@cabane/companion 0.6.24 → 0.6.26

package/dist/cli.js

@@ -6835,6 +6835,22 @@ function checkoutState(cwd) {
 function runKey(conversationId, agentId) {
   return `${conversationId}|${agentId}`;
 }
+var ERROR_BODY_LOG_CAP = 2e3;
+function describeErrorBody(body) {
+  if (body === void 0 || body === null) return void 0;
+  let text;
+  if (typeof body === "string") {
+    text = body;
+  } else {
+    try {
+      text = JSON.stringify(body);
+    } catch {
+      text = String(body);
+    }
+  }
+  if (text.length === 0) return void 0;
+  return text.length > ERROR_BODY_LOG_CAP ? `${text.slice(0, ERROR_BODY_LOG_CAP)}\u2026` : text;
+}
 function describeSubAgentError(status2, body) {
   const code = body && typeof body === "object" && "error" in body ? String(body.error) : void 0;
   switch (code) {
@@ -6860,6 +6876,13 @@ var Dispatcher = class {
   opts;
   // SJ383: per-(conversation, agent) abort registry.
   aborts;
+  // CT1109: pairs already told, in the conversation, that their session state is
+  // being refused. The failure repeats every single turn until someone fixes the
+  // payload, so without this the notice would be a per-turn drumbeat; one notice
+  // is the signal and the rest is noise. Deliberately in-memory and unbounded-free:
+  // a companion restart re-arms it, which is the behaviour we want — a restart is
+  // exactly when it's worth re-stating that memory is still being lost.
+  sessionWriteNotified = /* @__PURE__ */ new Set();
   notifyStart(info) {
     try {
       this.opts.observer?.onStart(info);
@@ -7430,6 +7453,7 @@ ${reason}`,
     ) : null;
     let sessionWritten = false;
     let sessionDegraded = false;
+    let sessionWriteRejected = false;
     let okResult = false;
     let resultReason;
     const turnRuntime = turnContext.runtime;
@@ -7550,10 +7574,25 @@ ${reason}`,
                 { agentSessionId: event.state }
               );
             } catch (err) {
-              turnLog.warn(
-                { err: err instanceof Error ? err.message : String(err) },
-                "dispatcher: session-id write failed (will retry next turn)"
-              );
+              const status2 = err instanceof ApiError ? err.status : void 0;
+              if (status2 !== void 0 && status2 >= 400 && status2 < 500) {
+                sessionWriteRejected = true;
+                turnLog.error(
+                  {
+                    err: err instanceof Error ? err.message : String(err),
+                    status: status2,
+                    responseBody: describeErrorBody(err instanceof ApiError ? err.body : void 0),
+                    stateLength: event.state.length,
+                    runtime: turnRuntime
+                  },
+                  "dispatcher: session-state write REJECTED (4xx) \u2014 this turn is not resumable and the next turn starts with no memory of it"
+                );
+              } else {
+                turnLog.warn(
+                  { err: err instanceof Error ? err.message : String(err) },
+                  "dispatcher: session-id write failed (will retry next turn)"
+                );
+              }
             }
           }
         } else if (event.type === "result") {
@@ -7680,6 +7719,10 @@ ${reason}`,
       if (sessionDegraded) {
         body.degraded = true;
       }
+      if (sessionWriteRejected && !this.sessionWriteNotified.has(key)) {
+        body.sessionWriteRejected = true;
+        this.sessionWriteNotified.add(key);
+      }
       this.opts.connectorHealth?.recordSettle(turnRuntime, {
         ok: okResult,
         errorReason: body.errorReason ?? null
@@ -8334,6 +8377,7 @@ var CompanionSupervisor = class {
   refreshing = false;
   stopped = false;
   draining = false;
+  restartPending = false;
   // CT484: latch so the companion/server version-skew warning is logged once, not
   // on every 30s heartbeat.
   versionSkewWarned = false;
@@ -8444,7 +8488,8 @@ var CompanionSupervisor = class {
         // actually landed (`harnessSignals` non-null) — an absent field means "we
         // didn't look this beat" and leaves the server's stored suggestion alone,
         // the same fail-soft contract `models` keeps.
-        ...this.harnessSignals ? { detectedRuntimes: detectedRuntimesFor(deriveHarnessSnapshot(this.harnessSignals)) } : {}
+        ...this.harnessSignals ? { detectedRuntimes: detectedRuntimesFor(deriveHarnessSnapshot(this.harnessSignals)) } : {},
+        ...this.restartPending ? { restarting: true } : {}
       });
       this.hub.setDevice({ deviceId: res.deviceId });
       this.deviceId = res.deviceId;
@@ -9082,32 +9127,41 @@ var CompanionSupervisor = class {
     );
   }
   async drainForRestart(graceMs) {
-    this.draining = true;
-    if (this.heartbeatTimer) clearInterval(this.heartbeatTimer);
-    if (this.pollTimer) clearInterval(this.pollTimer);
+    this.restartPending = true;
     if (this.inFlightHeartbeat) await this.inFlightHeartbeat;
     if (!this.deviceApi) throw new Error("cannot establish deploy drain before pairing");
-    await this.deviceApi.beginDrain();
-    for (const wr of this.workspaces.values()) wr.sub?.stop();
-    const turns = [...this.workspaces.values()].flatMap((wr) => [...wr.chains.values()]);
+    await this.sendHeartbeat();
+    const drainStartedAt = Date.now();
+    const deadline = Date.now() + Math.max(0, graceMs);
     let timedOut = false;
-    if (turns.length > 0) {
+    while (true) {
+      const turns = [...this.workspaces.values()].flatMap((wr) => [...wr.chains.values()]);
+      if (turns.length === 0) break;
+      const remainingMs = deadline - Date.now();
+      if (remainingMs <= 0) {
+        timedOut = true;
+        break;
+      }
       let timer;
       await Promise.race([
         Promise.allSettled(turns),
         new Promise((resolve) => {
-          timer = setTimeout(
-            () => {
-              timedOut = true;
-              resolve();
-            },
-            Math.max(0, graceMs)
-          );
+          timer = setTimeout(resolve, remainingMs);
           timer.unref?.();
         })
       ]);
       if (timer) clearTimeout(timer);
     }
+    this.log.info(
+      { waitedMs: Date.now() - drainStartedAt, timedOut },
+      timedOut ? "companion: deploy drain grace elapsed; fencing admission for restart" : "companion: deploy drain reached a quiet point; fencing admission for restart"
+    );
+    this.draining = true;
+    if (this.heartbeatTimer) clearInterval(this.heartbeatTimer);
+    if (this.pollTimer) clearInterval(this.pollTimer);
+    if (this.inFlightHeartbeat) await this.inFlightHeartbeat;
+    await this.deviceApi.beginDrain();
+    for (const wr of this.workspaces.values()) wr.sub?.stop();
     await Promise.allSettled(
       [...this.workspaces.values()].flatMap(
         (wr) => [...wr.agents.values()].map((agent) => agent.api.drainOutbox())

