|
Hello,
Running Gluster 3.7.2, distributed(3) / replicated(2) / 20Gb teamed network. Pretty much default options.
I am having an issue with self heal in Gluster. The second gluster node disconnected overnight, and now large files such as VM images are out of sync. The self-heal does not use the full disc IO, nor the full network bandwidth, but it does appear to slow down the volume during a heal, and lock up the files it is healing. A simple # df -h takes 1 - 2 minutes, where the volumes are FUSE mounted. The VMs using the healing image files freeze completely, also freezing Virt-Manager, and then all of the other VMs either freeze or become slow. In the interest of getting our office network operational this morning, I have removed the second node from the network - this has stopped the self heal and allowed us to work with just 1 storage node - but it is very slow and repeatedly freezes for a few seconds every minute or so. As I understand, gluster uses granular locking during a heal, so that a VM image may not entirely lock up during the heal / diff checking. However, it appears the entire VM hangs waiting for the heal to finish, with errors such as 'Device or Resource Busy'. The effect we are observing with the second node disconnected (few second freezing every few minutes) is was I would expect if it was healing with granular locking, where it would freeze each time it attempted to heal a block that was being accessed. I have been looking at the qemu / gluster libgfapi, but have been unable to make this work. From what I understand, this will only improve performance. Will this also allow a VM to be accessible during a self heal?
Other than not using the libgfapi for the VM images, is there anything else I am doing wrong?
Is there a way to use a VM image whilst it is healing?
Many thanks
|
_______________________________________________ Gluster-users mailing list Gluster-users@xxxxxxxxxxx http://www.gluster.org/mailman/listinfo/gluster-users
