feat: Add streaming tool-call parse buffer limit to prevent excessive memory usage#8811
Open
pskiran1 wants to merge 15 commits into
Open
Conversation
whoisj
previously approved these changes
Jun 4, 2026
| self.chat_template = load_chat_template(chat_template) | ||
|
|
||
| if self.tool_call_parser is not None: | ||
| print( |
Contributor
There was a problem hiding this comment.
interesting, why print("[INFO] ...") and use logger.info()?
Member
Author
There was a problem hiding this comment.
Currently, in openai_frontend, the root logger is not configured in main, as a result logging does not appear to be working. So, I have been using print statements, similar to the approach in fastapi_frontend.py.
Contributor
There was a problem hiding this comment.
You mean following lines doesn't work at all?
server/python/openai/openai_frontend/engine/triton_engine.py
Lines 757 to 763 in afee02f
| and len(previous_text) + len(delta_text) | ||
| > self.max_tool_call_parse_bytes | ||
| ): | ||
| print( |
Contributor
|
Docs? |
yinggeh
reviewed
Jun 5, 2026
| # streaming tool-call parser processes per request. | ||
| # Since the parser re-parses the entire buffer with each new chunk, | ||
| # this limit helps bound per-request CPU and memory usage. | ||
| DEFAULT_MAX_TOOL_CALL_PARSE_BYTES: int = 16 * 1024 |
Contributor
There was a problem hiding this comment.
Why 16 KiB? Can this limit be bigger?
yinggeh
reviewed
Jun 5, 2026
yinggeh
reviewed
Jun 5, 2026
…tils.py Co-authored-by: Yingge He <157551214+yinggeh@users.noreply.github.com>
…oolpparsing-can-oom-kill' of https://github.com/triton-inference-server/server into spolisetty/tri-1016-psirt-triton-openai-frontend-auto-toolpparsing-can-oom-kill
…tend-auto-toolpparsing-can-oom-kill
Member
Author
Sorry, I missed committing the README changes earlier. The documentation has now been updated. |
…tend-auto-toolpparsing-can-oom-kill
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
3 participants
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.
What does the PR do?
The streaming tool-call parser (
partial_json_parser.loads()) re-parses the full accumulated output on every chunk, resulting in excessive CPU and memory growth for large tool-call arguments. This PR adds a configurable per-request buffer cap--max-tool-call-parse-bytesthat truncates the stream gracefully when exceeded.Checklist
<commit_type>: <Title>Commit Type:
Check the conventional commit type
box here and add the label to the github PR.
Related PRs:
Where should the reviewer start?
Test plan:
Caveats:
Background
Related Issues: (use one of the action keywords Closes / Fixes / Resolves / Relates to)