Repository navigation
fix(core,server): keep StdioServerTransport running when a handler is cancelled #1101
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
Open
gibkim
wants to merge
2
commits into
modelcontextprotocol:main
Choose a base branch
from
gibkim:fix/1086-stdio-handler-cancellation
base: main
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
+71
−21
Open
Changes from all commits
Commits
Show all changes
2 commits
Select commit
Hold shift + click to select a range
File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
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
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
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
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
Oops, something went wrong.
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.
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.
Problem: keeping the loop alive is the right call, but it now leaves the request the cancelled handler was serving with no reply at all. In the concurrent dispatch phase
Protocol.onRequestalready turns a cancellation that escapes a live handler job into anINTERNAL_ERRORresponse (Protocol.kt:626-632, "so the peer does not hang until its own timeout"). The serial, pre-notifications/initializedphase takes thehandlerJob == nullbranch and rethrows instead — and until this change that rethrow landed here and tore the transport down, so the peer at least observed a closed connection. With the transport surviving, nothing goes out.Failure scenario: a client sends
initialize, thentools/callbefore the server has processednotifications/initialized(so dispatch is still inline), and the tool handler leaks a cancellation — e.g. an innerwithTimeoutexpiring. I ran this against the branch: the server answersinitialize, answers a later unrelatedping, and never answers thetools/call. The client blocks on that call until its own request timeout. Onmainthe same scenario closed the stream, so the client failed fast.Suggested fix: make the serial branch in
Protocol.onRequestanswer instead of rethrowing, using the same liveness check as here so a realclose()during an inline handler still propagates:Produced by Air Automations. Name: Code review / Run: https://air.jetbrains.cloud/org/05cf1a7f-6ab5-713b-abd3-29d0c8a05e2d/automations/409567ec-d3a7-40ac-ba95-8518ec577ac2?run=9ec05c83-ec7c-4a82-afff-a07cf8545630
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.
Fixed in 14c4d86, as suggested: the serial branch in
Protocol.onRequestnow callsensureActive()and answersINTERNAL_ERRORinstead of rethrowing. Added a core test for the leak (fails on the previous commit) and one that cancels the delivering coroutine and expects no response. This moves part of the fix into core, so I noted that in the PR description.