@bridge4dev/runner 0.36.0 → 0.38.0

package/dist/version.d.ts

@@ -1,2 +1,2 @@
-export declare const RUNNER_VERSION = "0.36.0";
+export declare const RUNNER_VERSION = "0.38.0";
 //# sourceMappingURL=version.d.ts.map

package/dist/version.js

@@ -1,3 +1,3 @@
 // Kept in sync with package.json by the release script (manual for now).
-export const RUNNER_VERSION = '0.36.0';
+export const RUNNER_VERSION = '0.38.0';
 //# sourceMappingURL=version.js.map

package/dist/ws-client.d.ts

@@ -1,4 +1,32 @@
 import { type GatewayFrame, type RunnerFrame } from './protocol.js';
+/**
+ * Mirror of `RUNNER_LIVENESS_*` in `@devbridge/shared` — the DevBridge side is the
+ * source of truth, exactly like `protocol.ts` mirrors the API's wire types.
+ * Copied rather than imported because this package is published to npm on its own
+ * (see the note in `recipe-schema.ts`).
+ *
+ * No inbound traffic for this long means the path is half-dead even though the
+ * socket still looks OPEN (QA-96 F16). The threshold has to stay comfortably
+ * BELOW the gateway's own budget (`RUNNER_GATEWAY_BUDGET_MS`, 60 s): a re-connect
+ * that lands before the gateway gives up is free — it replaces the old socket and
+ * the machine never leaves ONLINE — while one that lands after costs the user a
+ * «server is offline» banner over a working machine (QA-162 F2).
+ *
+ * It used to say «the gateway pings every 30s» and wait 90 s. The gateway has
+ * pinged every 15 s since devreport #117, so that 90 s had quietly become six
+ * missed pings instead of the three it was written for, and the runner sat blind
+ * for 90–120 s on every drop. Both ends now read one set of numbers.
+ */
+export declare const LIVENESS_TIMEOUT_MS = 35000;
+export declare const LIVENESS_CHECK_MS = 5000;
+/** Kept in step with `RUNNER_GATEWAY_BUDGET_MS`; asserted by `ws-liveness.test.ts`. */
+export declare const GATEWAY_BUDGET_MS = 60000;
+/**
+ * Worst-case cost of acting on the threshold: the check granularity, the first
+ * backoff step with its jitter, and a TLS+upgrade handshake. The invariant test
+ * uses it, because «the runner re-dials first» is only true with this included.
+ */
+export declare const RECONNECT_COST_MS: number;
 export interface WsClientEvents {
     frame: (frame: GatewayFrame) => void;
     open: () => void;

package/dist/ws-client.js

@@ -9,10 +9,34 @@ import { RUNNER_VERSION } from './version.js';
 const BACKOFF_BASE_MS = 1_000;
 const BACKOFF_MAX_MS = 60_000;
 const REPLACED_COOLDOWN_MS = 60_000;
-// The gateway pings every 30s; no inbound traffic for 90s means the path is
-// half-dead even though the socket looks OPEN (QA-96 F16).
-const LIVENESS_TIMEOUT_MS = 90_000;
-const LIVENESS_CHECK_MS = 30_000;
+/**
+ * Mirror of `RUNNER_LIVENESS_*` in `@devbridge/shared` — the DevBridge side is the
+ * source of truth, exactly like `protocol.ts` mirrors the API's wire types.
+ * Copied rather than imported because this package is published to npm on its own
+ * (see the note in `recipe-schema.ts`).
+ *
+ * No inbound traffic for this long means the path is half-dead even though the
+ * socket still looks OPEN (QA-96 F16). The threshold has to stay comfortably
+ * BELOW the gateway's own budget (`RUNNER_GATEWAY_BUDGET_MS`, 60 s): a re-connect
+ * that lands before the gateway gives up is free — it replaces the old socket and
+ * the machine never leaves ONLINE — while one that lands after costs the user a
+ * «server is offline» banner over a working machine (QA-162 F2).
+ *
+ * It used to say «the gateway pings every 30s» and wait 90 s. The gateway has
+ * pinged every 15 s since devreport #117, so that 90 s had quietly become six
+ * missed pings instead of the three it was written for, and the runner sat blind
+ * for 90–120 s on every drop. Both ends now read one set of numbers.
+ */
+export const LIVENESS_TIMEOUT_MS = 35_000;
+export const LIVENESS_CHECK_MS = 5_000;
+/** Kept in step with `RUNNER_GATEWAY_BUDGET_MS`; asserted by `ws-liveness.test.ts`. */
+export const GATEWAY_BUDGET_MS = 60_000;
+/**
+ * Worst-case cost of acting on the threshold: the check granularity, the first
+ * backoff step with its jitter, and a TLS+upgrade handshake. The invariant test
+ * uses it, because «the runner re-dials first» is only true with this included.
+ */
+export const RECONNECT_COST_MS = LIVENESS_CHECK_MS + BACKOFF_BASE_MS + 1_000 + 2_000;
 // Repeated HTTP 401/403 on the upgrade = bad token; slow way down.
 const AUTH_FAILURE_THRESHOLD = 5;
 const AUTH_FAILURE_COOLDOWN_MS = 10 * 60_000;
@@ -86,12 +110,17 @@ export class RunnerWsClient {
         if (this.stopped)
             return;
         log.info('ws: connecting', { url: this.wsUrl, attempt: this.attempts + 1 });
+        const dialedAt = Date.now();
         const socket = new WebSocket(this.wsUrl, {
             headers: { Authorization: `Bearer ${this.token}` },
             handshakeTimeout: 15_000,
         });
         this.socket = socket;
         socket.on('open', () => {
+            // Success used to be the only outcome that logged NOTHING, so «connected»
+            // and «hung mid-handshake» read identically in the journal — which is
+            // exactly where an hour went during QA-162. One line ends that.
+            log.info('ws: connected', { attempts: this.attempts + 1, elapsedMs: Date.now() - dialedAt });
             this.attempts = 0;
             this.authFailures = 0;
             this.lastInboundAt = Date.now();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bridge4dev/runner",
-  "version": "0.36.0",
+  "version": "0.38.0",
   "description": "DevBridge dev runner — connects a dev server to DevBridge and runs agent sessions (Claude Code / Codex)",
   "homepage": "https://bridge4.dev",
   "license": "MIT",