maxPending limit is enforced with some de-facto leniency under heavy load

[simonbasle]

Follow-up to #121 and #122.

With #122, the situation has improved and maxPending cap should be enforced with some leniency/margin of error. It might accumulate pending acquire calls over the configured limit, but within acceptable bounds (under heavy load). Previously, under heavy load it would grow past the limit indefinitely.

However, ideally the pool should manage to strictly enforce the maxPending limit.
This issue tracks that goal.

[simonbasle]

might be resolved by #155

[NeilOnasch0402]

As stated in reactor/reactor-netty#2261 the bug does still exist in reactor-core:0.2.8 / reactor-netty-core:1.0.19

Is there anything we can provide to help with this issue?

[simonbasle]

there is no easy way forward that I could see after having fixed the case where it would grow indefinitely. so with the current implementation I thought that the compromise would work out well as the amount of pending borrower going over the configured max should be limited...

in your case, it seems to be thousands despite a max at 50, which is definitely not okay. have you seen any improvements with these numbers when using 0.2.8 ?

[schreibaer]

Hi @simonbasle, I work with @NeilOnasch0402.

Even though we did upgrade our springboot version to 2.6.8 in an attempt to mitigate the problem, we were using 2.6.6 when this problem presented itself. Springboot 2.6.6 already uses reactor-pool 0.2.8 so unfortunately, the version does not seem to fix the issue in our case.