You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When a replica has appendonly yes it will first load any AOF found in disk, then connect to master and load a new dataset again.
There's no chance of PSYNC from an AOF, so the first step seems unnecessary.
Description of the feature
On startup, only load AOF from disk if we can't SYNC to master.
At least while the feature of restarting from AOF + PSYNC is not available.
Alternatives you've considered
Implement PSYNC for restarting from AOF (on both master and replica restart).
Additional information
If this is not acceptable, we could consider for the design of #9796 to have a way of knowing in advance whether a PSYNC will be possible, then skip loading AOF from disk when both are true:
No necessary information stored for PSYNC or dataset is too old
We managed to connect to master.
By that, I mean storing separately the information necessary for PSYNC, so we reduce chances of loading whole AOF unnecessarily and right after having to FULL SYNC from master.
Note: assuming that #9796 would be implemented for replicas as well
The text was updated successfully, but these errors were encountered:
The problem/use-case that the feature addresses
When a replica has
appendonly yes
it will first load any AOF found in disk, then connect to master and load a new dataset again.There's no chance of PSYNC from an AOF, so the first step seems unnecessary.
Description of the feature
On startup, only load AOF from disk if we can't SYNC to master.
At least while the feature of restarting from AOF + PSYNC is not available.
Alternatives you've considered
Implement PSYNC for restarting from AOF (on both master and replica restart).
Additional information
If this is not acceptable, we could consider for the design of #9796 to have a way of knowing in advance whether a PSYNC will be possible, then skip loading AOF from disk when both are true:
By that, I mean storing separately the information necessary for PSYNC, so we reduce chances of loading whole AOF unnecessarily and right after having to FULL SYNC from master.
Note: assuming that #9796 would be implemented for replicas as well
The text was updated successfully, but these errors were encountered: