@teamvibe/poller 0.2.3 → 0.2.4

package/dist/config.d.ts

@@ -16,6 +16,7 @@ declare const configSchema: z.ZodObject<{
     IDLE_WATCHDOG_WARN_MS: z.ZodDefault<z.ZodNumber>;
     IDLE_WATCHDOG_KILL_MS: z.ZodDefault<z.ZodNumber>;
     NETWORK_ERROR_KILL_MS: z.ZodDefault<z.ZodNumber>;
+    MAX_CONSECUTIVE_NET_FAILURES: z.ZodDefault<z.ZodNumber>;
     TEAMVIBE_DATA_DIR: z.ZodDefault<z.ZodString>;
     BRAINS_PATH: z.ZodDefault<z.ZodString>;
     DEFAULT_BRAIN_PATH: z.ZodDefault<z.ZodString>;
@@ -45,6 +46,7 @@ declare const configSchema: z.ZodObject<{
     IDLE_WATCHDOG_WARN_MS: number;
     IDLE_WATCHDOG_KILL_MS: number;
     NETWORK_ERROR_KILL_MS: number;
+    MAX_CONSECUTIVE_NET_FAILURES: number;
     TEAMVIBE_DATA_DIR: string;
     BRAINS_PATH: string;
     DEFAULT_BRAIN_PATH: string;
@@ -80,6 +82,7 @@ declare const configSchema: z.ZodObject<{
     IDLE_WATCHDOG_WARN_MS?: number | undefined;
     IDLE_WATCHDOG_KILL_MS?: number | undefined;
     NETWORK_ERROR_KILL_MS?: number | undefined;
+    MAX_CONSECUTIVE_NET_FAILURES?: number | undefined;
     TEAMVIBE_DATA_DIR?: string | undefined;
     BRAINS_PATH?: string | undefined;
     DEFAULT_BRAIN_PATH?: string | undefined;
@@ -112,6 +115,7 @@ export declare const config: {
     IDLE_WATCHDOG_WARN_MS: number;
     IDLE_WATCHDOG_KILL_MS: number;
     NETWORK_ERROR_KILL_MS: number;
+    MAX_CONSECUTIVE_NET_FAILURES: number;
     TEAMVIBE_DATA_DIR: string;
     BRAINS_PATH: string;
     DEFAULT_BRAIN_PATH: string;

package/dist/config.js

@@ -24,7 +24,8 @@ const configSchema = z.object({
     LOCK_RETRY_INTERVAL_MS: z.coerce.number().default(30000), // 30 seconds
     IDLE_WATCHDOG_WARN_MS: z.coerce.number().default(300000), // 5 minutes
     IDLE_WATCHDOG_KILL_MS: z.coerce.number().default(600000), // 10 minutes
-    NETWORK_ERROR_KILL_MS: z.coerce.number().default(120000), // 2 minutes of continuous network errors → kill
+    NETWORK_ERROR_KILL_MS: z.coerce.number().default(120000), // Per-session: kill stuck session after 2 min of network errors
+    MAX_CONSECUTIVE_NET_FAILURES: z.coerce.number().default(5), // Poll loop: DNS flush at N, exit(1) at 2×N consecutive failures
     // Paths
     TEAMVIBE_DATA_DIR: z.string().default(DEFAULT_DATA_DIR),
     BRAINS_PATH: z.string().default(''),

