[Remote Store] Redefine the insync allocations if remote translog is enabled #4506
Description
Is your feature request related to a problem? Please describe.
Whenever a primary fails and if replicas are available. The master choses to promote a replica that is insync (or active). Since the request to primary is not replicated in sync to the replicas. The definition of insync itself needs to be changed.
Couple of definitions in respect to a shard (in replication tracker) -
- inSync = whether this shard is treated as in-sync and thus contributes to the global checkpoint calculation
- tracked = whether this shard is tracked in the replication group, i.e., should receive document updates from the primary
- pendingInSync = This set contains allocation IDs for which there is a thread actively waiting for the local checkpoint to advance to at least the current global checkpoint
Describe the solution you'd like
all replicas in general can be assumed to be insync (need to find cases where it should not be not relating to checkpoints). This will make the rest of the updates on replica’s replication tracker as obsolete and then we can go on to the next step.
Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.
Additional context
Add any other context or screenshots about the feature request here.