oysterun 1.6.0-beta.2 → 1.6.0-beta.21

package/host-service/scheduler-service.mjs

@@ -27,6 +27,7 @@ import {
   recordHostTelemetryCounter,
   recordHostTelemetryFeature,
 } from "./host-telemetry.mjs";
+import { resolveSchedulerRunTimeout } from "./scheduler-run-timeout-policy.mjs";
 
 const SCHEDULER_LOOP_ACTIVE_REASON =
   "P12.2 in-session loop CLI dispatch is enabled";
@@ -59,7 +60,7 @@ function recordSchedulerRunTelemetry(status) {
   recordHostTelemetryCounter("scheduler_run_count");
   if (status === "dispatched") {
     recordHostTelemetryCounter("scheduler_run_success_count");
-  } else if (status === "failed") {
+  } else if (status === "failed" || status === "timed_out") {
     recordHostTelemetryCounter("scheduler_run_failed_count");
   }
   recordHostTelemetryFeature("scheduler");
@@ -404,6 +405,7 @@ function getHostScheduleRunOutcome(run) {
     ? metadata.provider_exit_code
     : null;
   if (run.status === "failed") return "failed";
+  if (run.status === "timed_out") return "timed_out";
   if (run.status === "skipped") return "skipped";
   if (exitCode === 0) return "success";
   return run.status || null;
@@ -423,8 +425,15 @@ function buildProviderRunMetadata({
     ? providerRun.exit_code
     : null;
   const signal = normalizeOptionalString(providerRun?.signal);
+  const timedOut =
+    signal === "timeout" ||
+    providerRun?.error_summary === "scheduler_provider_run_timed_out";
   const outcome =
-    providerRun?.ok === true && exitCode === 0 ? "success" : "failed";
+    providerRun?.ok === true && exitCode === 0
+      ? "success"
+      : timedOut
+      ? "timed_out"
+      : "failed";
   const startedAt = normalizeOptionalString(providerRun?.started_at);
   const completedAt = normalizeOptionalString(providerRun?.completed_at);
   return {
@@ -449,7 +458,7 @@ function buildProviderRunMetadata({
       providerRun?.log_summary ||
       fallbackLogSummary ||
       "Host scheduler direct provider run finished",
-    ...(outcome === "failed"
+    ...(outcome !== "success"
       ? {
           error_summary:
             providerRun?.error_summary ||
@@ -654,6 +663,7 @@ export class SchedulerService {
     portableDefinitionStore = null,
     clock = () => new Date(),
     recheckMs = SCHEDULER_BUSY_RECHECK_MS,
+    sessionLoopUnavailableGraceMs = undefined,
     sessionLoopService = null,
   } = {}) {
     this.store = store;
@@ -673,6 +683,9 @@ export class SchedulerService {
       new SessionLoopDefinitionService({
         clock,
         recheckMs,
+        ...(sessionLoopUnavailableGraceMs === undefined
+          ? {}
+          : { unavailableGraceMs: sessionLoopUnavailableGraceMs }),
         lifecycleStore: new SessionLoopLifecycleStore({
           configDir: store.configDir,
           clock,
@@ -916,6 +929,22 @@ export class SchedulerService {
     });
   }
 
