aglio-worker 0.1.0 → 0.6.0

package/agent.js

@@ -7345,8 +7345,10 @@ var NOTIFICATION_TYPES = [
   "run_finished",
   "run_failed",
   "review_feedback",
+  "pr_ready",
   "pr_merged",
-  "mention"
+  "mention",
+  "comment"
 ];
 var CREDENTIAL_SCOPES = ["workspace", "user"];
 var CREDENTIAL_AUTH_TYPES = ["api_key", "oauth_token", "session_bundle"];
@@ -27267,45 +27269,58 @@ var repoConnectionReportSchema = external_exports.object({
   repos: external_exports.array(repoConnectionCheckSchema)
 });
 
+// ../../packages/shared/src/notifications.ts
+var DEFAULT_NOTIFICATION_PREFERENCES = {
+  emailDigest: true,
+  emailQuestions: true,
+  emailFailures: true,
+  emailPrReady: true
+};
+var notificationPreferencesSchema = external_exports.object({
+  emailDigest: external_exports.boolean().catch(DEFAULT_NOTIFICATION_PREFERENCES.emailDigest),
+  emailQuestions: external_exports.boolean().catch(DEFAULT_NOTIFICATION_PREFERENCES.emailQuestions),
+  emailFailures: external_exports.boolean().catch(DEFAULT_NOTIFICATION_PREFERENCES.emailFailures),
+  emailPrReady: external_exports.boolean().catch(DEFAULT_NOTIFICATION_PREFERENCES.emailPrReady)
+});
+function isSlackWebhookUrl(value) {
+  if (typeof value !== "string" || !value.trim()) return false;
+  try {
+    const url2 = new URL(value.trim());
+    return url2.protocol === "https:";
+  } catch {
+    return false;
+  }
+}
+
 // ../../packages/shared/src/routing.ts
-var ROUTING_LABEL_KEYS = ["os", "mode", "tools", "size"];
 var ROUTING_OS_VALUES = ["linux", "macos"];
 var ROUTING_MODE_VALUES = ["docker", "host"];
-var ROUTING_TOOL_VALUES = ["ios", "xcode", "android", "node"];
-var ROUTING_SIZE_VALUES = ["small", "large"];
-var ROUTING_LABEL_VALUES = {
+var ROUTING_TOOL_VALUES = ["ios", "xcode", "android", "gpu"];
+var ROUTING_SIZE_VALUES = ["small", "medium", "large"];
+var ROUTING_GPU_VALUES = ["t4", "a100", "h100"];
+var ROUTING_ARCH_VALUES = ["x64", "arm64"];
+var ROUTING_REGION_VALUES = ["us-east", "us-west", "eu-west", "ap-southeast"];
+var ROUTING_LABEL_SUGGESTIONS = {
   os: ROUTING_OS_VALUES,
   mode: ROUTING_MODE_VALUES,
   tools: ROUTING_TOOL_VALUES,
-  size: ROUTING_SIZE_VALUES
-};
-var LABEL_PATTERN = /^([a-z]+):([a-z0-9][a-z0-9._-]*)$/;
-function parseOne(value) {
-  const match = LABEL_PATTERN.exec(value);
-  if (!match) {
-    throw new Error(`Invalid routing label "${value}". Use key:value, for example mode:host.`);
-  }
-  const key = match[1];
-  if (!ROUTING_LABEL_KEYS.includes(key)) {
-    throw new Error(
-      `Unknown routing label "${value}". Use os, mode, tools or size.`
-    );
-  }
-  const allowed = ROUTING_LABEL_VALUES[key];
-  if (!allowed.includes(match[2])) {
-    throw new Error(
-      `Unknown routing label "${value}". Use ${allowed.map((entry) => `${key}:${entry}`).join(", ")}.`
-    );
-  }
-  return { key, value: match[2] };
-}
+  size: ROUTING_SIZE_VALUES,
+  gpu: ROUTING_GPU_VALUES,
+  arch: ROUTING_ARCH_VALUES,
+  region: ROUTING_REGION_VALUES
+};
+var ROUTING_LABEL_SUGGESTION_LIST = Object.entries(
+  ROUTING_LABEL_SUGGESTIONS
+).flatMap(([key, values]) => values.map((value) => `${key}:${value}`));
+var ROUTING_LABEL_MAX_LENGTH = 64;
+var LABEL_PATTERN = /^([a-z][a-z0-9_-]*):([a-z0-9][a-z0-9._-]*)$/;
 function parseRoutingLabels(input2) {
   const values = Array.isArray(input2) ? input2 : typeof input2 === "string" ? input2.split(/[\s,]+/) : [];
   const seen = /* @__PURE__ */ new Set();
   const labels = [];
   for (const value of values) {
