Starwind 5 HA failback

TomHelp4IT · Thu Nov 12, 2009 11:53 pm

I wanted to clarify a couple of things about how the HA works since I can't find much documentation:

I've setup a pair of Win2k8 Starwind 5 servers with a single HA iSCSI target connected to a pair of ESXi4 servers, all seems to be working fine and performance is good. Although Starwind refers to one HA image as the "primary" and the other partner as the "secondary" ESXi didnt make any differentiation, just seeing one device with 2 paths, which makes sense. I rebooted one of the Starwind servers and the ESXi servers weren't bothered so the HA is working fine, however I noticed a couple of issues:

1) After the reboot the Starwind console showed a warning that the HA target was out of sync, I left this about 30 min and it didnt go away so I tried manually initiating a "full synchronisation". The warning message was a bit confusing about which device the sync should be initiated from making it quite easy to sync the wrong way if you are not careful. Anyway, does Starwind not automatically re-synchronise the cluster after a node has been offline?

2) What's the difference between a full and a fast re-sync? Since it appears the cluster loses its HA status until re-synced I would like to get it updated as soon as possible but is there a risk of data loss/corruption with the fast option?

Robert (staff) · Fri Nov 13, 2009 3:46 pm

Hi Tom,

Yes, we have not included automatic re-synchronization between the nodes alowing administrator to decide what way to synchronize the cluster.

Fast sync is an incremental synchronization that only synchronizes the data that was actually changed while one of the nodes was down. So, fast sync in terms of the speed works much faster than the full synchronization, which will make your HA cluster fully synchronized again much faster.

Thanks.

TomHelp4IT · Thu Nov 19, 2009 1:41 pm

OK, I can understand that there may be scenarios where you would want to control the direction of sync but the lack of an alerting facility concerns me. What happens if one of the HA server pair reboots itself for some reason, or the Starwind service is restarted - one half of the pair will then be out of sync so there is no fault tolerance until it is manually resynced?

If Starwind was able to send some sort of email alert saying the HA targets were out of sync then it wouldnt be a problem for someone to logon remotely and resync it, otherwise we would have to regularly check it just in case.

I have been experimenting on our test cluster with using the VMware "storage connection" alert to email us instead, which works ok except there is a bug in ESXi4 currently which doesnt clear the error properly after the paths are restored so it keeps sending out emails. I think there is a patch available now which will fix that though. Not much help though if you aren't running a full vSphere setup.

adns · Wed Nov 25, 2009 6:09 am

Amen! There should be the option to choose auto-synchronize and there should be alerting and logging on any errors (similar to a RAID card's alerting).

Robert (staff) · Fri Nov 27, 2009 4:52 am

This will definately be implemented in the future versions.

Thanks.

TomHelp4IT · Mon Dec 07, 2009 6:19 pm

Good, I just installed ESXi4 Update 1 on our test cluster and it doesn't seem to have fixed the storage connection alerting problem so that's still not an option unfortunately. In the meantime I'm investigating monitoring the Starwind service itself on each server but its not ideal.

Btw, have you done anything about new version announcements yet? With 4 we only ever found out when there was a new point release because the file name would change on the download

Thanks

Constantin (staff) · Fri Jan 15, 2010 2:48 pm

We are working on implementing notification functionality.
We will publish such information on site, and on forum. Next version, that will be released in the end of January is StarWind 5.5.