net/http: count a lost HTTP/2 ping exactly once, and never on a failed connection - #80923
Open
ali-sayyah wants to merge 1 commit into
Open
net/http: count a lost HTTP/2 ping exactly once, and never on a failed connection#80923ali-sayyah wants to merge 1 commit into
ali-sayyah wants to merge 1 commit into
Conversation
|
Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA). View this failed invocation of the CLA check for more information. For the most up to date status, view the checks section at the bottom of the pull request. |
…d connection
The HTTP/2 health-check timer is re-armed after every ReadFrame
return, including the final erroring one, and no close path stops it:
one SendPingTimeout after every connection close a post-mortem health
check pings the dead connection, fails immediately, and reports a
spurious CountError("conn_close_lost_ping"). A genuinely lost ping is
counted twice: the real detection, then the post-close echo.
Stop the timer when the read loop exits, publish the read loop's
terminal exit under cc.mu before the failure is counted, and make
lost-ping classification a single-claim close: the eligibility check
and the claim share one critical section in closeForLostPing, so
overlapping health checks, or a ping racing the read loop's terminal
exit, can never produce a second count. The CountError callback runs
outside the lock. A ping that fails on a connection that is live at
claim time is still counted and still closes the connection,
preserving write-blocked-ping detection.
The equivalent x/net change (golang/net#262) fixes the legacy
transport used on Go versions before 1.27 and with the http2legacy
build tag.
Fixes golang#80920
ali-sayyah
force-pushed
the
http2-close-healthcheck
branch
from
August 17, 2026 20:14
957dad6 to
7458cda
Compare
ali-sayyah
changed the title
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.
The HTTP/2 health-check timer is re-armed after every ReadFrame
return, including the final erroring one, and no close path stops it:
one SendPingTimeout after every connection close a post-mortem health
check pings the dead connection, fails immediately, and reports a
spurious CountError("conn_close_lost_ping"). A genuinely lost ping is
counted twice: the real detection, then the post-close echo.
Stop the timer when the read loop exits, publish the read loop's
terminal exit under cc.mu before the failure is counted, and make
lost-ping classification a single-claim close: the eligibility check
and the claim share one critical section in closeForLostPing, so
overlapping health checks, or a ping racing the read loop's terminal
exit, can never produce a second count. The CountError callback runs
outside the lock. A ping that fails on a connection that is live at
claim time is still counted and still closes the connection,
preserving write-blocked-ping detection.
The equivalent x/net change (golang/net#262) fixes the legacy
transport used on Go versions before 1.27 and with the http2legacy
build tag.
Fixes #80920