Query application status API triggers re-deployment?

Ray Serve
1

1. Severity of the issue:
Medium: Significantly affects my productivity but can find a workaround.

2. Environment:

  • Ray version: 2.45.0
  • Python version: 3.9.22
  • OS: ubuntu 20.04
  • Cloud/Infrastructure: GCP
  • Other libs/tools (if relevant):

3. What happened vs. what you expected:

  • Expected: Query application status through Python/REST API should not trigger any modification to the current deployment
  • Actual: Query application status through Python/REST API produce "Deploying new version of Deployment(name=‘xxx’, app=‘xxx’) " log

Github Issue: [Serve] Query application status API triggers re-deployment · Issue #53137 · ray-project/ray · GitHub

What happened + What you expected to happen

I am trying to deploy my application using Ray Serve and the application requires some local modules which are not in the working directory, so I pass them to runtime_env.py_modules.

But when I try to query the application status using either Python API serve.status() or REST API /api/serve/applications/, I notice there are always one log for EACH application:

INFO 2025-05-19 20:01:02,637 controller 14198 -- Deploying new version of Deployment(name='xxx', app='xxx') (initial target replicas: 1).

If I remove the py_modules field the log will not show when I run the query.

I check the return results from each query, and they are the same. So I suppose the controller does a try run of the deployment?

Reproduction script

from fastapi import FastAPI, Query, HTTPException
from ray import serve
import ray

app = FastAPI()

@serve.deployment(
    name="hello_world",
)
@serve.ingress(app)
class HelloWorld:
    @app.get("/hello")
    async def hello(self):
        return {"message": "Hello, world!"}
    

entrypoint = HelloWorld.bind()

ray.init(runtime_env={
        "working_dir": "./",
        "py_modules": ["https://storage.googleapis.com/kscale-text2img-app/deploy.zip"] # any other local dir/module have the same effect
    })

serve.run(entrypoint, name="hello_world")

# for local test only
import time
while True:
    time.sleep(3600)

And then run curl http://localhost:8265/api/serve/applications/, you will always get a log:

INFO 2025-05-19 19:54:40,388 controller 12939 -- Deploying new version of Deployment(name='hello_world', app='hello_world') (initial target replicas: 1).
2

Hey, thanks a lot for filing this report! You actually discovered a very old bug, I put up a fix in [serve] Fix runtime_env validation for py_modules by pcmoritz · Pull Request #53186 · ray-project/ray · GitHub

1 Like