Feb 17, 2011, 03:40 AM
PVA doesn't show all VEs on PVC 4.0 node
Just one more PVA issue remaining I hope :)
We have upgraded the PVA on our old PVC 4.0 HN to 4.6 by following the instructions in KB 9627. The PVA MN can succesfully connect to the HN, but somehow it doesn't show all containers.
When I do a vzlist -a on the HN it shows containers: 117, 121, 129, 137, 139 and 143. But the PVA overview only shows 117, 121, 137 and 143. So there are two missing. These used to show up in the old PIM installation on the hardware node, but that of course has been deleted since.
I'm really puzzled by this. Unregistering and re-registering the HN in PVA does not help. Is there a cache I can try to clean somewhere in the management node? For what it's worth, Power Panel does work for the missing containers.
Feb 17, 2011, 04:39 AM
Ugh.. and on top of that vzagent won't start on the old hardware node anymore. So we cannot manage backups for the missing containers using vzabackup/vzarestore.
[root@vps1 ~]# vzagent_ctl start
Starting vzagent: //opt/vzagent//bin/vzagent: symbol lookup error: //opt/vzagent//lib/libVZABasicFunctionalityLocal.so: undefined symbol: _ZN3VZL16VZLPipePrototype4postEPNS_10VZLMessageEN5 boost13intrusive_ptrINS_23VZLTaskHandlerPrototypeE EEPSsb
Feb 21, 2011, 06:52 AM
Ok so I've migrated one of the 'invisible' containers to a new HN (brand new, completely up-to-date) using the vzmigrate command-line tool. But I still have the same issue there: the VE shows up when running 'vzlist' in a shell, but it cannot be seen in PVA, while that does show other containers on the same HN.
I did notice that the other missing container on the old HN now does show up in PVA all of a sudden.. so some kind of refresh was triggered after all. How do I simulate this on our new HN?
Feb 21, 2011, 07:16 AM
I did notice this alert on the HN with the missing VE in PVA:
Virtuozzo group problem with <HN>. Subsystem: server_group_connection failed, reason can't update env config/status list, envcache unavailable [code -1].
Feb 21, 2011, 10:58 AM
Ok.. turns out the VE in question had an IPv6 address (which was added using vzctl) and the PVA MN tried to store this in it's database as a longint.. which failed miserably. So miserably that it completely stopped the agents further indexing for that node (which is why we had 2 invisible containers earlier).