Thank you!
Yes, that's logical and consistent.
I suspect description of system-defined failover policy in documentation is incomplete and misleading. This would imply two-node cluster needs different policy, and there is nothing anywhere about it. Also example in documentation contradicts description:
The following command displays information about the failover targets for all LIFs in a two-node cluster. The Failover Targets row shows the (prioritized) list of node-port combinations for a given LIF.
cluster1::> network interface show -failover Logical Home Failover Failover Vserver Interface Node:Port Policy Group -------- ------------- --------------------- --------------- --------- ... vs1 data1 node1:e0e system-defined bcast1 Failover Targets: node1:e0e, node1:e0f, node2:e0e, node2:e0f
I /hope/ that system-defined is intelligent enough to realize that in two-node cluster the only available failover targets are on HA partner. May be Justin can chime in...
-----Original Message----- From: Gelb, Scott [mailto:scott@redeight.com] Sent: Wednesday, July 01, 2015 11:48 PM To: momonth@gmail.com; Borzenkov, Andrei Cc: Parisi, Justin; Martin; toasters@teaparty.net Subject: RE: CDOT failover groups in 8.2.1P6
I upgraded 2 clusters today with overlapping failover groups... the upgrade leaves the failover groups intact and creates a broadcast-domain (and corresponding failover-group) with all of the combined ports...but leaves the lifs as is on the prior failover-group
For example on 8.2
fg-cluster-mgmt node1:e0a,node2:e0a fg-node1-mgmt node1:e0a,node1:e0M fg-node2-mgmt node2:e0a,node2:e0M
Prior to 8.3, we didn’t have a local only policy so had e0a overlapping between all 3 failover groups. That way the local management could also use e0M as backup. The end result was a new broadcast domain with all 4 ports (e0a and e0M on each controller named 10.x.x.x), a failover-group 10.x.x.x from the broadcast-domain, then the same manual failover-groups prior to upgrade.
One other note though...after upgrade, all SVM data lifs had a failover-policy of "system-defined" after the upgrade. This allows failover to the home-node and non-sfo partner...in a 2-node cluster there is no non-sfo partner. I modified these to "broadcast-domain-wide" after upgrade. In many cases we can eliminate manual failover groups and name broadcast-domains per vlan, but in this case above with a mix of 1Gb and e0M (or if you have 10Gb and 1Gb) it makes sense to have manual failover groups so slower speed ports are not used for data.
-----Original Message----- From: vladimir.zhigulin@gmail.com [mailto:vladimir.zhigulin@gmail.com] On Behalf Of Momonth Sent: Wednesday, June 10, 2015 4:04 AM To: Borzenkov, Andrei Cc: Gelb, Scott; Parisi, Justin; Martin; toasters@teaparty.net Subject: Re: CDOT failover groups in 8.2.1P6
I currently have non-overlapping failover groups on 8.2.1P3 nodes, can experiment a bit if needed. I'm about to upgrade those nodes to 8.3
Vladimir
On Mon, Jun 1, 2015 at 8:45 AM, Borzenkov, Andrei andrei.borzenkov@ts.fujitsu.com wrote:
What is unclear to me - failover groups can be overlapping (if my memory serves me right, I can add port to multiple failover groups) and broadcast domains cannot. This makes it impossible in general to create broadcast domain for each failover group.
If there is requirement for failover groups being non-overlapping I cannot find it in documentation.
So I wonder how upgrade handles this situation.