Hi,
I created a striped zfs pool on two nodes and then put replicated glusterfs on top. This is just testing cluster so no spare drive or raidz.
I tried to pull out one drive, the pool got into unavailable state and all vms on top of glusterfs froze.
I discovered that I need to set failmode to panic instead of wait with zpool set failmode=panic tank. Then after a failure, the affected server was restarted by other nodes in fence domain and vms were still running correctly.
What I would like to achieve is that gluster detects unavailable storage or filesystem and forward all data requests to a working node, without needed restart of problematic node. Or zfs dataset get unmounted somehow. Is it possible somehow? I did not find any answer to this question.
Thank you for your answers.
I created a striped zfs pool on two nodes and then put replicated glusterfs on top. This is just testing cluster so no spare drive or raidz.
I tried to pull out one drive, the pool got into unavailable state and all vms on top of glusterfs froze.
I discovered that I need to set failmode to panic instead of wait with zpool set failmode=panic tank. Then after a failure, the affected server was restarted by other nodes in fence domain and vms were still running correctly.
What I would like to achieve is that gluster detects unavailable storage or filesystem and forward all data requests to a working node, without needed restart of problematic node. Or zfs dataset get unmounted somehow. Is it possible somehow? I did not find any answer to this question.
Thank you for your answers.