How parallelism works in KubernetesExecutor?

[MaksymSofer]

Hello,
I use Airflow 2.10.1 in GKE, parallelism is set to 32
i have three dags that can bring (even on theoretical level only 12 simultaneous tasks).
but i see constant issue with Scheduler going into
Executor parallelism limit reached. 0 open slots.

Checking executor code, i see that open_slots defined as:

open_slots = self.parallelism - len(self.running)

I spotted that KubernetesExecutor leave "Succeeded" pods - is it possible that they calculated into "running" amount?

Please help me.

[MaksymSofer]

I enabled sending of traces to tempo and now i see that numbers actually count succeded pods

this correspond to tempo saved trace:

[MaksymSofer]

Scheduler again went into "no open slots" state and cluster overview (from airflow ui) show this:

while only two running tasks are:

which 18 tasks scheduler see with "no state" i dont know,,,