Issues with --limit-concurrency: Immediate Rejection and Inconsistent Behavior on Request Handling

[sharathkumar63]

Hi team,

I'm experiencing unexpected behavior with the --limit-concurrency option in Uvicorn and would like clarification or guidance on whether this is expected or a bug.

🔸 Setup:
Uvicorn run with: --workers 1 --limit-concurrency 1

Requests sent directly to Uvicorn

Issue 1: Immediate 503 Rejection on First Request
When I set --limit-concurrency=1, I expected Uvicorn to process one request at a time and queue or delay others.
However, the very first request itself gets immediately rejected with a 503 Service Unavailable, which shouldn't happen since the concurrency limit has not been breached yet.

Issue 2: Request Rejected After Previous Request Completes
With --limit-concurrency=2, the first request is processed as expected, but subsequent requests get rejected only after the first request finishes.
This behavior feels inconsistent: if the concurrency limit is already exceeded, the request should be rejected immediately—not after the previous one completes.

🔹 Expectation:
Requests beyond the concurrency limit should be either:

Immediately rejected (503)

Rejections should not happen after request completion; that feels like wasted wait time.

❓ Question:
Is this the expected behavior of --limit-concurrency?

I'm encountering multiple issues with this and would appreciate any insights or suggestions.

Thanks in advance!