Greetings,
we have recently installed ovirt as a hosted-engine with high availability on
six nodes over NFS storage (no Gluster), with power management through an
on-board IPMI device, and the setup was successful. All the nodes (from
Supermicro) are identical in every aspect, so no hardware differences exist and
no modifications to the servers' hardware were performed. The hosted-engine was
deployed on a second host, where two of the six hosts only were required to
host the HE VM.
The network interface on each node is bonded between two physical fiber optics
NICs in LACP mode with a VLAN on top, serving as the sole networking interface
for the server/node, no separate VM or storage networks were needed, as the
host OS, hosted-engine vm, and storage are required to be on the same network
and VLAN.
We started by testing the high-availability of the hosted-engine VM (as it was
deployed on two of the six nodes) by rebooting or powering off one of the
hosts, and the VM would migrate successfully to the second HE node. The main
goal of our experiments is to test the robustness of the setup, as it is
required for the cluster to remain functional even when up to two hosts are
brought down (whether due to a network or power issue), however, when rebooting
or powering off one of the hosts, the HE VM goes down and takes the entire
cluster with it, where we can't even access the web portal. Once the host is
rebooted, the HE VM and the cluster becomes functional again. Sometimes the HE
VM stays down for a set amount of time (5 to 6 minutes) and then goes back up,
and sometimes it goes down until the problematic host is back up. This behavior
happens to other VMs as well not the the HE.
We suspected an issue with the NFS storage, however, during ovirt operation it
is being mounted properly over /rhev/data-center/mnt/<nfs:directory>, while the
expected behavior is for the cluster to stay operational and any other VMs to
be migrated to other hosts. During one of the tests, we tried to mount the NFS
storage on a different directory and there was no problem, we were even able to
perform commands such as ls without any issues, as well as writing a text file
at the directory's root, and be able to modify it normally.
We suspected a couple of things the first being that the HE is unable to fence
the problematic host (the one we took down), however, power management is setup
properly.
The other thing we suspected is the cluster hosts (after taking down one of
them) are unable to acquire storage lease, which is weird since the host in
question is down and non-operational, hence no locks should be in place. The
reason behind this suspicion is the following two errors that we receive
frequently when one host or more goes down from the
engine\ovirt-engine\engine.log file:
1- "EVENT_ID: VM_DOWN_ERROR(119), VM HostedEngine is down with error. Exit
message: resource busy: Failed to acquire lock: Lease is held by another host."
2- "[<id>] Command 'GetVmLeaseInfoVDSCommand(
VmLeaseVDSParameters:{expectedEngineErrors='[NoSuchVmLeaseOnDomain]',
storagePoolId='<pool-id>', ignoreFailoverLimit='false', leaseId='<lease-id>',
storageDomainId='<domain-id>'})' execution failed: IRSGenericException:
IRSErrorException: No such lease: 'lease=<lease-id>'"
This is a third warning from the /var/log/vdsm/vdsm.log file
1- "WARN (check/loop) [storage.check] Checker
'/rhev/data-center/mnt/<nfs-domain:/directory>/<id>/dom_md/metadata' is blocked
for 310.00 seconds (check:265)"
All the tests are done without setting nodes into maintenance mode as we are
simulating an emergency situation. No HE configuration were modified via the
config-engine command, the default values are used.
Is this a normal behavior? Are we missing something? Do we need to tweak a
certain configuration using the config-engine command to get a better behavior
(e.g., shorter down period)?
Best regards
_______________________________________________
Users mailing list -- [email protected]
To unsubscribe send an email to [email protected]
Privacy Statement: https://www.ovirt.org/privacy-policy.html
oVirt Code of Conduct:
https://www.ovirt.org/community/about/community-guidelines/
List Archives:
https://lists.ovirt.org/archives/list/[email protected]/message/MUSSKYJMVEIMM5CT7WEPKBZW6Y7XZFRN/