     if (typeof value !== "string") continue;
-    const label = value.trim().toLowerCase();
+    const label = value.trim();
     if (!label || seen.has(label)) continue;
     seen.add(label);
     labels.push(label);
@@ -27314,24 +27329,16 @@ function parseRoutingLabels(input2) {
 }
 function validateRoutingLabels(input2) {
   const labels = parseRoutingLabels(input2);
-  const seenValues = /* @__PURE__ */ new Map();
-  const tools = /* @__PURE__ */ new Set();
   for (const label of labels) {
-    const { key, value } = parseOne(label);
-    if (key === "os" || key === "mode" || key === "size") {
-      const previous = seenValues.get(key);
-      if (previous && previous !== value) {
-        throw new Error(`Conflicting routing labels "${key}:${previous}" and "${label}".`);
-      }
+    if (label.length > ROUTING_LABEL_MAX_LENGTH) {
+      throw new Error(
+        `Invalid routing label "${label}". Labels are at most ${ROUTING_LABEL_MAX_LENGTH} characters.`
+      );
     }
-    seenValues.set(key, value);
-    if (key === "tools") tools.add(value);
-  }
-  const os = seenValues.get("os");
-  const mode = seenValues.get("mode");
-  if (tools.has("ios")) {
-    if (os && os !== "macos" || mode && mode !== "host") {
-      throw new Error("tools:ios requires os:macos and mode:host.");
+    if (!LABEL_PATTERN.test(label)) {
+      throw new Error(
+        `Invalid routing label "${label}". Use lowercase key:value, for example size:medium.`
+      );
     }
   }
   return labels;
@@ -27395,6 +27402,8 @@ var projectSettingsSchema = external_exports.object({
   runRetryBackoffSeconds: external_exports.number().int().min(10).max(600).default(60),
   runRetryMaxBackoffSeconds: external_exports.number().int().min(60).max(3600).default(600),
   notificationSound: external_exports.boolean().default(true),
+  /** Slack incoming webhook for question, failure and PR-ready alerts. Null disables Slack. */
+  slackWebhookUrl: external_exports.string().trim().max(500).refine(isSlackWebhookUrl, "Slack webhook must be an https URL").nullable().default(null),
   /** Rendered column strip: built-in columns and custom list ids. */
   boardColumnOrder: external_exports.array(external_exports.string()).optional(),
   /** Last repository connectivity check from Project settings or the worker gate. */
@@ -27594,11 +27603,12 @@ function isHostLocalSigningArtifact(name) {
 var WORKER_HEARTBEAT_SECONDS = 15;
 var WORKER_OFFLINE_AFTER_MISSES = 3;
 var WORKER_OFFLINE_AFTER_SECONDS = WORKER_HEARTBEAT_SECONDS * WORKER_OFFLINE_AFTER_MISSES;
+var WORKER_CLAIM_LONG_POLL_SECONDS = 25;
 var WORKER_MIN_VERSION = "0.1.0";
 var WORKER_VERSION_HEADER = "x-aglio-worker-version";
 var WORKER_MODES = ["docker", "host"];
 var WORKER_OSES = ["linux", "macos"];
-var WORKER_LABEL_MAX_LENGTH = 64;
+var WORKER_LABEL_MAX_LENGTH = ROUTING_LABEL_MAX_LENGTH;
 var WORKER_IOS_LABEL = "tools:ios";
 var WORKER_ENROLLMENT_TOKEN_PREFIX = "aglio_enr_";
 var WORKER_TOKEN_PREFIX = "aglio_wkr_";
@@ -30748,15 +30758,38 @@ function parseClaimedJob(value) {
   if (!lease || typeof lease.id !== "string" || typeof lease.token !== "string") return null;
   return value;
 }
+function claimHoldSeconds(config2) {
+  const requested = config2.claimTimeoutSeconds;
+  if (typeof requested !== "number" || !Number.isFinite(requested)) {
+    return WORKER_CLAIM_LONG_POLL_SECONDS;
+  }
+  return Math.min(Math.max(Math.round(requested), 5), 120);
+}
+var CLAIM_REQUEST_GRACE_MS = 2e4;
 async function claim2(config2, token, fetchImpl, signal) {
