Not handling growing concurrency past init_count very well

[MaxPower15]

Hi there, I'm using Riak in a system that potentially requires high concurrency and have run into an exception that's hard to handle. It looks like this:

21:03:45.777 [error] Process #PID<0.192.0> raised an exception
** (FunctionClauseError) no function clause matching in Riak.find/4
    (riak) lib/riak.ex:301: Riak.find(:error_no_members, :error_no_members, "abc", "def")
    (riak) lib/riak.ex:301: Riak.find/3
    (riak) lib/riak.ex:282: Riak.find/2

I set up a test repo with instructions to demonstrate: https://github.com/MaxPower15/riak_pooler_issue

I poked around myself with pooler, hoping to maybe get a PR fix, but I don't think I'm familiar enough with the underlying code to really do it correctly. I hope this demo helps though!

[edgurgel]

@MaxPower15 I would simply increase your pool https://github.com/drewkerrigan/riak-elixir-client#connection-pooling

Maybe each function should give a better error but it's fair to explode if no connection from the pool is ready to deal with the request.

[MaxPower15]

@edgurgel Yes, that's my current solution; I talk about it here: https://github.com/MaxPower15/riak_pooler_issue#explanation. I just think it should be handled better out of the box. :)

[MaxPower15]

I should mention, this occurs when requesting a connection when well below the max_count. It just needs to be above the previous minimum value, which is init_count in the demo.

[edgurgel]

Oh gotcha! I should have read the whole thing! Thanks for the report!

[edgurgel]

This issue seems to be the same you noticed: epgsql/pooler#54

[MaxPower15]

Aha, so perhaps a pooler issue. I wonder if it's introduced in a new version. Will try with an older one to see if it can still be reproduced.

[MaxPower15]

Well, so much for that idea. I couldn't compile pooler pre-1.5.0. I guess the question is if that's intended behavior or not with pooler.

[randysecrist]

Definitely think this is a pooler issue. I've done some house keeping with pooler recently and fixed some of the compilation issues and warnings including pushing some things up to hex.

Will close this, but will also mention this pooler PR. Might be worth considering pulling this in but I need to think about it more.

epgsql/pooler#72