fix(kernel,server): evict per-session caches on workflow termination (#54)
Two unbounded per-session leaks that never released after a workflow ended: - SessionOrchestrator.artifactContentCache (full file contents, keyed "<sessionId>:<path>") grew for the process lifetime. Added evictArtifactContentCache(sessionId) — session-prefix key removal, rehydrate-safe — called on both completeWorkflow and failWorkflow. cancellations was already evicted on both terminal paths. - NarrationSubscriber leaked a Channel + worker coroutine + lanes map entry per session forever. closeLane() now closes the lane channel on WorkflowCompleted/WorkflowFailed (draining the terminal narration first); the worker self-removes from lanes when its for-loop exits on close. Deferred: SqliteEventStore.subscriptions eviction — removing the SharedFlow mid-life would strand LiveArtifactRepository's downstream collector (a suspended-coroutine leak worse than the tiny empty-SharedFlow entry). Green: :core:kernel compile+detekt, :apps:server *NarrationSubscriber* (8).
This commit is contained in:
+32
-16
@@ -104,22 +104,28 @@ class NarrationSubscriber(
|
||||
stageId = p.stageId.value,
|
||||
),
|
||||
)
|
||||
is WorkflowCompletedEvent -> enqueue(
|
||||
sid,
|
||||
NarrationTrigger(
|
||||
kind = "workflow_completed",
|
||||
instruction = "The workflow finished successfully. Provide a brief summary.",
|
||||
stageId = p.terminalStageId.value,
|
||||
),
|
||||
)
|
||||
is WorkflowFailedEvent -> enqueue(
|
||||
sid,
|
||||
NarrationTrigger(
|
||||
kind = "workflow_failed",
|
||||
instruction = "The workflow failed in stage ${p.stageId.value}: ${p.reason}. Explain the failure to the user.",
|
||||
stageId = p.stageId.value,
|
||||
),
|
||||
)
|
||||
is WorkflowCompletedEvent -> {
|
||||
enqueue(
|
||||
sid,
|
||||
NarrationTrigger(
|
||||
kind = "workflow_completed",
|
||||
instruction = "The workflow finished successfully. Provide a brief summary.",
|
||||
stageId = p.terminalStageId.value,
|
||||
),
|
||||
)
|
||||
closeLane(sid)
|
||||
}
|
||||
is WorkflowFailedEvent -> {
|
||||
enqueue(
|
||||
sid,
|
||||
NarrationTrigger(
|
||||
kind = "workflow_failed",
|
||||
instruction = "The workflow failed in stage ${p.stageId.value}: ${p.reason}. Explain the failure to the user.",
|
||||
stageId = p.stageId.value,
|
||||
),
|
||||
)
|
||||
closeLane(sid)
|
||||
}
|
||||
// Surface the semantic reviewer's verdict conversationally instead of a raw findings
|
||||
// dump: the narrator turns "FAIL, 2 findings" into a plain-language explanation the
|
||||
// operator can act on. Only narrate when there's something to say (non-PASS or findings).
|
||||
@@ -221,6 +227,13 @@ class NarrationSubscriber(
|
||||
lanes[sessionId.value]?.pendingPauses?.clear()
|
||||
}
|
||||
|
||||
/** Closes a terminated session's lane so its worker drains any buffered narration (the terminal
|
||||
* one enqueued just before this) and then exits, self-removing the lane from [lanes]. Without
|
||||
* this, every session leaks its channel + worker coroutine + map entry forever. */
|
||||
private fun closeLane(sessionId: SessionId) {
|
||||
lanes[sessionId.value]?.channel?.close()
|
||||
}
|
||||
|
||||
private fun startLane(sessionId: SessionId): SessionLane {
|
||||
val channel = Channel<QueuedNarration>(capacity = Channel.UNLIMITED)
|
||||
val lane = SessionLane(channel, used = 0)
|
||||
@@ -249,6 +262,9 @@ class NarrationSubscriber(
|
||||
)
|
||||
}
|
||||
}
|
||||
// Channel closed on workflow termination: drop the lane so it stops leaking. A later event
|
||||
// for the same session (none expected post-terminal) would lazily start a fresh lane.
|
||||
lanes.remove(sessionId.value)
|
||||
}
|
||||
return lane
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user