@zq-silk/yui 0.7.0 → 0.8.1

package/dist/scheduler/roleRunStall.js

@@ -51,6 +51,8 @@ export function clearMatchingLeaderStallAttention(store, taskId, runId) {
  */
 export function projectRoleRunHealth(input) {
     const windowMs = input.windowMs ?? DEFAULT_STALL_WINDOW_MS;
+    const diagnosticAfterMs = input.diagnosticAfterMs
+        ?? DEFAULT_WORKFLOW_STALL_CANDIDATE_AGE_MS;
     if (!Number.isFinite(windowMs) || windowMs <= 0) {
         throw new Error("Role run stall window must be a positive number of milliseconds.");
     }
@@ -65,7 +67,7 @@ export function projectRoleRunHealth(input) {
     const candidate = Number.isFinite(candidateAge)
         && (input.deliveredAt === undefined
             ? candidateAge >= windowMs
-            : candidateAge >= DEFAULT_WORKFLOW_STALL_CANDIDATE_AGE_MS);
+            : candidateAge >= diagnosticAfterMs);
     const providerAcceptance = input.providerAcceptance
         ?? (input.deliveredAt === undefined ? "ambiguous" : "accepted");
     const hostLiveness = input.hostLiveness;
@@ -84,7 +86,7 @@ export function projectRoleRunHealth(input) {
                         : "matching";
     const resourceActivity = hostLiveness === "present"
         && nativeSession === "matching"
-        && resourceEvidenceIsFresh(input.resource, input.now, windowMs)
+        && resourceEvidenceIsFresh(input.resource, input.now, windowMs, diagnosticAfterMs)
         // Residency/RSS and an unchanged cumulative counter are not progress.
         && input.resource?.active === true
         && input.resource?.changed === true;
@@ -494,7 +496,7 @@ export function isRoleRunStalled(events, runId) {
  * structured waiting/working facts. No branch sends terminal bytes, retries,
  * replaces a Session, or changes Run status.
  */
-export async function reconcileStalledRoleRuns(store, delivery, now, selection, windowMs = DEFAULT_STALL_WINDOW_MS, liveStatuses, resourceEvidence) {
+export async function reconcileStalledRoleRuns(store, delivery, now, selection, windowMs = DEFAULT_STALL_WINDOW_MS, liveStatuses, resourceEvidence, diagnosticAfterMs = DEFAULT_WORKFLOW_STALL_CANDIDATE_AGE_MS) {
     // Dirty mailbox passes are intentionally not a second scheduler. Full
     // reconcile owns the all-active-Run scan; dirty passes may still route the
     // existing mailbox work without manufacturing another episode.
@@ -519,7 +521,7 @@ export async function reconcileStalledRoleRuns(store, delivery, now, selection,
                 session: store.getRoleSession(task.id, role.name, run.effective.agentId)
             }];
     })));
-    const stallCandidates = candidates.filter(({ run }) => isStallCandidate(run, now, windowMs));
+    const stallCandidates = candidates.filter(({ run }) => isStallCandidate(run, now, windowMs, diagnosticAfterMs));
     if (stallCandidates.length === 0)
         return [];
     const stallCandidateKeys = new Set(stallCandidates.map(({ task, role }) => (`${task.id}\0${role.name}`)));
@@ -689,7 +691,7 @@ export async function reconcileStalledRoleRuns(store, delivery, now, selection,
             && sessionUsable
             && expectedResourceIdentity !== undefined
             && resourceEvidenceMatchesCurrentRun(resourceSnapshot, expectedResourceIdentity, progressAt)
-            && resourceEvidenceIsFresh(resourceSnapshot, now, windowMs);
+            && resourceEvidenceIsFresh(resourceSnapshot, now, windowMs, diagnosticAfterMs);
         const resource = resourceIsCurrent ? resourceSnapshot : undefined;
         const health = projectRoleRunHealth({
             progressAt,
@@ -699,6 +701,7 @@ export async function reconcileStalledRoleRuns(store, delivery, now, selection,
                 : { deliveredAt: candidate.run.deliveredAt }),
             now,
             windowMs,
+            diagnosticAfterMs,
             hostLiveness: live,
             nativeSession: candidate.session,
             providerAcceptance: candidate.run.deliveredAt === undefined
@@ -825,7 +828,7 @@ export async function reconcileStalledRoleRuns(store, delivery, now, selection,
     }
     return raised;
 }
-function isStallCandidate(run, now, windowMs) {
+function isStallCandidate(run, now, windowMs, diagnosticAfterMs) {
     const startAt = run.deliveredAt ?? run.createdAt;
     const ageMs = now.getTime() - Date.parse(startAt);
     if (!Number.isFinite(ageMs))
@@ -834,7 +837,7 @@ function isStallCandidate(run, now, windowMs) {
     // window, but they never enter workflow-stall candidate filtering.
     return run.deliveredAt === undefined
         ? ageMs >= windowMs
-        : ageMs >= DEFAULT_WORKFLOW_STALL_CANDIDATE_AGE_MS;
+        : ageMs >= diagnosticAfterMs;
 }
 function classifyLeaderStall(store, taskId, observed, now, windowMs) {
     if (store.hasOpenInputRequest(taskId))
@@ -1020,7 +1023,7 @@ function resourceEvidenceMatchesCurrentRun(resource, expected, progressAt) {
 function hasResourceIdentityText(value) {
     return typeof value === "string" && value.trim().length > 0;
 }
-function resourceEvidenceIsFresh(evidence, now, windowMs) {
+function resourceEvidenceIsFresh(evidence, now, windowMs, diagnosticAfterMs = DEFAULT_WORKFLOW_STALL_CANDIDATE_AGE_MS) {
     if (evidence === undefined)
         return false;
     const observedAt = Date.parse(evidence.observedAt);
@@ -1028,5 +1031,5 @@ function resourceEvidenceIsFresh(evidence, now, windowMs) {
         return false;
     // A sample from an earlier scheduler window is not a current health signal.
     return observedAt <= now.getTime()
-        && now.getTime() - observedAt < Math.max(windowMs, DEFAULT_WORKFLOW_STALL_CANDIDATE_AGE_MS);
+        && now.getTime() - observedAt < Math.max(windowMs, diagnosticAfterMs);
 }