Reduces the memory requirement of a standalone Prefect server again #16651
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
In #16644, we were able to get the standalone Prefect server we run in `prefect server start` from 400MB to 343MB. By running `uvicorn` directly in the parent process, we further reduce that to *266MB*. I measured this using the same process as before, using a `cgroup` with no swap to constrain the memory until I found the minimum. There are two main changes here: 1. Making `prefect server start` a synchronous command in both the background and the foreground cases. This helps to simplify the handling of signals and removes the need for `anyio` or piping the streams of a child process manually. 2. Splitting the paths between background and foreground servers. In the case of the background server, we run the subprocess as usual and then exit leaving it running. In the case of the foreground server, we run it with `uvicorn.run` as the final call, letting `uvicorn` handle signals. `uvicorn` already handles `SIGINT` and `SIGTERM` to cleanly shutdown, so we don't need to do anything additional here. I'm removing the two tests that were testing what happens when the signals are sent in escalating succession, as they aren't materially different now. In both cases, `uvicorn` will attempt to shutdown cleanly. I also updated the tests to reflect that `SIGINT` will lead to a successful exit, while `SIGTERM` will lead to a -15 exit (pretty standard).
chrisguidry
requested review from
cicdw,
desertaxle and
zzstoatzz
as code owners
CodSpeed Performance ReportMerging #16651 will not alter performanceComparing Summary
|
chrisguidry
commented
Jan 8, 2025
| from prefect.server.api.server import create_app | ||
|
|
||
| with temporary_settings( | ||
| {getattr(prefect.settings, k): v for k, v in server_settings.items()} |
There was a problem hiding this comment.
Is this cheeseball? Should I just explicitly set the Prefect settings here? I couldn't decide...
There was a problem hiding this comment.
This seems reasonable to me, considering the current state of settings. We need to update temporary_settings to be more ergonomic now we're using pydantic settings, but this is the right way to do it now.
chrisguidry
commented
Jan 8, 2025
| server_env["PREFECT_API_SERVICES_SCHEDULER_ENABLED"] = str(scheduler) | ||
| server_env["PREFECT_SERVER_ANALYTICS_ENABLED"] = str(analytics) | ||
| server_env["PREFECT_API_SERVICES_LATE_RUNS_ENABLED"] = str(late_runs) | ||
| server_env["PREFECT_API_SERVICES_UI"] = str(ui) |
There was a problem hiding this comment.
While doing this, I discovered that PREFECT_API_SERVICES_UI isn't a thing, so I removed it. Not sure if was supposed to be another setting or just an old variable?
There was a problem hiding this comment.
Yeah, PREFECT_UI_ENABLED is the correct one, which also there, so this is good to remove.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In #16644, we were able to get the standalone Prefect server we run in
prefect server startfrom 400MB to 343MB. By runninguvicorndirectly in the parent process, we further reduce that to 266MB. I
measured this using the same process as before, using a
cgroupwithno swap to constrain the memory until I found the minimum.
There are two main changes here:
Making
prefect server starta synchronous command in both thebackground and the foreground cases. This helps to simplify the
handling of signals and removes the need for
anyioor piping thestreams of a child process manually.
Splitting the paths between background and foreground servers. In
the case of the background server, we run the subprocess as usual and
then exit leaving it running. In the case of the foreground server,
we run it with
uvicorn.runas the final call, lettinguvicornhandle signals.
uvicornalready handlesSIGINTandSIGTERMtocleanly shutdown, so we don't need to do anything additional here.
I'm removing the two tests that were testing what happens when the
signals are sent in escalating succession, as they aren't materially
different now. In both cases,
uvicornwill attempt to shutdowncleanly. I also updated the tests to reflect that
SIGINTwill lead toa successful exit, while
SIGTERMwill lead to a -15 exit (prettystandard).
Checklist
<link to issue>"mint.json.