-  const response = await request(
-    config2,
-    fetchImpl,
-    "/api/workers/claim",
-    token,
-    { version: config2.version, capacity: config2.capacity },
-    signal
+  const holdSeconds = claimHoldSeconds(config2);
+  const controller = new AbortController();
+  const onAbort = () => controller.abort(signal?.reason);
+  signal?.addEventListener("abort", onAbort, { once: true });
+  const timer = setTimeout(
+    () => controller.abort(new Error("Claim request did not answer in time.")),
+    holdSeconds * 1e3 + CLAIM_REQUEST_GRACE_MS
   );
+  timer.unref?.();
+  let response;
+  try {
+    response = await request(
+      config2,
+      fetchImpl,
+      "/api/workers/claim",
+      token,
+      { version: config2.version, capacity: config2.capacity, claimTimeoutSeconds: holdSeconds },
+      controller.signal
+    );
+  } finally {
+    clearTimeout(timer);
+    signal?.removeEventListener("abort", onAbort);
+  }
   if (response.status === 204) return { status: 204, job: null, reason: null, message: null };
   if (!response.ok) throw await errorFrom(response);
   const body = await responseBody(response);
@@ -30767,6 +30800,15 @@ async function claim2(config2, token, fetchImpl, signal) {
     message: typeof body.message === "string" ? body.message : null
   };
 }
+function isTransientControlPlaneError(error62) {
+  if (error62 instanceof WorkerHttpError) return error62.status >= 500 || error62.status === 429;
+  return true;
+}
+var CLAIM_BACKOFF_MS = [5e3, 1e4, 2e4, 3e4];
+function claimBackoffMs(attempt) {
+  const index = Math.min(Math.max(Math.floor(attempt), 1), CLAIM_BACKOFF_MS.length) - 1;
+  return CLAIM_BACKOFF_MS[index];
+}
 async function jobTokenRequest(config2, fetchImpl, path, token, init) {
   return fetchImpl(joinUrl(config2.apiUrl, path), {
     method: init.method,
@@ -30805,18 +30847,23 @@ function workerCardApi(config2, fetchImpl, signal) {
       return new Uint8Array(await response.arrayBuffer());
     },
     async updateRunStatus(input2) {
-      const response = await jobTokenRequest(
-        config2,
-        fetchImpl,
-        `/api/workers/jobs/${input2.jobId}/run/status`,
-        input2.token,
-        {
-          method: "POST",
-          body: JSON.stringify({ leaseId: input2.leaseId, ...input2.patch }),
-          signal
-        }
+      await retryTransient(
+        async () => {
+          const response = await jobTokenRequest(
+            config2,
+            fetchImpl,
+            `/api/workers/jobs/${input2.jobId}/run/status`,
+            input2.token,
+            {
+              method: "POST",
+              body: JSON.stringify({ leaseId: input2.leaseId, ...input2.patch }),
+              signal
+            }
+          );
+          if (!response.ok) throw await errorFrom(response);
+        },
+        { signal }
       );
-      if (!response.ok) throw await errorFrom(response);
     },
     async appendRunEvents(input2) {
       const response = await jobTokenRequest(
@@ -31004,6 +31051,24 @@ function jobApi(config2, fetchImpl, signal) {
     }
   };
 }
+var TRANSIENT_RETRY_DELAYS_MS = [500, 1500];
+async function retryTransient(call, options = {}) {
+  const sleep2 = options.sleep ?? ((ms) => new Promise((resolve6) => setTimeout(resolve6, ms)));
+  let lastError;
+  for (let attempt = 0; attempt <= TRANSIENT_RETRY_DELAYS_MS.length; attempt += 1) {
+    try {
+      return await call();
+    } catch (error62) {
+      lastError = error62;
