Add adaptive batching dispatch metrics#5637
Draft
RitwijParmar wants to merge 1 commit into
Draft
Conversation
Signed-off-by: Ritwij Aryan Parmar <ritwij.aryan.parmar@gmail.com>
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
This adds dispatcher-level metrics for adaptive batching so operators can see why batches are released and how much queue wait they are paying.
Today BentoML exposes batch size, but it is hard to tell whether an endpoint is filling batches, waiting on the optimizer window, or sitting with low item counts under load. This PR adds a lightweight observer on CorkDispatcher and wires it into both Service and Runner serving paths.
New Prometheus signals include dispatch reason, queued jobs at release time, item count after payload-size aware splitting, and oldest request queue delay. The existing batch-size metric is kept and moved to the same dispatch observer so it is recorded from one place.
Validation
python3 -m ruff check src/bentoml/_internal/marshal/dispatcher.py src/bentoml/_internal/server/runner_app.py src/_bentoml_impl/server/app.py tests/unit/_internal/test_dispatcher_metrics.pypython3 -m compileall -q src/bentoml/_internal/marshal/dispatcher.py src/bentoml/_internal/server/runner_app.py src/_bentoml_impl/server/app.py tests/unit/_internal/test_dispatcher_metrics.pygit diff --check/Users/ritwij/.cache/codex-runtimes/codex-primary-runtime/dependencies/python/bin/python3 -m pytest tests/unit/_internal/test_dispatcher_metrics.py -q