enbility / ship-go

93%

LAST BUILD ON BRANCH dev

branch: dev

CHANGE BRANCH

x

Reset

• dev
• auto_accept_stalls_ship
• bug/attempt_reconnect_when_device_reappears
• bug/autoAcceptAnnouncement
• bug/avahi_dead_lock
• bugfix/race-condition
• chore/redundant-nil-checks
• docs/testspec-coverage-comments
• feat/dynamicMdnsInterfaceManagement
• feat/hub-bound-port
• feature/action-update
• feature/analysis-docs
• feature/bugfixes
• feature/hub-improvements
• feature/json-parsing-improvements
• feature/maintenance
• feature/mdns-restart-improvements
• feature/register-improvements
• feature/shippairing-2
• feature/tls-improvements
• fix/connection-attempt-flag-leak-on-unknown-service
• fix/double-connection-outgoing-dedup
• fix/hub-connection-bugs
• fix/hub_connections_test
• fix/json-whitespace-tolerance
• fix/link-local-dial-loglevel
• fix/mdns-autoaccept-txt-update
• fix/mdns-deadlock-and-unreported-read-pump-panic
• fix/mdns-service-type-case-sensitivity
• fix/pairing-issues-and-trust-dupes
• fix/pairing-reactivate-listener-on-addCu-removal
• fix/pairing-startup-listener-guard
• fix/pairing-txt-format-validation
• fix/prevent-stuck-shutdown-test-goroutine-leak
• fix/protocol-handshake-error-envelope
• fix/qr-ski-spacing
• fix/race-job-timeout
• fix/reannounce-entrygroup-leak
• fix/remove-unused-announcement-cancel-ctx
• fix/ship-double-connection
• fix/ship-handshake-conformance
• fix/ship-hello-prolongation-increase
• fix/ship-sec-abort-tls-on-spoofed-cert
• fix/ship-tests-tc-ship-amdata-002
• fix/spine-write-error-teardown
• fix/trusted-reconnect-mdns-drop
• fix/ws-flush-queue-on-closing
• fix_issue_49
• main
• safer-connection-attempt-running-handling
• update-to-zeroconf-v3
• upstream/reset-connection-backoff

Build # 36413688379

Build Type

push

github

Committed by web-flow

Commit Message

fix(hub): don't strand the connection attempt flag when the service is gone (#123)

May fix some failed connection attempts by the living lab after
reregistering SKIs.

coordinateConnectionInitations sets connectionAttemptRunning for a SKI
and raises its attempt counter before it resolves the service entry.
When the entry is gone it returns early — leaving the flag set with no
delay timer attached.

The flag is only ever cleared by the delay timer's callback or by
cancelConnectionDelayTimer, and neither can run without a timer. Every
later call for that SKI then returns at the guard, so the SKI never
initiates a connection again until the process restarts.
UnregisterRemoteService removes the entry concurrently with mDNS
processing, which is exactly how the lookup comes back nil.

The stranded attempt counter has a second effect: it stays raised,
pushing the next delay into the 10–20 s band.

Resolving the service before taking any per-SKI state fixes both. If the
entry is removed after the lookup, the timer already exists and its
callback clears the flag.

Two regression tests: the attempt flag and counter must stay untouched
when no service entry exists, and a SKI must be able to connect again
after an unregister/register cycle.

Coverage Stats

9 of 9 new or added lines in 1 file covered. (100.0%)

4 existing lines in 1 file now uncovered.

6164 of 6626 relevant lines covered (93.03%)

14377.39 hits per line