+  resolveSessionLoopIncidentForGui({
+    hostSessionId,
+    agentId,
+    agentFolder,
+    scheduleId,
+    action,
+  }) {
+    return this.sessionLoopService.resolveIncidentForGui({
+      hostSessionId,
+      agentId,
+      agentFolder,
+      scheduleId,
+      action,
+    });
+  }
+
   clearSessionLoopRuntimeState({ hostSessionId, reason = "session_stop" }) {
     return this.sessionLoopService.clearRuntimeSession({
       hostSessionId,
@@ -962,6 +991,8 @@ export class SchedulerService {
     rule,
     timezone = getHostSystemTimezone(),
     enabled = true,
+    runTimeoutSeconds = undefined,
+    runTimeoutSource = null,
   }) {
     const normalizedRule = normalizeScheduleRule(rule, { timezone });
     const now = nowIso(this.clock);
@@ -985,6 +1016,10 @@ export class SchedulerService {
     const scheduleHostSessionId = `portable:${normalizedTargetBinding.setup_snapshot.agent_folder}`;
     const normalizedPrompt = normalizeHostSchedulePrompt(prompt);
     const active = enabled === true;
+    const runTimeout = resolveSchedulerRunTimeout(runTimeoutSeconds, {
+      present: runTimeoutSeconds !== undefined,
+      source: runTimeoutSource,
+    });
     return {
       hostSessionId: scheduleHostSessionId,
       agentId: normalizedAgentId,
@@ -996,6 +1031,7 @@ export class SchedulerService {
       availabilityReason: active
         ? HOST_SCHEDULE_ACTIVE_REASON
         : HOST_SCHEDULE_UI_PAUSED_REASON,
+      runTimeout,
       metadata: {
         source: "p12_5_host_scheduler_dashboard_ui",
         schedule_rule: normalizedRule,
@@ -1026,6 +1062,7 @@ export class SchedulerService {
     rule,
     timezone = getHostSystemTimezone(),
     enabled = true,
+    runTimeoutSeconds = undefined,
   }) {
     this.initialize();
     const normalized = this.normalizeDashboardHostScheduleInput({
@@ -1036,6 +1073,7 @@ export class SchedulerService {
       rule,
       timezone,
       enabled,
+      runTimeoutSeconds,
     });
     const runtimeSetupSnapshot = normalized.metadata.target_binding.setup_snapshot;
     const schedule = this.portableDefinitionStore.createSchedule(
@@ -1052,6 +1090,10 @@ export class SchedulerService {
         schedule_rule: normalized.metadata.schedule_rule,
         timezone: normalized.metadata.timezone,
         setup_snapshot: runtimeSetupSnapshot,
+        run_timeout_seconds: normalized.runTimeout.seconds,
+        run_timeout_source: normalized.runTimeout.source === "explicit"
+          ? "explicit"
+          : "default",
         metadata: normalized.metadata,
       },
       { agentFolder: runtimeSetupSnapshot.agent_folder }
@@ -1092,6 +1134,15 @@ export class SchedulerService {
     const enabled = enabledProvided
       ? input.enabled === true
       : existing.status === "active";
+    const runTimeout = inputHas("runTimeoutSeconds")
+      ? resolveSchedulerRunTimeout(input.runTimeoutSeconds, {
+          present: true,
+          source: "explicit",
+        })
+      : resolveSchedulerRunTimeout(existing.run_timeout_seconds, {
+          present: true,
+          source: existing.run_timeout_source,
+        });
     const normalized = this.normalizeDashboardHostScheduleInput({
       targetBinding,
       hostSessionId,
@@ -1100,6 +1151,8 @@ export class SchedulerService {
       rule,
       timezone,
       enabled,
+      runTimeoutSeconds: runTimeout.seconds,
+      runTimeoutSource: runTimeout.source,
     });
     const preserveNextRunAt =
       !inputHas("rule") &&
@@ -1124,6 +1177,8 @@ export class SchedulerService {
       schedule_rule: normalized.metadata.schedule_rule,
       timezone: normalized.metadata.timezone,
       setup_snapshot: normalized.metadata.target_binding.setup_snapshot,
+      run_timeout_seconds: runTimeout.seconds,
+      run_timeout_source: runTimeout.source,
       metadata: normalized.metadata,
     });
     return {
@@ -1293,6 +1348,7 @@ export class SchedulerService {
     prompt,
     rule,
     timezone = getHostSystemTimezone(),
+    runTimeoutSeconds = undefined,
   }) {
     this.initialize();
     const normalized = this.normalizeDashboardHostScheduleInput({
@@ -1303,6 +1359,7 @@ export class SchedulerService {
       rule,
       timezone,
       enabled: true,
+      runTimeoutSeconds,
     });
     const runtimeSetupSnapshot = normalized.metadata.target_binding.setup_snapshot;
     const schedule = this.portableDefinitionStore.createSchedule(
@@ -1319,6 +1376,10 @@ export class SchedulerService {
         schedule_rule: normalized.metadata.schedule_rule,
         timezone: normalized.metadata.timezone,
         setup_snapshot: runtimeSetupSnapshot,
+        run_timeout_seconds: normalized.runTimeout.seconds,
+        run_timeout_source: normalized.runTimeout.source === "explicit"
+          ? "explicit"
+          : "default",
         metadata: {
           ...normalized.metadata,
           source: "browser_root_portable_host_scheduler_model",
@@ -1451,7 +1512,12 @@ export class SchedulerService {
   }
 
