Object pool 'ReadIndex readers pool' has reached its max limit for items

Hi everyone

We have Eventstore running as a service under systemd. Our Reader-Queue is quite high while scavenging or while a projection has to catch up.
When we set the ReaderThreadsCount over 8, there will be a lot of log messages

Error during processing ReadStreamEventsBackward request.

Object pool ‘ReadIndex readers pool’ has reached its max limit for items: 14.

After a while Eventstore will crash with following exception:

46086036-9945fc80-c1a7-11e8-935d-aa2b830dc65c.png (1142×496)

All while the system is at 40% CPU and at max at 50% of IO-bytes.

How can we increase the limit of this 14 pool? Something with MONO_THREADS_PER_CPU?

Thank you very much in advance

I created a pull request. Can someone look into that?



We were also hit by this recently, resulting in nodes falling over and projections stopping with faults (they recovered when things were running again and prompted to start, but during the “catch up” phase, we were seeing more of the same error again, presumably as projections were putting more load on the system).

We’ve adjusted the number of reader threads down in response, but would love to see this PR merged.