HACMP doesn't deactivate a VG when falling back

[LeDam]

Hello everybody,
I had a very strange problem when testing HACMP failover-fallback on a AIX 4.3.3 cluster:
Sometimes, when doing a fallback, the only one VG from the only one ressource group stay online (but all FS are unmounted). In hacmp.out, all EVENTs are completed. No one failed.
On the default node, VG is varyon and applications are started without any errors!
But these crash after some minutes,...
My question is : why stay this VG online on the failover node?

 

[sbix]

What happens when you manually varyoff the vg (in that situation)?

 

[LeDam]

There is no more problem if we do a varyoffvg...
but normaly, HACMP do it itself...

 

[sbix]

Sometimes, by accident, VG are configured to be automatically varyed on ...if in a cluster a node has VG configured "autovary on" and another not, everything works fine when the 1st node has to get the control ... not so much when the other one has to take the resource .... it happens.
Dear LeDam, my question was ... when the problem happens can you MANUALLY vary off the vg on the node which is leaving the resource?

 

[LeDam]

yes i can...

 

[sbix]

Uhm ... what says log for cl_deactivate_vgs event?

 

[LeDam]

Nothing....
But you were right, one VG was autovaryon. This is very strange because when i setup this VG, i put it with autovaryon off...Without that, HACMP synchronization doesn't work isn't it?

 

[sbix]

Indeed it should fail in verify step

 

[jackson74]

ledam,

you may have to manually intervene when changes are made out of order or a process hangs during (re)integration.

you can solve most HA vg/lv/fs problems volume group is introduced to the failover node with ALL of it's necesary info:

take note of the VG major number and a pvname (from lspv) belonging to the vg. This should come from your host that can successfully varyon the vg.

on source node:

ha stopped
verify resource group fs unmounted
varyoffvg
do not export

on the target node:

ha stopped
exportvg (get rid of out of synch info)
***importvg WITH the major number and the pvname from source node. check man/docs
chvg -an (no auto-mount) -Qn (no quoroum) vgname
verify you can varyon vg
if successful varyoff the vg and start ha servies.


hope this helps...i think the process is explained in the HACMP administration (manual changes), or HACMP troubleshooting guide.