is_node_alive hangs on MCSG 11.12 and MCSG Extension for SAP B.03.04

[Mark van Hassel]

Hi all,

We are building a MCSG SAP cluster. The SAP packages source the /etc/cmcluster/sap.functions script that is part of the SAP extension toolkit. In this script is the is_node_alive function:
function is_node_alive
{
set -x
NODE=$1
RETURN=`ping $NODE 64 2 | tail -1 | awk '{print $1}'`

if [ $RETURN = "round-trip" ]; then
return 1
else
return 0
fi
}

Whenver this script is called it hangs on the ping. I ran some tests and discovered that the ping only sends one request and than hangs. Since the ping count in is_node_alive is set to 2 it hangs.
To make it even stranger: when I run is_node_alive outside the MCSG environment it does not hang ! This implies that the cmcld daemon is the culprit.

Has anyone encountered this before ?

[melvyn burnard · Accepted solution]

I believe the symptoms you are seeing are those documented in PHSS_22541:

11.Any application using SIGALRM will have problems with
ServiceGuard. Such applications will not start/work
properly when started from the ServiceGuard Package
Manager.




I have seen a few sites have these "hangs" using the Sybase software, but it is by no means limited to just Sybase.
This patch has been superseded and you should install PHSS_24305 available from this site.

HTH

[Carsten Krege]

Melvyn is right.

The unpatched ServiceGuard 11.12 causes SIGALRM to be blocked from any process started from a package including the package script itself.

ping uses SIGALRM and without the SG 11.12 patch he mentioned, your SGeSAP package will hang.

Carsten

[Mark van Hassel]

Melvyn,

Thanks, that is the solution. Strange though that the patch is not mentioned in the Release Notes.

[Carsten Krege]

Release Notes are created only for the initial (ie. unpatched) version of SG. The patch didn't exist when the release notes were posted.

Release Notes are not updated.

Carsten