   async drainSingleSessionLoopDue(loopSchedule, { now, dispatcher }) {
-    const { runtime, definition, state } = loopSchedule;
+    const current = this.sessionLoopService.refreshDueLoopForDispatch(loopSchedule);
+    if (!current) return {
+      status: "skipped", schedule_id: loopSchedule.definition.id,
+      reason: "loop_occurrence_superseded",
+    };
+    const { runtime, definition, state } = current;
     const triggeredAt = normalizeRequiredString(now, "now");
     const serializedSchedule = this.sessionLoopService.serializeForFeedback({
       definition,
@@ -1611,6 +1677,8 @@ export class SchedulerService {
         claimed_target_binding: structuredClone(targetBinding),
         trigger_basis: "schedule_rule_after_actual_trigger_time",
         actual_triggered_at: triggeredAt,
+        run_timeout_seconds: schedule.run_timeout_seconds,
+        run_timeout_source: schedule.run_timeout_source,
         ...buildHostScheduleDispatchProvenance({
           dispatchOrigin,
           dispatchAuthority,
@@ -1652,6 +1720,9 @@ export class SchedulerService {
         mailCapability
       );
       if (safeProviderRun?.ok !== true) {
+        const timedOut =
+          safeProviderRun?.signal === "timeout" ||
+          safeProviderRun?.error_summary === "scheduler_provider_run_timed_out";
         const reason =
           safeProviderRun?.log_summary ||
           safeProviderRun?.error_summary ||
@@ -1662,7 +1733,7 @@ export class SchedulerService {
         const statusAfter = nextRunAt ? "active" : "stopped";
         const failedRun = this.store.updateScheduleRun(run.id, {
           completedAt: safeProviderRun?.completed_at || triggeredAt,
-          status: "failed",
+          status: timedOut ? "timed_out" : "failed",
           error: reason,
           metadata: {
             ...claimedMetadata,
@@ -1703,7 +1774,7 @@ export class SchedulerService {
           });
         }
         return {
-          status: "failed",
+          status: timedOut ? "timed_out" : "failed",
           schedule_id: schedule.id,
           run_id: failedRun.id,
           reason,
@@ -1863,6 +1934,8 @@ export class SchedulerService {
         claimed_target_binding: structuredClone(targetBinding),
         trigger_basis: "manual_dashboard_test_run",
         actual_triggered_at: triggeredAt,
+        run_timeout_seconds: schedule.run_timeout_seconds,
+        run_timeout_source: schedule.run_timeout_source,
         next_run_at_preserved: schedule.next_run_at,
         ...buildHostScheduleDispatchProvenance({
           dispatchOrigin,
@@ -1906,6 +1979,9 @@ export class SchedulerService {
         mailCapability
       );
       if (safeProviderRun?.ok !== true) {
+        const timedOut =
+          safeProviderRun?.signal === "timeout" ||
+          safeProviderRun?.error_summary === "scheduler_provider_run_timed_out";
         const reason =
           safeProviderRun?.log_summary ||
           safeProviderRun?.error_summary ||
@@ -1914,7 +1990,7 @@ export class SchedulerService {
             : "scheduler_direct_provider_run_failed");
         const failedRun = this.store.updateScheduleRun(run.id, {
           completedAt: safeProviderRun?.completed_at || triggeredAt,
-          status: "failed",
+          status: timedOut ? "timed_out" : "failed",
           error: reason,
           metadata: {
             ...claimedMetadata,
@@ -1956,10 +2032,10 @@ export class SchedulerService {
         const retainedSchedule =
           this.portableDefinitionStore.getSchedule(schedule.id);
         return {
-          status: "test_run_failed",
+          status: timedOut ? "test_run_timed_out" : "test_run_failed",
           schedule_id: schedule.id,
           run: {
-            status: "failed",
+            status: timedOut ? "timed_out" : "failed",
             schedule_id: schedule.id,
             run_id: failedRun.id,
             reason,
@@ -2609,6 +2685,10 @@ export class SchedulerService {
       provider_command: metadata.provider_command || null,
       exit_code: exitCode,
       signal: metadata.provider_signal || null,
+      run_timeout_seconds:
+        metadata.run_timeout_seconds || schedule.run_timeout_seconds,
+      run_timeout_source:
+        metadata.run_timeout_source || schedule.run_timeout_source,
       log_summary: redactHostScheduleTextForApi(
         metadata.log_summary,
         tokens
@@ -2656,12 +2736,19 @@ export class SchedulerService {
       targetBinding,
       inputText
     );
+    const runTimeout = resolveSchedulerRunTimeout(schedule.run_timeout_seconds, {
+      present: schedule.run_timeout_seconds !== undefined,
+      source: schedule.run_timeout_source,
+    });
     return {
       id: schedule.id,
       host_session: serializeHostScheduleSessionBoundary(schedule),
       agent: serializeHostScheduleAgentBoundary(schedule),
       schedule_kind: schedule.schedule_kind,
       status: schedule.status,
+      run_timeout_seconds: runTimeout.seconds,
+      run_timeout_source: runTimeout.source,
+      run_timeout_explicit: runTimeout.explicit,
       next_run_at: schedule.next_run_at,
       rule: metadata.schedule_rule
         ? serializeScheduleRuleForSummary(metadata.schedule_rule)

