Docker network error when running uv run python run_swelancer.py

[bingyuyap]

1. I had to update my default address pool for Docker in order and make sure that there are enough subnets for the LocalClusters

RuntimeError: Too many docker networks have been created. Most machines/laptops default to allowing
16 LocalCluster instances.

2. Running uv run python run_swelancer.py after gives after network related error like:

APIError: 409 Client Error for
http+docker://localhost/v1.47/containers/b82a01e75fc56dd741fc6a9160d7ed3cd1c2d2b4b2314cd6e4d6a36fe44
cf9c0/exec: Conflict ("container b82a01e75fc56dd741fc6a9160d7ed3cd1c2d2b4b2314cd6e4d6a36fe44cf9c0 is
not running")

and

APIError: 403 Client Error for
http+docker://localhost/v1.47/networks/6754a0439c1a914d5c6430e5feb3939ec6241db72f590974d8c4359f98ff6
5ea: Forbidden ("error while removing network: network
tinydockernet-alcatraz-10b58905-e743-4315-a9f0-8bcca776063f id
6754a0439c1a914d5c6430e5feb3939ec6241db72f590974d8c4359f98ff65ea has active endpoints")

Error: 403 Client Error for http+docker://localhost/v1.47/networks/cccc2e44b63756397009bc43756083b1d89593fbba1caa5b91dde15c3854a15c: Forbidden ("error while removing network: network tinydockernet-alcatraz-5b653b9c-a749-4b18-88af-ff72b74eace4 id cccc2e44b63756397009bc43756083b1d89593fbba1caa5b91dde15c3854a15c has active endpoints")

Using Macbook M3 Max and Orbstack

[leon-openai]

Could you try lowering your concurrency below 16, and seeing if that helps?

[bingyuyap]

@leon-openai i think that would help the Too many docker networks created but putting concurrency = 1 still gives me this error:

TypeError: StopIteration interacts badly with generators and cannot be raised into a Future

[leon-openai]

Based on the error they're getting, if they're trying to run everything they should check out our documentation about Running at Scale! Our guidance to folks trying to run SWELancer is to check out the README first and then reach out for support.

Here's the writeup to running at scale, if helpful

[dchichkov]

The issue is that nanoeval is initializing all the dockers upfront, during the init. And the concurrency is not used during the init. The relevant code is in the project/alcatraz/alcatraz/clusters/local.py.

I've tried to reconfigure/change that (to either reuse networks or to create containers lazily), but the code is heavy on asyncio/ relatively difficult to debug, its not a trivial change.