-
Notifications
You must be signed in to change notification settings - Fork 6.9k
[Serve.llm][P/D] Fix health check in prefill disagg #53937
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[Serve.llm][P/D] Fix health check in prefill disagg #53937
Conversation
|
|
||
| self.response_postprocessor = ResponsePostprocessor() | ||
|
|
||
| @property |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This is useless. I don't know why we had it. removing.
|
|
||
| try: | ||
| return await asyncio.wait_for(self.engine.check_health(), timeout=15) | ||
| await asyncio.wait_for(self.engine.check_health(), timeout=15) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
removed the timeout time since ray serve has an adjustable timeout per deployment anyways.
| ): | ||
| yield chunk | ||
|
|
||
| async def check_health(self) -> None: |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
These must be removed. In general the health check of a deployment is not bounded to the health check of its child deployments.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
cc @lk-chen fyi
|
|
||
| async def check_health(self): | ||
| await self._init_completed.wait() | ||
| await asyncio.gather( |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
same thing applies here.
| model="Qwen/Qwen2.5-0.5B-Instruct", | ||
| dtype="auto", | ||
| disable_log_stats=False, | ||
| enforce_eager=True, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Making the tests faster
|
|
||
| await router.check_health() | ||
|
|
||
| assert server.check_health.remote.call_count == 1 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
testing router's health check has nothing to do with server's health check.
Signed-off-by: Kourosh Hakhamaneshi <kourosh@anyscale.com>
Signed-off-by: Kourosh Hakhamaneshi <kourosh@anyscale.com> Signed-off-by: elliot-barn <elliot.barnwell@anyscale.com>
Original PR #53937 by kouroshHakha Original: ray-project/ray#53937
…isagg Merged from original PR #53937 Original: ray-project/ray#53937
Also needs vllm-project/vllm#19821 to be merged.