How gracefully does it "fail" if there is a loss in iSCSI connectivity?

[bill_french]

I'm getting ready to do some experiments on ESX and iSCSI, but I'm curious if anyone has any experience with how an ESX server handles a loss in connectivity to an iSCSI system. How quickly does it fail? How does it fail? If a VM has two vmdks, one on FC, one on iSCSI, and the iSCSI connection drops, what can I expect? Thanks, --Bill

[acr]

With any failover it wont instant, but it works well, just remember with iSCSI you need to make sure your switches are set with portfast (esp for Cisco), otherwise you'll get some interesting results..

[bill_french]

Thanks for the reply. I haven't been on these boards in a while, everything has been working so well... i guess too well, now I have to go playing around with iSCSI to find ways to blow it up.

Anyway, just to clarify my question, I'm not talking about "failover" -- i'm asking more about "Fail". I'm considering having some iSCSI stuff in a building seperate from the ESX servers, and am wondering what would happen if a total loss of connectivity to the iSCSI happened. When the iSCSI initiator is within the VM (Windows) itself, typically the drive just disappears, and reappears when the connection is reestablished. It's fairly graceful from a "hardware" perspective, although various applications don't typically appreciate data temporarily vanishing.

But if the iSCSI initiator is within ESX, i suspect the behavior is going to be different, and, i fear, more drastic. I'm going to try and and see for myself, but I am curious as to what others have seen.

[ronzo]

Hi Bill-

I was looking at some logs today on a host that lost the iSCSI.. It caused the host to lock hard... required a reboot on the host..

thanks

ron

[Dave_Mishchenko]

Hi Bill, I have a similar setup on the home lab and should the iSCSI target fail / reboot / etc

1) The VMs die (only drive in the VM stored on the iSCSI LUN).

2) ESX will mark the VMs on those LUNs as inaccessible.

3) In some cases the CPU usage in the service console would run really high for a while (but that was mostly with 3.0.0).

4) Other VMs not on the iSCSI target continue to run fine.

I've never had to reboot an ESX host due to an iSCSI target failure.

[RParker]

That is completely different from Fibre. Loss of Fibre locks the entire host, and I can't reboot at all. I have to kill power, and cold boot.

We are considering iSCSI for future, because our SAN won't failover Fibre.. sadly, but it appears iSCSI is more forgiving, which is good to know.

[BUGCHK]

I was able to recover an ESX 2.5.4 and two 3.x servers from an accidental shutdown of an EVA4000. Rescanning the adapters brought back the data.

[wila]

I have seen the exact same thing happening over here in the lab using the software iscsi solution when i ran some tests, but i also believe that the complete answer isn't as simple as that.

It also depends a bit on your configuration. For example is the host booting from SAN or local disks? As long as the ESX host itself has the capability to read/write its temp files and read its own configuration, it should stay alive.

But i'm sure there will be some exceptions to this such as the exact operation the client VMs were in at the time of the connection loss, your hardware configuration, etc...

If your VM was running at the time of the disconnect i would personally choose to trash it and restore a backup of the VM to use instead.