+      if (!isTransientControlPlaneError(error62) || options.signal?.aborted) throw error62;
+      const delay = TRANSIENT_RETRY_DELAYS_MS[attempt];
+      if (delay === void 0) break;
+      options.onRetry?.(error62, attempt + 1);
+      await sleep2(delay);
+    }
+  }
+  throw lastError;
+}
 function fatalExitCode(error62) {
   if (error62.status === 401 && error62.code === "worker_revoked") return 0;
   if (error62.status === 426 || error62.code === "incompatible_worker_version") return 2;
@@ -31121,10 +31186,27 @@ async function runRemoteWorker(config2, deps = {}) {
       1,
       Math.min(workerCapacity(config2.capacity, resolved.capacity), workerCapacity(resolved.capacity))
     );
+    let claimAttempt = 0;
+    const backOffClaim = async (error62) => {
+      claimAttempt += 1;
+      const delay = claimBackoffMs(claimAttempt);
+      if (claimAttempt === 1) {
+        log("claim unavailable, backing off", {
+          ...error62 instanceof WorkerHttpError ? { status: error62.status, code: error62.code } : {},
+          error: error62 instanceof Error ? error62.message : String(error62),
+          retryInSeconds: Math.round(delay / 1e3)
+        });
+      }
+      await sleep2(delay);
+    };
     const runSlot = async () => {
       while (!controller.signal.aborted) {
         try {
           const result = await claim2(config2, resolved.token, fetchImpl, controller.signal);
+          if (claimAttempt > 0) {
+            log("claim recovered", { attempts: claimAttempt });
+            claimAttempt = 0;
+          }
           if (result.status === 204) {
             lastReason = null;
             continue;
@@ -31194,11 +31276,13 @@ async function runRemoteWorker(config2, deps = {}) {
           }
         } catch (error62) {
           if (controller.signal.aborted) break;
+          if (isTransientControlPlaneError(error62)) {
+            await backOffClaim(error62);
+            continue;
+          }
           if (error62 instanceof WorkerHttpError) {
             fail2(error62, "claim");
             if (controller.signal.aborted) break;
-          } else {
-            log("claim failed", { error: error62 instanceof Error ? error62.message : String(error62) });
           }
           await sleep2(5e3);
         }
@@ -43330,6 +43414,8 @@ async function main() {
   const runnerImage = defaultRunnerImage();
   const dockerSocket = process.env.AGLIO_WORKER_DOCKER_SOCKET?.trim() || "/var/run/docker.sock";
   const mcpPort = Number(process.env.AGLIO_WORKER_MCP_PORT ?? 4310);
+  const claimTimeoutRaw = Number(process.env.AGLIO_WORKER_CLAIM_TIMEOUT_SECONDS);
+  const claimTimeoutSeconds = Number.isFinite(claimTimeoutRaw) && claimTimeoutRaw > 0 ? claimTimeoutRaw : void 0;
   const agentContainerId = process.env.HOSTNAME?.trim() || hostname3();
   const runnerNetwork = mode === "docker" ? resolveRunnerNetworkMode({
     override: process.env.AGLIO_WORKER_RUNNER_NETWORK,
@@ -43362,6 +43448,7 @@ async function main() {
     stateFile,
     // Simulators and Xcode share one signing context on a Mac host.
     capacity: workerCapacityForMode(process.env.AGLIO_WORKER_CAPACITY ?? 1, mode),
+    claimTimeoutSeconds,
     workspaceRoot,
     hostWorkspaceRoot,
     runnerImage,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "aglio-worker",
-  "version": "0.1.0",
+  "version": "0.6.0",
   "description": "Host machine agent for Aglio: claims jobs over the API and starts aglio-runner containers.",
   "type": "module",
   "bin": {