* MM-68179: Run sendLoop workers on all HA nodes
In HA clusters, sendLoop worker goroutines only ran on the leader node.
When an API request to send a channel invite landed on a non-leader node,
SendMsg enqueued the task to a local in-memory channel but no goroutine
consumed it, silently losing the message. Fix by starting sendLoop workers
in Start() on all nodes, independent of the leader-only ping lifecycle.
- Separate sendLoop lifecycle (Start/Shutdown) from ping lifecycle
(pingStart/pingStop on leader change)
- Rename resume/pause to pingStart/pingStop for clarity
- Change Active() to mean "service started" via atomic.Bool
- Remove SetActive (no longer needed; tests use Start())
* address review comment
* Added idempotency guard to Start()
* Start() and Shutdown(): CompareAndSwap instead of Load/Store — eliminates races where concurrent calls could both proceed. Only the winner of the CAS executes; the loser returns nil
immediately.
Ping test: replaced time.Sleep with assert.Never/assert.Eventually — no more brittle fixed sleeps. Uses assert.Never to verify no pings fire on non-leader, and assert.Eventually to
verify pings stop after losing leadership (snapshot-then-compare pattern).
* make unit tests parallel capable
---------
Co-authored-by: Mattermost Build <build@mattermost.com>
