[Bugs] [Bug 1506487] glusterfs / glusterfsd processes may not start properly upon reboot/restart

[bugzilla at redhat.com]

https://bugzilla.redhat.com/show_bug.cgi?id=1506487

jack.wong at laserfiche.com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jack.wong at laserfiche.com



--- Comment #26 from jack.wong at laserfiche.com ---
(In reply to Milind Changire from comment #16)
> Sylvain,
> Could you check if adding/changing "option transport.listen-backlog 100" in
> /etc/glusterfs/glusterd.vol helps any bit ?

We have run into this issue too. Thank you for your suggestion. It was quite
helpful.

Tweaking "transport.listen-backlog" fixed the problem for us. One thing I want
to note is that 100 may be too low. We are running about 40 Gluster volumes on
a single server. We had to set "transport.listen-backlog" higher than 1024
before all our glusterfsd processes consistently started up. Because 1024 is
higher than the default net.core.somaxconn kernel configuration value of 128,
we also had to increase net.core.somaxconn to make that take effect. Otherwise,
the kernel would silently truncate the listen backlog to SOMAXCONN
(https://manpages.debian.org/stretch/manpages-dev/listen.2.en.html#NOTES).

We only had to edit /etc/glusterfs/glusterd.vol. We did not have to set the
transport.listen-backlog on any of our Gluster volumes.

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