[Gluster-users] nfs and unexpected automounter behaviour

[John Sellens]

Hi - new to gluster, having an unexpected issue with NFS mounts via
the automounter.

I've noticed two things
- I'm used to the automounter unmounting inactive mounts
  after a while - the gluster NFS mounts seem to hang around indefinitely
- "amq -m" showws the gluster hosts as down, and non-gluster hosts up

For a short time after a gluster FS is NFS mounted, "amq -m" shows the
host as starting, and then it changes to showing as down:

myclient# amq -m
"root"                                                                    root    1   localhost is up
/etc/amd.map                             /host                            toplvl  1   localhost is up
/etc/amd.map                             /net                             toplvl  1   localhost is up
hpb:/net/hpb                             /.amd_mnt/hpb/host               host    1   hpb.mydomain.com is down
hpa:/net/hpa                             /.amd_mnt/hpa/host               host    1   hpa.mydomain.com is starting
myclient# amq -m
"root"                                                                    root    1   localhost is up
/etc/amd.map                             /host                            toplvl  1   localhost is up
/etc/amd.map                             /net                             toplvl  1   localhost is up
hpb:/net/hpb                             /.amd_mnt/hpb/host               host    1   hpb.mydomain.com is down
hpa:/net/hpa                             /.amd_mnt/hpa/host               host    1   hpa.mydomain.com is down

I had a quick look in the amq source but I couldn't get a good idea
of what "down" actually means in this case - files are still serving,
the machines are up.


NFS file access seems to work as expected, but when I reboot a gluster
server, the transition doesn't seem as smooth as I would have hoped and
I wonder if it's related to the amd automounter not being happy.


I have a very simple gluster setup - two servers (centos 6.5 with
glusterfs-3.4.3-2.el6.x86_64 packages), ctdb managing two shared IPs
(the hpa and hpb in the mounts show above).  Gluster configuration is
the defaults with 3, 2 replica volumes defined.  Client side is FreeBSD
8.2 with stock automounter config (NFS version 3, tcp).

rpcbind is running everywhere, dns forward and reverse, the network is
reliable (not busy), I don't see anything in /var/log/glusterfs/*
or the system logs that looks suspicious to me.

I've tried google-ing and document-ing looking for clues, but I've
come up with nothing.  I would be grateful for any insight or
suggestions anyone could provide.

Thanks!

John
jsellens at syonex.com