We are running Data OnTAP 8.0.2P7 on a FAS3270 HA Pair. While we have iSCSI running on both instances of the HA Pair, we need to run isolated instances of iSCSI in two new vFilers we are planning to stand up. I am aware that iSCSI should work fine within a vFiler, however I am unsure about what (if any) configurations I need to make to ensure that failover would work in this scenario.
Typically, to ensure failover, you make sure that the interfaces on each node of the HA Pair have matching portal group tags. In this case, however, the interface is affiliated with the vFiler, and iSCSI is configured with the vFiler. Do I need to convert to IP-based portal group tags for this to work, or can I leave the tags as they are? In other words, as long as I create a matching failover interface like I would for any vFiler, is that enough to ensure failover of iSCSI functionality?
I'm not sure I understand how portal group is related to failover. During failover virtual instance of controller is started on partner. It has exactly the same configuration, there is nothing to match against. So you simply need to ensure that every interface has matching partner interface, that's all.
Thanks for your reply aborzenkov! The following is from page 28 of the Block Management Guide for iSCSI and FC (emphasis mine), and this is what I'm basing my concern on.
How iSCSI works with HA pairs
HA pairs provide high availability because one system in the HA pair can take over if its partner fails. During failover, the working system assumes the IP addresses of the failed partner and can continue to support iSCSI LUNs.
The two systems in the HA pair should have identical networking hardware with equivalent network configurations. The target portal group tags associated with each networking interface must be the same on both systems in the configuration. This ensures that the hosts see the same IP addresses and target portal group tags whether connected to the original storage system or connected to the partner during failover.
Interesting. I have never noticed this before; and I have never seen it mentioned in any document from NetApp (technical reports, training materials etc).
Even though I’m rather skeptical here, why not. Actually usually configuration is symmetrical, so same interfaces get the same target portal tag. But that’s exactly what makes me skeptical - there is no reason to assume that a) interface configuration is symmetrical or b) that both controllers even have the same interfaces.
Sure, but paragraph from block management guide quoted by OP does not say a single word about multistore (it talks about HA pair), nor does multistore guide say anything about target portal groups. That is why I find it extremely confusing.
David, the comment by aborzenkov is exactly my issue. I had read the Multistore section titled "LUNs on a vFiler Unit", however this section doesn't address the comment from the "Block Management Guide for iSCSI and FC" about Target Portal Group Tags needing to be the same on both sides.
I have read about IP Based Portal Group Tags, however this appears to be only needed/recommended for migrating a vFiler that contains LUNs. I would rather not use them if they aren't necessary. Let me know if anyone else has thoughts!