@atolis-hq/wake 0.3.84 → 0.3.85

package/dist/src/bootstrap/self-update-failure-log.js

@@ -1,5 +1,13 @@
 import { mkdir, readFile, rename, rm, writeFile } from 'node:fs/promises';
 import { dirname, join } from 'node:path';
+// recordUpdateFailure (self-update-application.ts) records every failed
+// attempt through this same log regardless of stage — including one that
+// never reached rollout.deploy() at all, e.g. a quiesce timeout waiting on
+// active Runs. "rolled back" would misreport those as a completed deploy
+// that had to be undone, when nothing was ever deployed.
+export function describeSelfUpdateFailure(failure) {
+    return `update to ${failure.tag} failed at ${failure.occurredAt}: ${failure.message}`;
+}
 /**
  * Persists the most recent self-update rollback so it can surface on the
  * operator health screen instead of relying on a third-party notification —

package/dist/src/bootstrap/surface-api-applications.js

@@ -3,7 +3,7 @@ import { ApiCommandStatus, fromWorkItemKey, presentBoardCard, presentResource, p
 import { analyticsProjection } from './analytics-projection.js';
 import { boardConditionCounts, boardProjection, } from './board-projection.js';
 import { primaryExternalRef } from './external-ref.js';
-import { createSelfUpdateFailureLog } from './self-update-failure-log.js';
+import { createSelfUpdateFailureLog, describeSelfUpdateFailure, } from './self-update-failure-log.js';
 import { createExecutionApplications } from './surface-api-execution-applications.js';
 import { projectionMeta, sampledMeta } from './surface-api-metadata.js';
 import { projectionPage } from './surface-api-projection-pages.js';
@@ -200,7 +200,7 @@ function createSystemApplications(root, now) {
                     : {
                         name: 'self-update',
                         status: 'degraded',
-                        detail: `rolled back from ${selfUpdateFailure.tag} at ${selfUpdateFailure.occurredAt}: ${selfUpdateFailure.message}`,
+                        detail: describeSelfUpdateFailure(selfUpdateFailure),
                     },
             ];
             const adapters = root.providers.flatMap((instance) => (instance.health?.() ?? []).map((check) => ({
@@ -309,15 +309,25 @@ async function performTick(root, now, command, sequence) {
         result: await readControlPlaneStatus(root, now),
     };
 }
-async function readControlPlaneStatus(root, now) {
+export async function readControlPlaneStatus(root, now) {
     const stored = await root.projections.read(ControlStreamKind.Global, 'global');
     const meta = await projectionMeta(root.journal, stored === null ? [] : [stored], now());
+    const lease = await root.maintenance.read();
     return {
         data: {
             paused: stored?.value.pausedUntil !== null && stored?.value.pausedUntil !== undefined,
             ...(stored?.value.pausedUntil == null ? {} : { pausedUntil: stored.value.pausedUntil }),
             ...(stored?.value.reason === undefined ? {} : { reason: stored.value.reason }),
             updatedAt: meta.asOf,
+            ...(lease === null
+                ? {}
+                : {
+                    maintenanceLease: {
+                        phase: lease.phase,
+                        startedAt: lease.startedAt,
+                        ...(lease.failure === undefined ? {} : { failure: lease.failure }),
+                    },
+                }),
         },
         meta,
     };

