Detaching Filesystem Results in iSCSI Becoming Unavailable to vSphere

[dobler]

Hello,

I have a backup array that I attach to the system. Whenever I detach the backup array the iSCSI for the active vSphere datastores becomes unresponsive forcing me to reboot the vSphere server to regain the connection. Is this normal behavior? I'd like to know if there is a way to avoid this type of issue from happening. I have the vSphere attached through a 4g fiber with a backup 1g ethernet connection.

 

[dlavigne]

I asked one of the devs, who suggested this:

It is hard to diagnose the problem with provided information, but I have a guess. ZFS very dislike when imported pool looses its critical mass
(or completely disconnected). Such pool may hang requests sent to it, and IIRC even completely block the storage subsystem, causing problems for other pools.

I would 1) make sure that backup pool is always exported before detach, and 2) removed swap partitions from those disks.

 

[dobler]

Please let me know what pertinent info that you need me to supply to help diagnose. I'd like to add that the filesystem is encrypted. Haven't had a chance to test with unencrypted in case the enryption is the issue. Also, it is not just the export that causes iSCSI loss, import causes loss as well.

Regarding what you have suggested:
- I am always exporting before detaching. It is the export function that causes the iSCSI loss, not the physical disconnect.
- Not sure what you mean by swap partitions (l2arc, zil?) It is just a single hd.

 

[cyberjock]

Is your system dataset on the zpool you are detaching? The .system dataset is critical for many critical services and processes that run normally. If you suddenly remove the zpool that has the .system dataset (or even just delete the .system dataset) then the system quickly becomes unusable.

 

[dobler]

Its not the FreeNAS itself that is becoming unresponsive. It is that the iSCSI share turns off for a brief moment rendering the connected vsphere server inoperable during the export/import routine. The FreeNAS remains stable, but the vSphere is unable to recover from the disconnect.

The system dataset is not on the zpool being detached. It is just a backup zpool.

Next I may try switching the pathway to gig ethernet to see if maybe the issue is FreeNAS's fiber channel support. Also, I'm considering trying exporting a non encrypted volume to see if it is an issue with FreeNAS's encryption system.

 

[cyberjock]

I won't lie.. I'm not really aware of anyone personally that has had 2 zpools, and while running iSCSI and in use, tried to unmount the second zpool. So there may be some kind of limitation here. I definitely would like to see if there is a bug here or not, I just don't have the ability to verify this is/isn't the case. Please don't let this fall off the radar. If you end up in a position where you cannot test further please let me know by updating this ticket or by PM so I can see if I can get a test environment for this in a lab.

I will say that FC support isn't all that well supported at present, so there's definitely a possibility that FC is related.

 

[dobler]

thank you cyberjock, I'll report back as soon as I have time to test.