@ricsam/r5d-worker 0.0.178 → 0.0.180

package/dist/cjs/package.json

@@ -1,5 +1,5 @@
 {
   "name": "@ricsam/r5d-worker",
-  "version": "0.0.178",
+  "version": "0.0.180",
   "type": "commonjs"
 }

package/dist/mjs/internal-r5dctl.cjs

@@ -20605,7 +20605,7 @@ function resolveEntrypointPath(entrypoint) {
   }
 }
 function getR5dctlVersion() {
-  if (true) return "0.0.178";
+  if (true) return "0.0.180";
   const entrypoint = process.argv[1] ? resolveEntrypointPath(process.argv[1]) : null;
   let current = entrypoint ? import_node_path2.default.dirname(entrypoint) : process.cwd();
   for (let index = 0; index < 12; index += 1) {

package/dist/mjs/main.mjs

@@ -7,7 +7,7 @@ import { startManagerRpc } from "./runtime/releases/rpc-main.mjs";
 import { ManagerRpcConfig } from "./runtime/releases/rpc-protocol.mjs";
 const args = process.argv.slice(2);
 if (args.includes("--version")) {
-  console.log(`r5d-worker ${true ? "0.0.178" : "development"}`);
+  console.log(`r5d-worker ${true ? "0.0.180" : "development"}`);
 } else if (!args.length || args.includes("--help")) {
   console.log(
     "Usage: r5d-worker start --label <label> [--root <dir>] [--base-url <url>] [--token <worker-token>] [--initial-sync merge|reset]\n       r5d-worker executor /absolute/private/config.json\n       r5d-worker manager /absolute/private/config.json\n\nRun independently supervised durable runtime services. Provision configuration with r5dinfra."
@@ -15,7 +15,7 @@ if (args.includes("--version")) {
 } else if (args[0] === "start") {
   const runtime = await startPersonalWorker(
     parsePersonalWorkerOptions(args.slice(1)),
-    true ? "0.0.178" : "development"
+    true ? "0.0.180" : "development"
   );
   console.log(`Worker connected: ${runtime.resourceId}`);
   let closing = false;

package/dist/mjs/package.json

@@ -1,5 +1,5 @@
 {
   "name": "@ricsam/r5d-worker",
-  "version": "0.0.178",
+  "version": "0.0.180",
   "type": "module"
 }

package/dist/mjs/personal/client.mjs

@@ -25,6 +25,8 @@ class PersonalResponseError extends Error {
   rejectedBeforeAdmission;
 }
 const storageErrorCodes = /* @__PURE__ */ new Set(["invalid_input", "conflict", "not_found", "busy", "stale_fence", "unknown", "unsafe_path", "git_failed"]);
+const POLL_TIMEOUT_MS = 1e4;
+const isStaleFence = (error) => error?.code === "stale_fence";
 function parsePersonalWorkerOptions(args, environment = process.env) {
   const values = {};
   const allowed = /* @__PURE__ */ new Set(["label", "base-url", "token", "api-key", "config", "root", "initial-sync"]);
@@ -90,13 +92,13 @@ async function startPersonalWorker(options, version) {
     }
   }
   const endpoint = `/api/personal/resources/worker/${encodeURIComponent(identity.resourceId)}`;
-  async function request(pathname, body) {
+  async function request(pathname, body, timeoutMs = 3e4) {
     const response = await fetch(options.baseUrl + pathname, {
       method: "POST",
       headers: { authorization: `Bearer ${options.credential}`, "content-type": "application/json" },
       redirect: "error",
       body: JSON.stringify(body),
-      signal: AbortSignal.timeout(3e4)
+      signal: AbortSignal.timeout(timeoutMs)
     });
     const data = await response.json();
     if (!response.ok) {
@@ -105,7 +107,7 @@ async function startPersonalWorker(options, version) {
     return data;
   }
   const cli = await resolvePersonalCliEntrypoint(options.cliEntrypoint, version);
-  const bootId = randomUUID();
+  let bootId = randomUUID();
   const metadata = () => ({
     version,
     bootId,
@@ -203,19 +205,47 @@ async function startPersonalWorker(options, version) {
   const terminalStream = startTerminalStream({ url: streamUrl.toString(), credential: options.credential, version, runtime });
   const loop = (async () => {
     while (!stopped) {
+      let response;
       try {
         const sendMetadata = Date.now() - metadataAt > 6e4;
-        const response = await request(`${endpoint}/poll`, {
-          instanceId: identity.instanceId,
-          ...sendMetadata ? { metadata: metadata() } : {}
-        });
+        response = await request(
+          `${endpoint}/poll`,
+          { instanceId: identity.instanceId, ...sendMetadata ? { metadata: metadata() } : {} },
+          options.pollTimeoutMs ?? POLL_TIMEOUT_MS
+        );
         if (sendMetadata) metadataAt = Date.now();
-        await runtime.renew({ ...grant, ...response });
-        scheduler.accept(response.actions ?? []);
       } catch (error) {
         process.stderr.write(`[r5d-worker] ${safeError(error)}; reconnecting without replacing the local keeper
 `);
+        await Bun.sleep(250);
+        continue;
+      }
+      try {
+        await runtime.renew({ ...grant, ...response });
+      } catch (error) {
+        if (!isStaleFence(error)) {
+          process.stderr.write(`[r5d-worker] ${safeError(error)}; reconnecting without replacing the local keeper
+`);
+          await Bun.sleep(250);
+          continue;
+        }
+        try {
+          bootId = randomUUID();
+          const rotated = PersonalWorkerGrant.parse(
+            await request("/api/personal/resources/register", { kind: "worker", ...identity, label: options.label, metadata: metadata() })
+          );
+          metadataAt = Date.now();
+          await runtime.renew(rotated);
+          process.stderr.write(`[r5d-worker] executor authority lapsed locally; re-registered with a new boot id (epoch ${rotated.workerFence.epoch})
+`);
+        } catch (rotation) {
+          process.stderr.write(`[r5d-worker] ${safeError(rotation)}; authority rotation pending
+`);
+          await Bun.sleep(250);
+          continue;
+        }
       }
+      scheduler.accept(response.actions ?? []);
       await Bun.sleep(250);
     }
   })();
@@ -224,11 +254,17 @@ async function startPersonalWorker(options, version) {
     root,
     async close() {
       stopped = true;
-      await terminalStream.close();
-      await loop;
-      await scheduler.drain();
-      await runtime.close();
-      ledger.close();
+      try {
+        await terminalStream.close();
+        await loop;
+        await scheduler.drain();
+      } finally {
+        try {
+          await runtime.close();
+        } finally {
+          ledger.close();
+        }
+      }
     }
   };
 }

