network ping using -node uses wrong lif

[BobMcKim]

I have a 4-node cluster with two lifs in the cluster vserver on each node - an intercluster lif and a node mgmt lif.  When running a network ping, with the -node specified, the traffic flows over the node mgmt lif on nodes 3 & 4.  Works as expected.  On nodes 1 & 2 the traffic flows over the intercluster lifs - problem.  Using the -vserver and -lif on the ping works on all 4 nodes.

The default gateway is specified on the node-mgmt lif.  The intercluster lif has no routing tables.  So why would a ping, to a network that has no entries in the node routing tables (node run node xxx netstat -r), flow over an interface with no routing information?????

How does one change which lif to use when using the -node parameter. 

[nitish]

In order to use a specific lif on a node for ping operations use the following syntax

network ping -lif-owner <vserver-name/node-name> -lif <lif-name> -destination <destination-ip-address>

Check the Intercluster and Node-mgmt routing/group "metric" value

routing-groups show -routing-group ic*|n*

Is the metric value same for all node?  

[BobMcKim]

 There is only the default route for the cluster SVM and it's on the mgmt network.

I should also point out the the pktt trace shows the node assigned the intercluster IP and it's in the losk interface trace, not in the trace for the interface where the IP for the intercluster lif resides (a0a-1232).

[BobMcKim]

Actually it looks like we've hit something similar to BUG 537982.  The conditions aren't exactly as described, we're not specifying the source LIF but using -node instead.  I'll open a case.

[BobMcKim · Accepted solution]

It ended up being a stale route cached in the first two nodes.