[ppenguin]

Startup races in the event/action dispatch.

Problem	Fix
Services.Start runs EventServiceWS.Start in a goroutine; YSK subscribes right after → Start resets subscriberChannels → that subscription is dropped (cards stop updating, depending on scheduling)	Start keeps an existing subscriber map (event + action)
Publish / Trigger before Start: select on (*s.ctx).Done() with a nil ctx → panic	drop (and log) until Start ran

Tests

• subscription made before Start receives an event
• publish before Start returns
• red → green: both fail on main (timeout / nil-pointer panic)
• ysk tests: already poll; comment updated (the boot wait stays: no readiness signal)

Context: ReCasaOS/CasaOS#11

🤖 Generated with Claude Code

[inkly]

The event-service fix looks right.

Please apply the same locking to ActionServiceWS: Start's nil check, Subscribe and Trigger are still unsynchronised (and Start still resets s.mutex), so a pre-Start subscription can be lost and the reads race with Start. A test for that path would help, since CI runs -race.

publishUntilReceived republishes every 10 ms, so a slow first delivery under -race yields a duplicate that the next iteration receives, or a leaked forwarder that trips goleak. Consider a ready channel closed in Start: it would remove the republish loops and the 1 s sleep in ysk_test.go. Dropping before Start instead of queueing is fine with us; Warn would be enough for that log line.

We will squash-merge with a plain commit message (we keep tool attribution trailers out of main's history), so nothing to change there.

[ppenguin]

Done in f850e35:

• ActionServiceWS: Start no longer resets the mutex; Start setup/teardown, Trigger's read, Subscribe, dispatch and heartbeat under it
• Ready() on both services: closed by Start once its state is in
  • tests wait on it: republish loops and the 1 s sleep in pkg/ysk gone
• Trigger waits for the dispatch like Publish (was: dropped when the dispatch was busy). Routes trigger from their own goroutine. Needed for a deterministic test after Ready(); say if you'd rather keep the drop.
• before Start: Publish/Trigger drop with Warn
• tests (-race): action subscription before Start survives it, Trigger before Start, 8× Subscribe/Trigger concurrent with Start; -count=30 clean

[inkly]

Thanks, this addresses both points: the action service now locks like the event service, and Ready() removes the republish loops and the sleep. Approving.

One small follow-up, older than this PR: Start's teardown closes the inbound channel while a Trigger/Publish may already hold it, and with ctx.Done also ready select can pick the send and panic (send on closed channel; reproduced with triggers racing cancel()). Since senders already select on ctx.Done(), setting the field to nil without closing it is enough. Optionally: subscriberBuffer for action subscribers too (still 1), and a timeout on the receive in TestEventTypeService.

We'll squash-merge with a plain commit message.

[ppenguin]

Follow-up: #7 (teardown no longer closes the inbound channel, action subscriber buffer 64, receive timeout in TestEventTypeService).