Ran Shidlansik

Provided PR: #11245

@oranagra I think the word "threads" here is wrong. however in the specific case of some logic registered to the atexit() I think it is better to keep aligned with...

@polaris-alioth I think that the main motivation here is saving the slowlog in cases the engine crashed. Maybe we should instead just allow adding the slowlog to the exception information...

> the number of slowlogs stored in the memory is limited. and persistent storage can allow users to view full slowlogs. according to `logfile`, it can generate a separate slowlog...

I think that applications are capable to do the math externally. but in order to be more backword compatible I would suggest having all commands which have subcommands include the...

> I'm not sure i understand your suggestion. what's `cmdstat_cluster|` (the second line in commandstats output)? I meant that in order to distinguish between the aggregated values of the parent...

Hello @HanGuangyu. I work with @madolson in AWS and we are interested in this great effort to improve the Redis performance! without diving too deep in to the implementation I...

@guybe7 I went shortly through the change (will make a deeper attempt later) but do we also plan to change the dbOverwrite? Currently it will only trigger stream types. I...

@madolson just to make sure I understand the complete suggestion: I also agree that investing in pub/sub cluster memory performance is important and we can also reduce the memory overhead...

> > I think that maintaining a buffer list per channel would help reduce the memory overhead of large clusters (when sharding is not used) somewhat similar to what was...