mkdir. Hence, the newly added brick would be considered for layout assignments. However, the directory creation itself might've been failed because of parents not being present on new brick and subsequently when a file that is about to be created hashes to the new brick, it would fail with ENOENT, which is propagated back to application.

The fix involves treating parent being absent on a sub-volume (in this case because the directory hierarchy is yet to be constructed on the newly added brick) as ESTALE error (as opposed to ENOENT). As a result, the newly added brick is not considered for layout assignment, thereby fixing 'No such file or directory' failures on the mount point on subsequent directory entry creations.
shylesh 2014-08-21 12:39:22 UTC Status ON_QA VERIFIED
errata-xmlrpc 2014-09-22 17:55:43 UTC Status VERIFIED RELEASE_PENDING
errata-xmlrpc 2014-09-22 19:28:57 UTC Status RELEASE_PENDING CLOSED
Resolution --- ERRATA
Last Closed 2014-09-22 15:28:57 UTC
Rejy M Cyriac 2015-05-13 17:01:10 UTC Target Release --- RHGS 3.0.0
Target Milestone RHS 3.0.0 ---

Back to bug 1006809