package/dist/poller.js

@@ -13,6 +13,10 @@ import { writeMessage, pickNextMessage, markProcessed, markFailed, hasUnprocesse
 import { Cron } from 'croner';
 import { isControlMessage, handleControlMessage } from './control-handler.js';
 import { checkForUpdate } from './auto-update.js';
+import { exec } from 'child_process';
+import { promisify } from 'util';
+import { platform } from 'os';
+const execAsync = promisify(exec);
 // Track active message processing
 const processingMessages = new Set();
 // Poller auto-update cron (separate from brain heartbeats)
@@ -49,6 +53,38 @@ function stopAutoUpdateCron() {
         autoUpdateJob = null;
     }
 }
+// --- Circuit breaker ---
+let consecutiveNetFailures = 0;
+const NETWORK_ERROR_CODES = ['ENOTFOUND', 'ETIMEDOUT', 'ECONNRESET', 'ECONNREFUSED'];
+function isNetworkError(error) {
+    if (error instanceof Error) {
+        const code = error.code || '';
+        const message = error.message || '';
+        return (NETWORK_ERROR_CODES.includes(code) ||
+            message.includes('getaddrinfo') ||
+            message.includes('socket hang up'));
+    }
+    return false;
+}
+async function attemptDnsFlush() {
+    const cmd = platform() === 'darwin'
+        ? 'dscacheutil -flushcache'
+        : platform() === 'linux'
+            ? 'resolvectl flush-caches'
+            : null;
+    if (!cmd) {
+        logger.warn(`Circuit breaker: DNS flush not supported on platform "${platform()}"`);
+        return;
+    }
+    logger.warn('Circuit breaker: Attempting DNS cache flush...');
+    try {
+        await execAsync(cmd, { timeout: 5000 });
+        logger.info('Circuit breaker: DNS cache flush completed');
+    }
+    catch (err) {
+        logger.warn(`Circuit breaker: DNS cache flush failed: ${err}`);
+    }
+}
 // Track which threads have an active processing loop
 const activeLoops = new Set();
 async function runInboxLoop(threadId, kbPath, queueMessage, sessionLog) {
@@ -314,6 +350,7 @@ async function pollLoop() {
             const availableSlots = config.MAX_CONCURRENT_SESSIONS - getActiveProcessCount();
             logger.debug(`Polling for up to ${availableSlots} messages...`);
             const messages = await pollMessages(availableSlots);
+            consecutiveNetFailures = 0; // Successful poll — reset circuit breaker
             if (messages.length === 0) {
                 continue;
             }
@@ -347,8 +384,34 @@ async function pollLoop() {
             Promise.all(processPromises);
         }
         catch (error) {
-            logger.error('Error in poll loop:', error);
-            await sleep(5000);
+            if (isNetworkError(error)) {
+                consecutiveNetFailures++;
+                logger.error(`Network error in poll loop (${consecutiveNetFailures}/${config.MAX_CONSECUTIVE_NET_FAILURES}):`, error);
+                if (consecutiveNetFailures === config.MAX_CONSECUTIVE_NET_FAILURES) {
+                    await attemptDnsFlush();
+                }
+                if (consecutiveNetFailures >= config.MAX_CONSECUTIVE_NET_FAILURES * 2) {
+                    logger.error(`Circuit breaker triggered: ${consecutiveNetFailures} consecutive network failures. Exiting for restart.`);
+                    emitEvent('poller.circuit_breaker', '', {
+                        reason: 'sustained_network_failure',
+                        count: consecutiveNetFailures,
+                    });
+                    // Hard timeout on shutdown — network may be down, don't hang
+                    await Promise.race([
+                        shutdown('CIRCUIT_BREAKER', 1),
+                        sleep(15000),
+                    ]);
+                    process.exit(1); // Fallback if shutdown didn't call process.exit (e.g. timeout)
+                }
+                // Exponential backoff: 5s, 10s, 20s, 40s... capped at 60s
+                const backoffMs = Math.min(5000 * Math.pow(2, consecutiveNetFailures - 1), 60000);
+                await sleep(backoffMs);
+            }
+            else {
+                // Non-network errors don't affect the circuit breaker counter
+                logger.error('Error in poll loop:', error);
+                await sleep(5000);
+            }
         }
     }
 }
@@ -357,7 +420,7 @@ function sleep(ms) {
 }
 // Graceful shutdown
 let shuttingDown = false;
-async function shutdown(signal) {
+async function shutdown(signal, exitCode = 0) {
     if (shuttingDown)
         return;
     shuttingDown = true;
@@ -382,7 +445,7 @@ async function shutdown(signal) {
     // Final flush after all processes stopped — captures errors emitted during shutdown
     await flushEventQueue();
     logger.info('Shutdown complete');
-    process.exit(0);
+    process.exit(exitCode);
 }
 export async function startPoller() {
     logger.info('TeamVibe Poller starting...');

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@teamvibe/poller",
-  "version": "0.2.3",
+  "version": "0.2.4",
   "type": "module",
   "main": "./dist/index.js",
   "bin": {