ServiceGuard errors: Segmented LAN

[Thomas J. Harrold_1]

Let me start by saying that I am an HP-UX CSE, and I've been building SG clusters for years. This problem has me stumped, so I'd like to see if anyone out there can offer some suggestions...

I attempt to build a cluster config on a pair of rx4640's. Each has two quad 10/100 network cards, as well as the built in dual GB card.

When I try to use the quad cards for primary and standby LANs (one port on each card), I get an error about the network being segmented. This is not possible, as I am using a simple layer 2 HUB (for testing purposes - production network is not yet available).

I can run linkloop between the interfaces, and there does not appear to be ANY issues!

If I switch, and use the 2-port ethernet card, I have no problems.

Is there an issue with the quad ethernet cards and SG?

This is a strange one...

Any troubleshooting ideas are welcome...

-tjh

[Thomas J. Harrold_1]

Some more details:

Each server has a quad ethernet card in PCI slots 4 and 5, as I recall.

I connect the top port (lan2 and lan6) of each card (on both servers) to the hub.

I run cmquerycl with the appropriate options, and a cluster ascii file is created with no errors. I then make some customizations, and try to create the cluster binary, but it fails, stating that my LAN is segmented: lan2 (node1) and lan2 (node2) are in different segments.

I have used linkloop to validate that these interfaces can communicate properly.

Am I missing something?

I have validated that both lan cards are configured the same: 100FD.

The swich is an old netgear, unmanaged hub.

Thanks in advance...

-tjh

[Geoff Wild]

To answer "Is there an issue with the quad ethernet cards and MC/SG?"

There shouldn't be - I have quads in my MC/SG environments.

That said - I only have rp's not rx's.

I would place a call with HP support and ask them.

Rgds...Geoff

[baiju_3]

We have 4 port 10/100 cards in our MC/SG Setup .

Please test the below things before proceeding .


1) Is any of the node you try to configure is already a part of cluster ? I mean any of your cluster is not updated properly after removing this node ?

2)Is these LAN cards are part of APA ?


Just few tips , your problem is strange .

thx,
bl.

[Helen French]

Sometimes, MC/SG discovery does not match with the result you get from linkloop command, since it uses a slightly different method. In this case you can use an 'unsupported' tool - 'dlpiping'to check connectivity:



# dlpiping 2 1 (test local PPA1 to local PPA2)

# dlpiping -n remote_node 2 1 (test local PPA1 to remote PPA2)

[melvyn burnard]

Well thuis may be a number of things.
It could be something silly like subnet masks, or devices on the net preventing certain traffic types or ports from going through.
I also have seen duplicate MAC addreses cause issues.

I do know that there have been fixes for issues like this in patches in the past. Do you have the latest Serviceguard and Networking patches installed on all nodes?

A cmquerycl does a general broadcast whereas cmapplyconf uses specific information.

[Thomas J. Harrold_1]

Nodes are new, and not part of a previous cluster.

The hub is just that...a dumb hub, and it's the ONLY network connection point between these nodes. I doubt there are any routing issues.

I'll try the unsupported utility, and I'll check for relevent patches.

Thanks for the suggestions, and if anyone else has any more ideas, please pass them along.

-tjh

[Sameer_Nirmal]

There should not be any problem using Quad-ethernet cards with SG.

I could make out following from the whole scenario.
When in built GB cards are used, the auto-negotiation might be enabled which would have worked with that hub.
For the quad cards , making 100FD ports are not being supported by the hub.
As per the error , lan2 ports are not being seen by each other. It just the matter the way SG probes the link level connectivity which it does correctly and found lan6 seeing each other, but it can't for lan2.

Check if it work by turning on auto-negotiation for quad ethernet card.

I would follow these commands to cross-check all network aspects.
ioscan
ifconfig
lanscan
cmviewconf
cmquerycl -l net -v
linkloop
dlpiping
nettl
cmsetlog -M NET

[Stephen Doud]

Have you tried different combinations of the quad cards? Eg. 1 in each server vs. 2 in each server.
Also as Melvyn said, don't rule out fundamental network configuration mistakes such as speed/duplex mismatch, subnet mask gaffs, duplicate MAC addresses.
Also beware of attaching two network devices like hubs with a regular lan cable when a crossover cable is needed.
Can you also please share the actual text messages of the errors that you get? Without them, we're just guessing.

[Steve Lewis]

Nasty problem.

I got that error when I got 2 cables wrongly swapped over - plugged in each other's interface port.

For what its worth (probably not much) I suggest you simplify the cluster config as much as possible, unplug most cables, then add each new interface one at a time rebuilding the cluster and packages until you get the error.

Maybe one of the cables is bad.