You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I'm using 0.6.4.1 on CentOS 7, exporting zvols to ESXi via scst/vdisk_blockio/iscsi
I have a 4-disks/raidz1 pool (I know, not ideal) and an SSD. L2ARC definitely seems to help, until I have to remove it from the pool.
From what I see, removing the L2ARC cache device can take a very long time (up to 5 minutes measured) and IOs to pool zvols are completely blocked during that time (as reported by scst).
I already reported this to the mailing list a few weeks ago on 0.6.3 but wanted to wait for 0.6.4 for some additional tests. It looks like it didn't change.
It might be a "by design" issue, but it makes ZFS almost impractical for a SAN cluster setup where you expect to switch pools from one node to another in just a few seconds (I would say under 30 seconds).
I can add that the time required for removing the cache device is proportional to how much the L2ARC is used (as reported by arcstat.sh).
The text was updated successfully, but these errors were encountered:
I'm using 0.6.4.1 on CentOS 7, exporting zvols to ESXi via scst/vdisk_blockio/iscsi
I have a 4-disks/raidz1 pool (I know, not ideal) and an SSD. L2ARC definitely seems to help, until I have to remove it from the pool.
From what I see, removing the L2ARC cache device can take a very long time (up to 5 minutes measured) and IOs to pool zvols are completely blocked during that time (as reported by scst).
I already reported this to the mailing list a few weeks ago on 0.6.3 but wanted to wait for 0.6.4 for some additional tests. It looks like it didn't change.
It might be a "by design" issue, but it makes ZFS almost impractical for a SAN cluster setup where you expect to switch pools from one node to another in just a few seconds (I would say under 30 seconds).
I can add that the time required for removing the cache device is proportional to how much the L2ARC is used (as reported by arcstat.sh).
The text was updated successfully, but these errors were encountered: