fix: serialize Dial/Listen queue handoff to prevent router token loss#573

Open

raballew wants to merge 24 commits intojumpstarter-dev:mainfrom

raballew:fix-listen-queue-race

Commits on Apr 17, 2026

fix: use CompareAndDelete for race-free listenQueue cleanup

raballew
and
claude
committed
fix: wrap listenQueue channel in struct to prevent stale cleanup race
raballew
committed
fix: retry CompareAndSwap on failure in concurrent Listen() reconnect
raballew
committed
fix: give each Listen() goroutine its own channel to prevent stale reader race
raballew
committed
fix: reject Dial send to superseded listenQueue via done channel pre-check

raballew
and
claude
committed
fix: close done channel on all Listen exit paths using sync.Once

raballew
and
claude
committed
fix: swap defer order to close done channel before map deletion
raballew
committed
Improve test quality for listen queue race fix

raballew
and
claude
committed
fix: serialize Dial/Listen queue handoff with per-lease mutex

raballew
and
claude
committed
fix: pass Dial context to sendToListener to prevent mutex deadlock under backpressure

raballew
and
claude
committed
refactor: replace goto labels with drainChannel helper in test
raballew
committed
fix: clean up leaseLocks entries when Listen exits without replacement
raballew
committed
test: add end-to-end deadlock chain test
raballew
committed
ci: enable race detector in controller test target
raballew
committed
fix: resolve golangci-lint goconst and gofmt violations

raballew
and
claude
committed
fix: use sendToListener in concurrent dial reconnection test
raballew
committed
fix: use service methods in listen queue tests instead of raw sync.Map ops
raballew
committed
fix: serialize closeDone under lease mutex and stop deleting leaseLocks entries
raballew
committed
fix: use non-blocking send in sendToListener to prevent mutex starvation

raballew
and
claude
committed
fix: replace raw lease mutex with ref-counted leaseLock to enable safe cleanup

raballew
and
claude
committed
test: fix race in concurrent lease lock test and add overlapping listener test

raballew
and
claude
committed
ci: add E2E stress test workflow for flakiness detection

raballew
and
claude
committed
Revert "ci: add E2E stress test workflow for flakiness detection"
raballew
committed
fix: drain buffered tokens on Listen supersession and assert gRPC status codes

raballew
and
claude
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix: serialize Dial/Listen queue handoff to prevent router token loss#573

fix: serialize Dial/Listen queue handoff to prevent router token loss#573
raballew wants to merge 24 commits intojumpstarter-dev:mainfrom
raballew:fix-listen-queue-race

Commits on Apr 17, 2026

fix: use CompareAndDelete for race-free listenQueue cleanup

fix: wrap listenQueue channel in struct to prevent stale cleanup race

fix: retry CompareAndSwap on failure in concurrent Listen() reconnect

fix: give each Listen() goroutine its own channel to prevent stale reader race

fix: reject Dial send to superseded listenQueue via done channel pre-check

fix: close done channel on all Listen exit paths using sync.Once

fix: swap defer order to close done channel before map deletion

Improve test quality for listen queue race fix

fix: serialize Dial/Listen queue handoff with per-lease mutex

fix: pass Dial context to sendToListener to prevent mutex deadlock under backpressure

refactor: replace goto labels with drainChannel helper in test

fix: clean up leaseLocks entries when Listen exits without replacement

test: add end-to-end deadlock chain test

ci: enable race detector in controller test target

fix: resolve golangci-lint goconst and gofmt violations

fix: use sendToListener in concurrent dial reconnection test

fix: use service methods in listen queue tests instead of raw sync.Map ops

fix: serialize closeDone under lease mutex and stop deleting leaseLocks entries

fix: use non-blocking send in sendToListener to prevent mutex starvation

fix: replace raw lease mutex with ref-counted leaseLock to enable safe cleanup

test: fix race in concurrent lease lock test and add overlapping listener test

ci: add E2E stress test workflow for flakiness detection

Revert "ci: add E2E stress test workflow for flakiness detection"

fix: drain buffered tokens on Listen supersession and assert gRPC status codes