Would a partner interface function correctly in a takeover if the interface is tagged but with no assigned ip?

[mallison]

All,

With a local configuration in the rc:

----

ifconfig NFS-972 10.72.12.12 netmask 255.255.254.0 mtusize 1500 -wins partner NFS-972

-----

and a HA-pair ifconfig of:

-----

NFS-972: flags=0x710a862<BROADCAST,RUNNING,MULTICAST,TCPCKSUM,NOWINS> mtu 1500

    ether 02:a0:98:27:01:ca (Enabled virtual interface)

----

with no IP assigned to the partner's obviously ifgrped and tagged interface, will a takeover work?

We have no way to check it without a test take/give, but if there is an IP assigned, there is probably a higher chance that it will work than otherwise, no?

In looking at another configuration a few moments later, I see this:

NFS-981: flags=0x310a862<BROADCAST,RUNNING,MULTICAST,TCPCKSUM> mtu 1500

    partner NFS-981 (not in use)

    ether 02:a0:98:26:3d:0e (Enabled virtual interface)

Indicating to me that there is information missing in the first (the "partner" in the ifconfig -a statement) that shows that the takeover would not work. Does that make sense?

Thanks,

Mike

[aborzenkov]

In this case my first concern is that interface is not up.

You can test this behavior in 7.x simulator which supports HA pair simulation (pity it was dropped in 8.x).

[mallison]

The interface not being up is a problem for me, but I suspect that is only because there is no IP on it. It is on a physical interface that is tagged with several other functioning vlans, so layer-1 is not an issue.

Other's note this is a "standby", I understand that, but there is no way to tell, without an IP, whether or not the tagging on the switch port is functional.

Thanks!

[scottgelb]

Good points. You could manually ifconfig an ip to the node and see if it works the. Ifconfig 0.0.0.0 to remove it after the test but leaving it as a standby partner interface.

Sent from my iPhone 4S

[scottgelb]

An interface with a partner but no IP is a standby interface and works just fine... although check the partner and why it doesn't have partner showing in the ifconfig.  The partner interface works the same on takeover whether or not it has an IP address (or additional ip addresses with aliases) or no ip address it still picks it up on the partner.  What is the ifconfig statement on the partner? The ifconfig -a output is listed but not the commands in the rc file that look like it doesn't have partner defined.

[mallison]

I should have been more specific the "ifconfig -a" are from the partner, while the ifconfig statement is from the local controller's rc.

The local controller's "ifconfig -a" for that interface is:

NFS-972: flags=0x794a863<UP,BROADCAST,RUNNING,MULTICAST,TCPCKSUM,NOWINS> mtu 1500

    inet 10.72.12.12 netmask-or-prefix 0xfffffe00 broadcast 10.181.239.255

    partner NFS-972 (not in use)

    ether 02:a0:98:24:68:52 (Enabled virtual interface)

Interestingly, the partner's ifconfig for NFS-972 doesn't have the "partner" statement correct, but the local's "ifconfig -a" shows correctly. The partner's ifconfig statement is configured correctly in rc and shows such in the local "ifconfig -a" output.

What it boils down to now, is whether or not the ambiguity of the missing "partner" statement in the partner's ifconfig going to cause an issue on a takevoer?

Does that mean that the  partner has not "accepted" that the local controller has declared a specific interface is expecting to land on "that" interface?

[sridharchevendra]

I just tested this i.e taking out the partner statement in my other node in the cluster and the interface did not fail over.

You need to add this statement in the /etc/rc for that virtual interface.

Thanks for the good one.

Also adding/removing Partner statements needs a reboot.It did not take it with source command.

[scottgelb]

I hit a similar issue at a customer where the standby interface also needed the partner setting (set to partner on both sides...even though one needed one-way)... not sure if a feature or bug but we tested takeover/giveback since a new install and found this behavior... may have changed in ONTAP since.. I think 8.0.1 or 8.0.2 on that one....also with ifgrp, vlans and the vlan in a non-default ipspace with a vfiler.

[aborzenkov]

Local controller has no idea about configuration of partners interfaces. The statement “ifconfig ifX partner ifY” means that when partner in takeover mode tries to configure interface ifY, it will be running via local interface ifX. Notice that nowhere does it refer to any “partner“ statement on partner controller.

Which means that failover of partner to this controller should work, but failover of this controller to partner won’t.

[brauntvr2swiss]

Hi Mike

In our environment we configure every partner interface with an IP... So we know always that this Interface would prober takeover the Partnerinterface.

I recommend to configure every HA Environment balanced. That means you have all VLANs/IPs active on both Controllers.........

Adapted to your enironment:

netapp1 ifconfig NFS-972 10.72.12.12 netmask 255.255.254.0 mtusize 1500 -wins partner NFS-972

netapp2 ifconfig NFS-972 10.72.12.13 netmask 255.255.254.0 mtusize 1500 -wins partner NFS-972

Hope this help...

Thomas

[sridharchevendra]

This will work fine without any trouble.

One thing you need to make sure is

1.VLAN has been tagged on the network switches.

2.This needs to be added to /etc/rc with the VLAN and IP address.

[mallison]

Thomas,

Thanks, I understand that part plenty well, this is a customer's configuration. I always want to see symmetry in the network configurations (as well as the vfiler setups), but I had never noticed this sort of discrepancy before.

Thanks!