Hello,
I am currently running a 2k2R2 cluster with shared storage, mostly as a file share. Yesterday the HBA on one of the nodes went bad and showed intermittent connectivity. I would have guessed Windows Clustering Service to detect this loss of disk and automatically migrate the affected services to another, unaffected node. Instead it didn't raise any warning on the cluster or client side, continued to present the defunct share to our users and still allowed full access. For writes I found a Event 50 (Lost Delayed-Write Data) as a warning in the nodes local system log. The modifications seem to have never reached the disk, so while the user thought he successfully saved a file it actually went to /dev/null. I've seen a similar behaviour on another cluster in the past.
Is there a way to tell clustering to fail over on such failures? While I probably could get it to react how I would like to see by disabling write cache this doesn't really seem like a good idea performancewise.