cmviewcl time-out

[Lorenzo Meneguzzi]

Hi all,
some of my shell scripts use the cluster command "cmviewcl".
This result of this command is immediate in case all the nodes of the cluster are present. But if one node is not present, there's a time-out (about 15s) before the command returns.
Does anyone knows if it's possible to decrease this time-out ?
Thanks for your help.
Lorenzo.

[Karthik S S]

Hi,

I think that depends on the NODE_TIMEOUT value in the cluster.asc file (2000000 micro sec by default). Do a cmcheckconf on your cluster and see if you have set it to a higher value.

-KarthiK S S

[Carsten Krege]

Hi,

no, these timeout are hardcoded and there is no way to decrease them. Actually the ~15 seconds are the result of two queries that are sent out by cmviewcl. One on TCP with a 10s timeout and the other on UDP (a broadcast) with a 3s timeout that was sent after the TCP query timed out.

You should be aware that you shouldn't issue cmviewcl too often to check the status of the cluster, e.g. not every seconds. This might cause high CPU usage for cmcld on the local and on remote nodes. Once every minute should be sufficient.

Carsten

[Hoefnix]

Hi,
You can do a workarround if you keep track on which nodes are down in you cluster and then create a script that does the next:
cmviewcl -n node1 -n node3 -n node4
When node2 is down it will not try to figure out the state of the node.

HTH,
Peter

[Lorenzo Meneguzzi]

Thanks Carsten,
I thought it was something like that.
But apparently in case of a node quit the cluster with the command cmhaltnode, the command cmviewcl should return quickly. It was question on the MCSerGuard Forum and the reply from Stephen Doud (HP) was to refer to the doc UMCSGKBRC00013544.
But I can't find this doc. Do you know where I can get it ?
Thanks.
Lorenzo.
N.B. : if you want to find the question I'm refering to, enter the keywords "ServiceGuard cmviewcl command"

[Carsten Krege]

Hi Lorenzo,

no idea how to access the document on ITRC, at least I couldn't find it. I'm not a frequent user of the ITRC knowledge databases though - so it might be my ignorance. :) I'll give Stephen a heads-up to point this out.

However, I can see that Stephen opened an enhancement request regarding this issue in Nov 2003 which is still under consideration by the lab. Service request number (which will show up in a potential patch) is SR 8606338820.

Carsten

[Kent Ostby]

Lorenzo --

For whatever reason, that document is no longer a part of our certified database or has a new doc ID.

Reading through the old, internal document, it essentially just documents the behavior that you noted.

A discussion in another internal document indicates that by design, cmviewcl must wait at least 10 seconds before timing out in order to ensure that we are reporting correct status (i.e ensure that we dont say a node is down when we are having some sort of networking delay).

Starting with A.11.13, this delay should be closer to 10 seconds as that is the change that was put in at that time.

There is no way to make the delay less then 10 seconds when a node is not present.

Best regards,

Kent M. Ostby

[Stephen Doud]

DocId: UMCSGKBRC00013544
TITLE: cmviewcl responds slowly when a node is not present
Updated: 3/11/04 5:33:00 AM (submitted to the customer-viewable database today)

PROBLEM
When I perform cmviewcl, I noticed that it takes much longer (15 seconds) to
report than it normally does. What could be the cause of this?

RESOLUTION

One confirmed reason is that one of the nodes in the cluster is no longer
reachable:

WITH ALL NODES RESPONDING:

# timex cmviewcl | grep real

real 0.87 (seconds)
user 0.08
sys 0.09


WITH A NODE OUT OF THE CLUSTER:

# timex cmviewcl | grep real

real 14.85 (seconds)
user 0.09
sys 0.11


As Carsten stated, I submitted a change request to improve SG to skip the departed node. Unfortunately until the lab does something about this annoyance, you are stuck with the delay.

Regards,
Stephen Doud

[Lorenzo Meneguzzi]

Thanks for all the details.
I think I will change the way for my scripts.
Lorenzo.