[Bugs] [Bug 1187296] No way to gracefully rotate the libgfapi Samba vfs_glusterfs logfile.

[bugzilla at redhat.com]

https://bugzilla.redhat.com/show_bug.cgi?id=1187296

Jeff Byers <jbyers at stonefly.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
              Flags|needinfo?(jbyers at stonefly.c |
                   |om)                         |



--- Comment #4 from Jeff Byers <jbyers at stonefly.com> ---
(In reply to Shyamsundar from comment #3)
> Current situation:
> The regular gluster deamons, use SIGHUP to get notified on log rotations, so
> when the logrotate cron job runs, it notifies glusterfs(d), glusterd, et.
> al. via a HUP to close the old fd/FILE and open a new one. logrotate in the
> background has already renamed the file to its backup name etc.
> 
> gfapi cannot override HUP as it is being used by another service and is not
> a deamon buy itself. As a result, there is no way currently to notify gfapi
> that a log rotation has been performed.
> 
> Here are the options for gfapi consumers:
> 
> 1) Add a logrotate entry for the log file that is configured via SAMBA (or
> any other service that uses gfapi) with the copytruncate flag in the rotate
> command.
> 
> copytruncate, will let the deamon operate on the same fd/FILE whereas copy
> the old contents and truncate the original file, as a result we achieve
> rotation.
> 
> @Jeff, possibly even academic, but could you try this option? If not, just
> remove the NEEDINFO flag against yourself.
> 
> 2) Provide an API that the consumer of gfapi can call, to gracefully rotate
> logs, and consumer would need to handle when this API is called (say in
> their HUP signal handler).
> 
> This requires changes (as stated by Niels) in gfapi and its consumers.

I did not use logrotate, but performed the test manually by hand. The
logrodate copytruncate option (or the functional equivalent) should
work fine. This implies that there is no logfile offset knowledge, or
seeking within the logfile, other than to the EOF, which seems likely.
Thanks.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
Unsubscribe from this bug https://bugzilla.redhat.com/token.cgi?t=QM6SBRcmHS&a=cc_unsubscribe