-
Notifications
You must be signed in to change notification settings - Fork 50
Description
Is your feature request related to a problem? Please describe.
I'm always frustrated when it takes about a month to synchronize a cluster. Policer has some fixed rate of objects per cycle.
Describe the solution you'd like
Add some "boost" mode when policer detects (and recovers) some missing replicas. Normally it'll run at its ordinary speed, but when recovery is made it'll try to pick/check 3× or 5× (configurable) number of objects for the next cycle. This way if we have some nodes added/removed it'll run faster until all replicas are settled in place. Same for node outages and similar events.
Describe alternatives you've considered
The mechanics can be more complex, it can theoretically react to netmap change event as well, but this required additional external data while replica recovery is what policer knows anyway and this can be even more informative.
Additional context
It's hard to recover mainnet nodes with ~40M objects.