[Bugs] [Bug 1665145] Writes on Gluster 5 volumes fail with EIO when "cluster.consistent-metadata" is set

[bugzilla at redhat.com]

https://bugzilla.redhat.com/show_bug.cgi?id=1665145

Ravishankar N <ravishankar at redhat.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |Triaged
             Status|NEW                         |ASSIGNED
                 CC|                            |ravishankar at redhat.com
           Assignee|bugs at gluster.org            |ravishankar at redhat.com



--- Comment #2 from Ravishankar N <ravishankar at redhat.com> ---

> $ echo foo > /mnt/gluster/a; ls -l /mnt/gluster/a
> bash: /mnt/gluster/a: Input/output error
> -rw-r--r--. 1 root root 0 10. Jan 15:53 /mnt/gluster/a

It is the truncate that is failing. This seems to be a regression introduced by
commit c9bde3021202f1d5c5a2d19ac05a510fc1f788ac (performance/readdir-ahead:
keep stats of cached dentries in sync with modifications). When you set
.consistent-metadata on, AFR zeroes out the ctime (commit
4c4624c9bad2edf27128cb122c64f15d7d63bbc8).  When ctime is zero, readdir-ahead
xlator seems to zero out the entire stat buf including the ia_type and ia_gfid.

Let me attempt a fix. As a workaround, if you want to enable
consistent-metadata, you can disable 'performance.readdir-ahead'. That should
solve the issue interim.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are the assignee for the bug.