Karl Nilsson

> @kjnilsson: Do you mind if I wait a real use case before merging this? I mean, when you have a feature flag which would benefit from this, then we...

I don't have any feature flags apart from a dummy one so no. Ok I will take a look

Needs https://github.com/rabbitmq/rabbitmq-server/pull/5262

Looks like the second channel isn't detecting the deletion of the classic queue, either quickly enough or there is a bug causing it not to detect it at all. This...

Quorum queues recover in parallel already (as you noted as multiple queues make better use of the disk) so I am not sure what this request is for. When recovering...

Do you have any profiling data to suggest that recovering the segment references is what is taking time during queue recovery? if anything it would be reading the log from...

`ra_log_segment:open` does not read the whole file. It reads and recovers the fixed size segment index that resides at the start of the file. Later during state machine recovery it...

in particular the call to `filename:absname/1` is very slow and possibly not even needed.

This might help a bit: https://github.com/rabbitmq/ra/pull/222 It won't be a night and day difference but may improve certain scenarios

https://github.com/rabbitmq/ra/pull/223/ This is the PR but I am still experimenting quite heavily with it. for my test case of a 1M message QQ the recovery time went from 1 minute...