2 VSA's in a VMware HA Cluster?

[MGovus]

Hi,

I am trying to setup a remote office with the following setup: - 2 x ESX servers and 1 x ESXi. The ESXi is running a Failover Manager. The ESX servers are running a VSA each. The VSA's are configured to be in a VSA cluster, have 2-way replication and are presenting the volume back to the two ESX servers via iSCSI.

The ESX servers are in a VMware HA/DRS cluster with a failover capacity of 1 out of the two hosts and DRS is set to fully auto. The VSA's have been set to "Disabled" in the "HA Restart Priority" and "Power Off" in the "HA Isolation response". The VSA's have also been disabled in the DRS section. The VSA's have the correct reservations for memory and CPU.

A Test VM has been created and stored in the Lun presented by the VSA's. The VM is set to be restarted by HA and moved by DRS.

The problem is that when we fail an ESX server to simulate a host failure (this is a power off, not a shutdown) the VMware HA either fails to start the test vm up on the other host or will take anything between 10-20 minutes to work. vMotion and cold migration work perfectly when everything is up. Network times are spot on.

Can someone tell me if this is a configuration that isn't supported? What is the best way to get ESX HA in a remote office using VSA's?

After two weeks in a basement computer room I need help!! :o)

[Uwe Zessin]

see the user guide:

* HP LeftHand P4000 VSA User Guide
TA688-96002__20090423 (April 2009) - c01727958

Page 12:
HP LeftHand Networks does not support the following configurations or procedures.

- Use of any shared storage as the data location of the VSA. This includes shared DAS, Fibre Channel, iSCSI, and NAS.
- Use of VMware snapshots, VMotion, HA, or DRS on the VSA itself.
...
- Co-location of a VSA and other virtual machines on the same VMFS datastore.

[Uwe Zessin]

Other question: did you implement a Failover Manager (FOM) VM?

[MGovus]

I've read the user guide. As stated in my post I have disabled the VSA's from being a part of any vMotion, DRS and HA.

A failover manager is running on an ESXi server and works flawlessly.

[Uwe Zessin]

How quick does the VSA failover happen?

I ask because the first implemention of HA did attempt to power-on the lost VMs after 15 seconds(!) on the surviving ESX server(s) which could be too quick and fail if the datastores are not ready.

Have not checked if that was increased recently as I do set "das.failuredetectiontime" to 60..120 seconds anyway.

[MGovus]

Soooo potentially the issue might be that VMware HA is trying to failover before the VSA is ready?

I've checked and the HA time is 15 seconds. I've not changed that on our setup.

I'm assuming if I look at the Management group status and see how long it takes for the VSA to be marked offline in order to get that time?

Thanks for your help.

Mark.

[MGovus]

VSA failover time is 5 seconds.

[Keith Martin_3]

Hi,
I'm in the process of setting up a similar environment and wondered if you ever got things working. It seems like a strange environment considering that you're creating VMs within VMs(the VSAs) and I was interested to see how you were able to configure HA for just a select few VMs, and exclude the VSA VMs from the HA failover.

Please let me know if you were able to get this working in your environment. Thanks.
Keith

[teledata]

Technically you don't really need to "exclude" the VSAs from HA/DRS, because they inherintly WON'T atempt to v-motion or failover because they have virtual disks that are local storage only. However for thoroughness you can control individual VM behavior (for HA/DRS) under the cluster settings in vcenter: Right click your Cluster -> Edit Settings -> under VMware HA -> Virtual Machine Options: Select the vms you want and change their restart priority to disabled.

Same for DRS:Right click your Cluster -> Edit Settings -> under VMware DRS -> Virtual Machine Options: Select the vms you want and change their Automation Level to Manual (or partially automated (DRS decides where to start it on power on)


One of the biggest challenges I ran into (when running production VMs on the same esx host that hosts your VSAs) was the graceful startup and shutdown of the VSA, and to get the ESX datastores to come up automatically after a reboot. That was fixed through a collaborative effort on the forum here (post a reply and Adam or myself can find the link for the script) that consisted of a script that re-runs the "auto-start" script AFTER the VSA VM comes back online after a esx host reboot, and then during a shutdown it gracefully stops the VSA, and disables iSCSI (to prevent volume UNmounting for volumes that are missing.)

[damiri]

Hello,

we are also considering VSA on two packed DL380 G6 servers to have failover. However, questions are:

- two servers are enough ?
- I am talking for networks with 50 to 100 users. Is VSA viable solution for this environment runnig file, mssql, exchange or lotus domino servers ?

Thanks,

Damir