package/host-service/scheduler-store.mjs

@@ -79,6 +79,7 @@ function normalizeRunStatus(value) {
       "dispatched",
       "skipped",
       "failed",
+      "timed_out",
       "canceled",
     ].includes(status)
   ) {
@@ -240,7 +241,7 @@ export class SchedulerStore {
         scheduled_for TEXT NOT NULL,
         claimed_at TEXT,
         completed_at TEXT,
-        status TEXT NOT NULL CHECK (status IN ('planned', 'claimed', 'dispatched', 'skipped', 'failed', 'canceled')),
+        status TEXT NOT NULL CHECK (status IN ('planned', 'claimed', 'dispatched', 'skipped', 'failed', 'timed_out', 'canceled')),
         dispatched_message_id TEXT,
         error TEXT,
         metadata_json TEXT,
@@ -352,6 +353,12 @@ export class SchedulerStore {
     if (hasForeignKeys(this.db, "schedule_runs")) {
       this.rebuildScheduleRunsWithoutForeignKeys();
     }
+    const scheduleRunsSql = this.db
+      .prepare("SELECT sql FROM sqlite_master WHERE type = 'table' AND name = 'schedule_runs'")
+      .get()?.sql;
+    if (!String(scheduleRunsSql || "").includes("'timed_out'")) {
+      this.rebuildScheduleRunsWithoutForeignKeys();
+    }
     if (hasForeignKeys(this.db, "schedule_run_logs")) {
       this.rebuildScheduleRunLogsWithoutForeignKeys();
     }
@@ -374,7 +381,7 @@ export class SchedulerStore {
         scheduled_for TEXT NOT NULL,
         claimed_at TEXT,
         completed_at TEXT,
-        status TEXT NOT NULL CHECK (status IN ('planned', 'claimed', 'dispatched', 'skipped', 'failed', 'canceled')),
+        status TEXT NOT NULL CHECK (status IN ('planned', 'claimed', 'dispatched', 'skipped', 'failed', 'timed_out', 'canceled')),
         dispatched_message_id TEXT,
         error TEXT,
         metadata_json TEXT,