[FEATURE] Ability to manually specify number of threads

[darrynten]

First off, amazing project, absolutely mind blowingly cool.

Problem

Restricting the number of CPUs a container can have access to does not get passed along to the app.

If I limit the container to 2 out of 8 available cores the worker still tries to spawn 8 threads, and 6 of them error with the following message:

2026-02-05 08:41:02.339811204 [E:onnxruntime:Default, env.cc:234 ThreadMain] pthread_setaffinity_np failed for thread: 232, index: 0, mask: {1, }, error code: 22 error msg: Invalid argument. Specify the number of threads explicitly so the affinity is not set.

It would be great if we could specify a desired limit on the number of threads spawned so that we can be more in control of how many cores we dedicate to the audiomuse worker.

[NeptuneHub]

Hi and thanks for the feedback. So you have the last version of AudioMuse-AI? And you’re testing where? Like Kubernetes? Can you show how you set the CPU?

Anyway actually I leave onnx determine the correct number of thread during the Analysis and this till now is a more stable way than just give a fixed number. But off course I can try to give a look.

[Coloradohusky]

I'm also having this issue - I'm running mine in a Docker container within a Proxmox LXC, I gave the LXC 4 cores out of a possible 16.

[lucidfoxes]

Same issue for me running in a Docker container. The system has an i5-13500 with 64GB RAM, and I'm trying to restrict the number of cores and RAM used, so other containers can work normally and the server remains stable. I tried to do this by using these settings for the worker container -

cpuset: "0-7"
mem_limit: 16g
memswap_limit: 16g

But I get a lot of these errors in the logs -

[E:onnxruntime:Default, env.cc:234 ThreadMain] pthread_setaffinity_np failed for thread: 1787, index: 4, mask: {10, 11, }, error code: 22 error msg: Invalid argument. Specify the number of threads explicitly so the affinity is not set.

[lucidfoxes]

I'm not sure if this is the right approach, but I managed to avoid these errors by changing the variables in the docker compose to -

CLAP_PYTHON_MULTITHREADS: "true"

and

cpu_period: 100000 cpu_quota: 800000 mem_limit: 16g memswap_limit: 16g

[scottia]

These are my results regarding CPU use on my Docker / LXC

Worker is configured with:
cpus: "4.0" (host has 10 cores available)
mem_limit: 16g (host has 32g)
memswap_limit: 16g
CLAP_PYTHON_MULTITHREADS: "true"
no cpuset

The worker still logs repeated pthread_setaffinity_np failed errors from ONNX Runtime and causes recurring load spikes during analysis.

I don't think this is limited to cpuset usage as still a problem with simple CPU cap
Environment:
AudioMuse v0.9.3
Docker in LXC
Redis + Postgres external
worker using SERVICE_TYPE=worker

RESOLVED :
Prox 8.4.17 /
Resolved setting /etc/pve/lxc xxx.conf to total host cores vs. LXC cores assigned of 10.
lxc.cgroup2.cpuset.cpus: 0-19
No more affinity errors and better performance.