package/dist/runtime.js

@@ -6454,6 +6454,22 @@ function checkoutState(cwd) {
 function runKey(conversationId, agentId) {
   return `${conversationId}|${agentId}`;
 }
+var ERROR_BODY_LOG_CAP = 2e3;
+function describeErrorBody(body) {
+  if (body === void 0 || body === null) return void 0;
+  let text;
+  if (typeof body === "string") {
+    text = body;
+  } else {
+    try {
+      text = JSON.stringify(body);
+    } catch {
+      text = String(body);
+    }
+  }
+  if (text.length === 0) return void 0;
+  return text.length > ERROR_BODY_LOG_CAP ? `${text.slice(0, ERROR_BODY_LOG_CAP)}\u2026` : text;
+}
 function describeSubAgentError(status, body) {
   const code = body && typeof body === "object" && "error" in body ? String(body.error) : void 0;
   switch (code) {
@@ -6479,6 +6495,13 @@ var Dispatcher = class {
   opts;
   // SJ383: per-(conversation, agent) abort registry.
   aborts;
+  // CT1109: pairs already told, in the conversation, that their session state is
+  // being refused. The failure repeats every single turn until someone fixes the
+  // payload, so without this the notice would be a per-turn drumbeat; one notice
+  // is the signal and the rest is noise. Deliberately in-memory and unbounded-free:
+  // a companion restart re-arms it, which is the behaviour we want — a restart is
+  // exactly when it's worth re-stating that memory is still being lost.
+  sessionWriteNotified = /* @__PURE__ */ new Set();
   notifyStart(info) {
     try {
       this.opts.observer?.onStart(info);
@@ -7049,6 +7072,7 @@ ${reason}`,
     ) : null;
     let sessionWritten = false;
     let sessionDegraded = false;
+    let sessionWriteRejected = false;
     let okResult = false;
     let resultReason;
     const turnRuntime = turnContext.runtime;
@@ -7169,10 +7193,25 @@ ${reason}`,
                 { agentSessionId: event.state }
               );
             } catch (err) {
-              turnLog.warn(
-                { err: err instanceof Error ? err.message : String(err) },
-                "dispatcher: session-id write failed (will retry next turn)"
-              );
+              const status = err instanceof ApiError ? err.status : void 0;
+              if (status !== void 0 && status >= 400 && status < 500) {
+                sessionWriteRejected = true;
+                turnLog.error(
+                  {
+                    err: err instanceof Error ? err.message : String(err),
+                    status,
+                    responseBody: describeErrorBody(err instanceof ApiError ? err.body : void 0),
+                    stateLength: event.state.length,
+                    runtime: turnRuntime
+                  },
+                  "dispatcher: session-state write REJECTED (4xx) \u2014 this turn is not resumable and the next turn starts with no memory of it"
+                );
+              } else {
+                turnLog.warn(
+                  { err: err instanceof Error ? err.message : String(err) },
+                  "dispatcher: session-id write failed (will retry next turn)"
+                );
+              }
             }
           }
         } else if (event.type === "result") {
@@ -7299,6 +7338,10 @@ ${reason}`,
       if (sessionDegraded) {
         body.degraded = true;
       }
+      if (sessionWriteRejected && !this.sessionWriteNotified.has(key)) {
+        body.sessionWriteRejected = true;
+        this.sessionWriteNotified.add(key);
+      }
       this.opts.connectorHealth?.recordSettle(turnRuntime, {
         ok: okResult,
         errorReason: body.errorReason ?? null
@@ -7953,6 +7996,7 @@ var CompanionSupervisor = class {
