Blocker Description
A cluster node lease is renewed periodically to indicate that a given clusterId is in use. This happens twice the async delay before the lease expires. With the default configuration this means the lease is renewed two seconds before it expires. Load test with an oak cluster shows that the lease sometimes is not renewed in time and the LastRevRecoveryAgent kicks in even though the cluster nodes are still running.
I see a couple of options to fix this:
- Renew the lease earlier. E.g. after half of the lease timeout, or even with every background operation run. This would be roughly once a second.
- Decouple the lease renew from the background operation thread where also more expensive operations are executed (_lastRev updates and cache invalidation).