Cluster Failover Question

[kofchur]

I have a customer that has a filer cluster with 2 stacks that are MP-HA.  He wants to add another stack that will serve only as a backup and, for various reasons, wants to connect it non-MP-HA, that is only one connection to one filer head and another connection to another filer head.  He also does not want the filer to failover due to a shelf/cable/GBic issue on the backup stack/loop.  Now, if we leave the default cf.takeover.on_disk_shelf_miscompare to "off", if we loose a cable, GBic, or ESH module on the backup stack, the cluster will not fail over.  Is this thinking correct, or am I missing something?  Thanks!

[andrc · Accepted solution]

That sounds correct; as long as they're aware that if either controller loses connectivity to the shelf, then they will not be able to access data in volumes based on disks in that stack.

It is also the case that if either controller loses connectivity to the stack and an aggregate goes offline because of it, the controller will panic which could trigger a takeover anyway depending on if the cf.takeover.on_panic option is enabled.

[aborzenkov]

cf.takeover.on_disk_shelf_miscompare is ignored for software based disk ownership (should should be default today). So you are correct – as long as filer does not have online aggregate on the shelf, it should not panic and imitate failover.

[thomas_glodde]

With the planed setup, the filer will failover if you loose a cable, gbic or esh module! you have online volumes and online aggregates in that machine, if you loose connectivity to the disks you will have a multiple disk missing error and the filer will panic!

you might consider buying a FAS2020 with internal S-ATA disks for the purpose of "just a bit plain disk storage" instead of endangering your productive HA-system.