   refreshing = false;
   stopped = false;
   draining = false;
+  restartPending = false;
   // CT484: latch so the companion/server version-skew warning is logged once, not
   // on every 30s heartbeat.
   versionSkewWarned = false;
@@ -8063,7 +8107,8 @@ var CompanionSupervisor = class {
         // actually landed (`harnessSignals` non-null) — an absent field means "we
         // didn't look this beat" and leaves the server's stored suggestion alone,
         // the same fail-soft contract `models` keeps.
-        ...this.harnessSignals ? { detectedRuntimes: detectedRuntimesFor(deriveHarnessSnapshot(this.harnessSignals)) } : {}
+        ...this.harnessSignals ? { detectedRuntimes: detectedRuntimesFor(deriveHarnessSnapshot(this.harnessSignals)) } : {},
+        ...this.restartPending ? { restarting: true } : {}
       });
       this.hub.setDevice({ deviceId: res.deviceId });
       this.deviceId = res.deviceId;
@@ -8701,32 +8746,41 @@ var CompanionSupervisor = class {
     );
   }
   async drainForRestart(graceMs) {
-    this.draining = true;
-    if (this.heartbeatTimer) clearInterval(this.heartbeatTimer);
-    if (this.pollTimer) clearInterval(this.pollTimer);
+    this.restartPending = true;
     if (this.inFlightHeartbeat) await this.inFlightHeartbeat;
     if (!this.deviceApi) throw new Error("cannot establish deploy drain before pairing");
-    await this.deviceApi.beginDrain();
-    for (const wr of this.workspaces.values()) wr.sub?.stop();
-    const turns = [...this.workspaces.values()].flatMap((wr) => [...wr.chains.values()]);
+    await this.sendHeartbeat();
+    const drainStartedAt = Date.now();
+    const deadline = Date.now() + Math.max(0, graceMs);
     let timedOut = false;
-    if (turns.length > 0) {
+    while (true) {
+      const turns = [...this.workspaces.values()].flatMap((wr) => [...wr.chains.values()]);
+      if (turns.length === 0) break;
+      const remainingMs = deadline - Date.now();
+      if (remainingMs <= 0) {
+        timedOut = true;
+        break;
+      }
       let timer;
       await Promise.race([
         Promise.allSettled(turns),
         new Promise((resolve) => {
-          timer = setTimeout(
-            () => {
-              timedOut = true;
-              resolve();
-            },
-            Math.max(0, graceMs)
-          );
+          timer = setTimeout(resolve, remainingMs);
           timer.unref?.();
         })
       ]);
       if (timer) clearTimeout(timer);
     }
+    this.log.info(
+      { waitedMs: Date.now() - drainStartedAt, timedOut },
+      timedOut ? "companion: deploy drain grace elapsed; fencing admission for restart" : "companion: deploy drain reached a quiet point; fencing admission for restart"
+    );
+    this.draining = true;
+    if (this.heartbeatTimer) clearInterval(this.heartbeatTimer);
+    if (this.pollTimer) clearInterval(this.pollTimer);
+    if (this.inFlightHeartbeat) await this.inFlightHeartbeat;
+    await this.deviceApi.beginDrain();
+    for (const wr of this.workspaces.values()) wr.sub?.stop();
     await Promise.allSettled(
       [...this.workspaces.values()].flatMap(
         (wr) => [...wr.agents.values()].map((agent) => agent.api.drainOutbox())

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@cabane/companion",
-  "version": "0.6.24",
+  "version": "0.6.26",
   "type": "module",
   "description": "The Cabane Companion (headless): connect a coding agent on your machine to your Cabane workspace as a responder — drive work against your own codebase, files, and MCP servers without putting any of it in Cabane.",
   "license": "UNLICENSED",