Gluster operations hangs when one node fail

glusterfshigh-availabilitymount

I have 3 nodes gluster server. Most of mounts from it are readonly and are to localhost like:

mount -t glusterfs localhost:/volume /mnt/mountpoint/ -o ro

volume is type replicate with 3 replicas (one brick on every node)

I've configured for volume options like:

cluster.quorum-count: 1
cluster.quorum-type: fixed
cluster.server-quorum-type: none

I would expect that fail/reboot of one node doesn't interrupt read operations from localy mounted volume. Unfortunally – when i reboot any node or any node is fail – even read operations like 

ls -l /mnt/mountpoint/somefile

hangs until failed node is up or network.ping-timeout is reached.
I've tried versions up to current 3.8.

Is there any way to configure glusterfs to serve (local!) content always, independent from failed nodes?

Best Answer

If someone is still looking for an answwer...

GlusterFS has a default timeout of 42 seconds. You can execute this command:

gluster volume set VOLUME_NAME network.ping-timeout 5

Where 5 is the amount of seconds you want your gluster to hang before becoming operational again. You only have to set this option on one of the nodes as it is a shared option.

Related Topic