packages halt with snmp start

[Rick Garland]

Hi all:

Got MC/SG 11.16 on HPUX 11.23 running on rx4640 hardware.

Have discovered that when stopping/starting the SNMP processes, the SG monitoring senses a network hiccup and drops the packages. Needless to say, on PRD systems this is not good.

Have done this countless times before with no ill effects. Been doing this well over a year. Now it is all of the sudden.

Any ideas as to what is happening?

[Prashant Zanwar_4]

Could be that the port service is already in use and causing problem..
just trace using lsof..

Rgds
Prashant

[Sameer_Nirmal]

Hi Rick,

One possiblity..

Check if you have these daemons running
cmsnmpd - It provides HA for Cluster MIB objects. It is also used by ServiceGuard Manager. Although it is optional, it might have been used in your case.
snmpdm - It is used by cmsnmpd.

You can refer the man pages of these daemons to know the dependancy.

[Rick Garland]

Thanks for the replies thus far. No luck with either of the suggestions.

So far HP does not have an answer. I have provided them with the control files, the build parameters, etc. and they cannoty reproduce this. This is being escalated to WTEC.

Any other ideas out there?

Many thanks

[John Bigg · Accepted solution]

If you have packages that define EMS resources then those EMS resources may require both EMS and SNMP to be running. If you halt these then at the next defined poll period Serviceguard will be unable to deterimne the resource value and will therefore halt the package. I would expect the following sorts of messages in syslog should this occur:

cmcld: Resource monitor for resource XXX is having problems.
cmcld: Resource XXX is assumed to be unavailable.
cmcld: Resource XXX does not meet package ZZZ for package YYY.
cmcld: Resource XXX in package YYY does not meet ZZZ.
cmcld: Executing 'SSS stop' for package YYY, as service PKG*NNNNN.

It is possible that you get lucky and manage to halt and restart SNMP between 2 polls and not notice any problems, but if a poll happens to occur while SNMP is down then the package will be brought down.

If you need to re-start SNMP you should halt any packages (or move to another node if needed) that have EMS resources defined unless of course you like playing russian roulette.

You should hear this from WTEC soon :-)

[Rick Garland]

Hi John:

Sounds as if you are/were working this issue at HP.

All sounds good. And no, I do not like playing roulette.

Thanks much. I am expecting WTEC to say the same. Will let you know.

[John Bigg]

I am the WTEC engineer.

[Rick Garland]

John:

I should have known ...

Many thanks!