package/dist/src/bootstrap/surface-cli-applications.js

@@ -9,7 +9,7 @@ import { IntakeHost, ResidentHost, TickHost } from '../control-plane/index.js';
 import { ExecutionCancellationReason, ExecutionFailureCode, RunStatus, loadPromptTemplate, } from '../execution/index.js';
 import { EventActorKind, correlationId } from '../kernel/index.js';
 import { ResourceCorrelationRole, resourceId } from '../resources/index.js';
-import { DockerProcessError, createApiDispatcher, createApiHttpServer, createLoggedDockerCli, createPackagedAssetSource, createProcessLogSink, createSandboxDockerPort, drainProcessOutput, runDoctor, runSandbox, runSandboxEntrypoint, runSandboxSetup, runSelfUpdateLatestLoop, runTargetSmoke, verifyResidentStart, waitForActiveRuns, } from '../surfaces/index.js';
+import { DockerProcessError, createApiDispatcher, createApiHttpServer, createLoggedDockerCli, createPackagedAssetSource, createProcessLogSink, createSandboxDockerPort, drainProcessOutput, runDoctor, runSandbox, runSandboxEntrypoint, runSandboxSetup, runSelfUpdateLatestLoop, runTargetSmoke, verifyResidentStart, waitForActiveRuns, waitForever, } from '../surfaces/index.js';
 import { WorkStreamKind, workItemId } from '../work/index.js';
 import { loadConfig } from './config/load-config.js';
 import { runtimeProjectionDefinitions } from './projection-runtime.js';
@@ -456,6 +456,7 @@ async function doctorDiagnostics(root) {
         }
     }
     await checkProviders(root, failures, notices);
+    await checkMaintenanceLease(root, failures);
     notices.push(...(await dockerSandboxHealthNotices(root)));
     for (const [name, path] of Object.entries({
         events: root.paths.eventsRoot,
@@ -490,6 +491,21 @@ function referencedPromptTemplateNames(workflows) {
     }
     return [...names];
 }
+// A retained maintenance lease pauses every resident loop (intake and
+// dispatch) regardless of phase -- including Failed, since it's the
+// operator's decision whether a failed attempt is safe to retry or clear.
+// Without this check the pause is invisible: the process ticks normally and
+// logs nothing, so a stuck lease from an update that couldn't quiesce active
+// Runs looks identical to a healthy idle system.
+async function checkMaintenanceLease(root, failures) {
+    const lease = await root.maintenance.read();
+    if (lease === null)
+        return;
+    const detail = lease.failure === undefined ? '' : ` (${lease.failure})`;
+    failures.push(`update maintenance lease is held in phase "${lease.phase}" since ${lease.startedAt}${detail} -- ` +
+        'every resident loop stays paused until it is cleared or resumes; run `wake self-update` ' +
+        'to retry, or clear the lease manually if the attempt is abandoned');
+}
 async function checkProviders(root, failures, notices) {
     for (const provider of root.providers) {
         if (provider.adapter.trim().length === 0) {
@@ -661,7 +677,7 @@ function createSandboxEntrypointDependencies(root) {
         waitForExit: async (pid) => children.get(pid) ?? 1,
         writeFile: (path, content) => writeFileContent(path, content, 'utf8'),
         sleep: (milliseconds) => new Promise((resolve) => setTimeout(resolve, milliseconds)),
-        waitForever: () => new Promise(() => { }),
+        waitForever,
         log: (message) => process.stdout.write(`${message}\n`),
     };
 }

package/dist/src/bootstrap/version.js

@@ -108,4 +108,4 @@ export function resolveWakeVersion(options = {}) {
         return `g${headHash.slice(0, 7)}`;
     return '0.1.0-dev';
 }
-export const wakeVersion = "g80c78a2";
+export const wakeVersion = "g6de20bb";

package/dist/src/persistence/filesystem/file-event-journal.js

@@ -15,6 +15,15 @@ export class FileEventJournal {
         return this.changeSignalSource;
     }
     cached;
+    // Every read (readStream/readAll/readLatest/latestGlobalPosition) reaches
+    // scan(), and the resident host's tick loop calls those many times a
+    // second while work is progressing. Without coalescing, calls that land
+    // concurrently each independently decide the cache is stale and each
+    // re-read and re-parse the entire on-disk journal — under I/O pressure
+    // (e.g. a concurrently running build/test process competing for disk),
+    // enough of these can overlap at once to exhaust the heap. Sharing one
+    // in-flight decode among concurrent callers makes that impossible.
+    inFlightScan;
     async append(stream, expectedSequence, drafts) {
         return withFileLock(join(this.root, 'locks', 'event-journal.lock'), async () => {
             const current = await this.scan();
@@ -97,7 +106,17 @@ export class FileEventJournal {
             : [...priorEntries, updatedEntry];
         this.cached = { entries, events: [...priorEvents, ...newEnvelopes] };
     }
-    async scan() {
+    scan() {
+        if (this.inFlightScan !== undefined)
+            return this.inFlightScan;
+        const run = this.scanUncoalesced().finally(() => {
+            if (this.inFlightScan === run)
+                this.inFlightScan = undefined;
+        });
+        this.inFlightScan = run;
+        return run;
+    }
+    async scanUncoalesced() {
         const directory = join(this.root, 'events');
         let files;
         try {

package/dist/src/surfaces/cli/commands/sandbox-entrypoint.js

@@ -1,3 +1,17 @@
+/**
+ * Never resolves, and never lets the process exit either. The supervised
+ * child is spawned detached and unref'd (see spawnDetached), so an
+ * unresolved Promise alone is not enough here — a Promise executor registers
+ * no libuv handle, and once other pending I/O quiets down Node treats the
+ * still-pending top-level await as unsettled and exits anyway. A ref'd
+ * timer is a real handle, so it keeps the process alive for as long as this
+ * Promise is meant to.
+ */
+export function waitForever() {
+    return new Promise(() => {
+        setInterval(() => { }, 1 << 30);
+    });
+}
 /**
  * Restarts `wake start` on every exit — a first-boot crash (e.g. missing
  * sandbox auth) must not stop the retry loop, since the operator's only path