package/dist/mjs/personal/runtime.mjs

@@ -509,10 +509,18 @@ exec ${quote(executable)} ${quote(cli)} "$@"
       async close() {
         closing = true;
         clearInterval(publicationTimer);
-        await publication;
-        await tcp.close();
-        await authority.close();
-        await daemon.close();
+        try {
+          await publication;
+        } catch (error) {
+          process.stderr.write(`[r5d-worker] final publication cycle failed: ${error instanceof WorkspaceError ? error.code : "publication_failed"}
+`);
+        }
+        try {
+          await tcp.close();
+          await authority.close();
+        } finally {
+          await daemon.close();
+        }
       }
     };
   } catch (error) {

package/dist/mjs/runtime/workspace/authority.mjs

@@ -2031,6 +2031,7 @@ class WorkspaceAuthority {
     await Promise.allSettled([...this.actions]);
     await Promise.allSettled([...this.queues.values()]);
     for (const b of this.benches.values()) await this.refresh(b);
+    await this.drainLiveRuns();
     if ([...this.benches.values()].some(
       (b) => b.state.blocked || Object.values(b.state.runs).some((r) => !["completed", "cancelled", "rejected_capacity"].includes(r.state))
     ))
@@ -2043,6 +2044,32 @@ class WorkspaceAuthority {
     await fs.rm(path.join(this.config.root, ".sync-lock"), { recursive: true });
     this.closed = true;
   }
+  /** A stop ends what the keeper tracks. Live runs (an open shell, an agent
+   * command) are cancelled through the executor and waited for, the way a
+   * supervisor drain would, so that an orderly stop leaves cancelled receipts
+   * behind rather than a lock nobody can release: the process is going away
+   * either way. Outcomes that stay unknown after the drain still retain the
+   * lock for review. */
+  async drainLiveRuns() {
+    const live = (b) => Object.entries(b.state.runs).filter(([, run]) => !["completed", "cancelled", "rejected_capacity", "unknown"].includes(run.state));
+    if (![...this.benches.values()].some((b) => live(b).length)) return;
+    for (const b of this.benches.values()) {
+      for (const [operationId, run] of live(b)) {
+        try {
+          const identity = { userId: b.config.userId, sessionId: run.sessionId ?? b.config.sessionId };
+          const route = await this.route(b, identity);
+          await route.client.cancel({ ...identity, operationId, fence: route.workerFence }, `drain-${this.lockId.slice(0, 8)}-${sha256(operationId).slice(0, 16)}`);
+        } catch {
+        }
+      }
+    }
+    const deadline = Date.now() + 15e3;
+    for (; ; ) {
+      for (const b of this.benches.values()) await this.refresh(b);
+      if (![...this.benches.values()].some((b) => live(b).length) || Date.now() >= deadline) return;
+      await new Promise((resolve) => setTimeout(resolve, 200));
+    }
+  }
 }
 export {
   WorkspaceAuthority

package/dist/types/personal/client.d.ts

@@ -11,6 +11,7 @@ export type PersonalWorkerOptions = {
     root: string;
     cliEntrypoint?: string;
     initialSync?: PersonalWorkerInitialSync;
+    pollTimeoutMs?: number;
 };
 export declare function parsePersonalWorkerOptions(args: string[], environment?: Record<string, string | undefined>): PersonalWorkerOptions;
 /** HTTP reconnects never replace the local source/PTY keeper. */

package/dist/types/runtime/workspace/authority.d.ts

@@ -500,4 +500,11 @@ export declare class WorkspaceAuthority {
     cancel(identity: WorkspaceIdentity, run: RunIdentity, actionId: string): Promise<import("..").ActionReceipt>;
     close(): Promise<void>;
     private finishClose;
+    /** A stop ends what the keeper tracks. Live runs (an open shell, an agent
+     * command) are cancelled through the executor and waited for, the way a
+     * supervisor drain would, so that an orderly stop leaves cancelled receipts
+     * behind rather than a lock nobody can release: the process is going away
+     * either way. Outcomes that stay unknown after the drain still retain the
+     * lock for review. */
+    private drainLiveRuns;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ricsam/r5d-worker",
-  "version": "0.0.178",
+  "version": "0.0.180",
   "type": "module",
   "main": "./dist/mjs/main.mjs",
   "module": "./dist/mjs/main.mjs",
@@ -21,7 +21,7 @@
     "r5d-worker": "dist/mjs/main.mjs"
   },
   "dependencies": {
-    "@ricsam/r5d-api": "^0.0.178",
+    "@ricsam/r5d-api": "^0.0.180",
     "node-pty": "1.1.0",
     "zod": "^4.1.13",
     "picomatch": "^4.0.3"