VMware, ZoL, and targetcli odd issue...

[kemic]

A quick overview of the setup:

• VMware ESXi (5.5 and 6.0) on a C6100 node (for now)...
• Fedora Server 21 - ZFS on Linux - (Tested Fedora Server 23 also...) - 8x 3TB SATA 7200RPM drives on a Dell h310 (LSI-9211-8i flash). Mirrored pair vdevs for now.
• Brocade 200E Fabric Switch
• QLogic 2462 4Gbps HBA in all servers

Using targetcli I've exported a 200G ZVOL to the ESXi hosts and everything is fine... I can create Centos6/7 VMs no problem, no issues. Here's the weird one, when I try to install Fedora 23 Server on a VM it gets to 25% of the installing packages progress bar and the SAN server bombs, kernel panics, hard lockup... I can try to provide a log if anyone is curious. A lot of terrible red messages related to the qla2xxx module... I was pretty shocked to see a VM OS install killing my storage server.

I can only theorize that the kernel in Fedora 4.x is doing something fancy with the storage it's detecting from VMware... I don't need Fedora 23, it was merely a test case for my infrastructure. Pulled my hair out for a while before I decided to try another guest OS.

So as you can see, I have no idea where to post this (might be better in Linux forum, mods feel free to move if you agree).

 

[kemic]

For anyone curious:

I ended up fixing this! Looks like for whatever reason when you create a ZVOL on Linux the underlying block device is flagged as an SSD (Fedora 21 test server).

In turn when the block device is added to targetcli, the attribute on the backstore, is_nonrot, is set to 1 (as it detects an SSD from the underlying storage)... Manually setting this attribute to 0 causes ESXi to detect the device as a non-SSD and all is right with my test setup!

I would have to guess that Fedora 23 (guest OS) and or kernel 4.x in general was trying to do something fancy with the underlying SSD storage that it thought it had...

 

[T_Minus]

Thanks for following up on this @Jim great to reference inthe future.