docs: runner lifecycle

engine/contrib-docs/RUNNER_LIFECYCLE.md

@@ -0,0 +1,172 @@
+# Runner Lifecycle
+
+## Connection
+
+```mermaid
+sequenceDiagram
+	participant R as Runner
+	participant P as Pegboard
+	participant RW as Runner Workflow
+
+	note over R,RW: Phase 1: WebSocket Connection
+
+	R->>P: WebSocket open
+	R->>P: ToServerInit (name, version, totalSlots, lastCommandIdx)
+	note over R: start ping interval (3s)
+	note over R: start command ack interval (5min)
+
+	P->>RW: Forward (ToServerInit)
+	note over RW: ProcessInit activity
+	note over RW: load state, process prepopulated actors
+
+	note over R,RW: Phase 2: Initialize Runner
+
+	RW->>P: ToClientInit (runnerId, lastEventIdx, metadata)
+	P->>R: ToClientInit
+
+	note over R: store runnerId
+	note over R: store runnerLostThreshold from metadata
+
+	note over R,RW: Phase 3: Resend Pending State
+
+	note over R: processUnsentKvRequests
+	note over R: resendUnacknowledgedEvents
+	note over R: tunnel.resendBufferedEvents
+
+	note over R,RW: Phase 4: Send Missed Commands
+
+	RW->>P: ToClientCommands (missed commands)
+	P->>R: ToClientCommands
+	note over R: handleCommands
+
+	note over R,RW: Phase 5: Complete Connection
+
+	note over RW: InsertDb activity
+	note over RW: write runner to database
+	note over RW: update allocation indexes
+
+	note over R: config.onConnected callback
+```
+
+## Reconnect
+
+```mermaid
+sequenceDiagram
+	participant R as Runner
+	participant P as Pegboard
+	participant RW as Runner Workflow
+
+	note over R,RW: Phase 1: Detect Disconnection
+
+	alt WebSocket error/close
+		P--xR: connection lost
+		note over R: start runner lost timeout (if threshold configured)
+		note over R: schedule reconnect with backoff
+		note over R: config.onDisconnected callback
+	end
+
+	note over R,RW: Phase 2: Reconnect
+
+	note over R: calculate backoff delay
+	note over R: increment reconnectAttempt counter
+
+	R->>P: WebSocket open (reconnect)
+	R->>P: ToServerInit (lastCommandIdx preserved)
+
+	note over R: clear reconnect timeout
+	note over R: clear runner lost timeout
+	note over R: reset reconnectAttempt = 0
+
+	P->>RW: Forward (ToServerInit)
+	RW->>P: ToClientInit (lastEventIdx)
+	P->>R: ToClientInit
+
+	note over R,RW: Phase 3: Resynchronize
+
+	note over R: if runnerId changed, clear event history
+
+	note over R: processUnsentKvRequests
+	note over R: resendUnacknowledgedEvents (from lastEventIdx)
+	note over R: tunnel.resendBufferedEvents
+
+	alt missed commands exist
+		RW->>P: ToClientCommands (missed commands)
+		P->>R: ToClientCommands
+		note over R: handleCommands
+	end
+
+	note over R: config.onConnected callback
+```
+
+## Shutdown
+
+```mermaid
+sequenceDiagram
+	participant R as Runner
+	participant P as Pegboard
+	participant RW as Runner Workflow
+	participant A as Actors
+
+	note over R,RW: Phase 1: Initiate Shutdown
+
+	alt graceful shutdown
+		R->>P: ToServerStopping
+		P->>RW: Forward (ToServerStopping)
+	else forced stop
+		RW->>RW: receive Stop signal
+	end
+
+	note over R,RW: Phase 2: Drain Runner
+
+	note over RW: handle_stopping
+	note over RW: set state.draining = true
+	note over RW: ClearDb activity (update_state = Draining)
+	note over RW: remove from allocation indexes
+	note over RW: set drain_ts, expired_ts
+
+	note over RW: FetchRemainingActors activity
+	loop for each actor
+		RW->>A: GoingAway signal
+		note over A: actor workflows begin stopping
+	end
+
+	note over R,RW: Phase 3: Wait for Actors
+
+	note over R: waitForActorsToStop (max 120s)
+	loop check every 100ms
+		alt all actors stopped
+			note over R: continue shutdown
+		else websocket closed
+			note over R: force continue shutdown
+		else timeout reached
+			note over R: force continue shutdown
+		end
+	end
+
+	note over R,RW: Phase 4: Close WebSocket
+
+	note over R: send ToServerStopping (if not sent)
+	R->>P: WebSocket close (code=1000, reason=pegboard.runner_shutdown)
+	note over R: clear ping interval
+	note over R: clear ack interval
+	note over R: tunnel.shutdown
+
+	note over R: config.onShutdown callback
+
+	note over R,RW: Phase 5: Complete Workflow
+
+	note over RW: workflow exits drain loop after runner_lost_threshold
+
+	note over RW: ClearDb activity (update_state = Stopped)
+	note over RW: remove from active indexes
+	note over RW: set stop_ts
+
+	note over RW: FetchRemainingActors activity
+	loop for each remaining actor
+		RW->>A: Lost signal
+		note over A: reschedule actors if needed
+	end
+
+	RW->>P: ToClientClose
+	note over RW: workflow complete
+```