@teamvibe/poller 0.2.2 → 0.2.4

package/dist/cli/doctor.js

@@ -1,17 +1,23 @@
 import * as fs from 'fs';
 import * as path from 'path';
 import { execSync } from 'child_process';
-import { PLIST_PATH, SERVICE_LABEL, TEAMVIBE_DIR, LOGS_DIR } from './plist.js';
+import { PLIST_PATH, SERVICE_LABEL, TEAMVIBE_DIR, LOGS_DIR, isContainerRuntime } from './plist.js';
+function ok(name, message) {
+    return { name, status: 'ok', message };
+}
+function fail(name, message) {
+    return { name, status: 'fail', message };
+}
+function skipped(name, reason) {
+    return { name, status: 'skipped', message: `Skipped (${reason})` };
+}
 function check(name, fn) {
     try {
         const result = fn();
-        if (result === true) {
-            return { name, ok: true, message: 'OK' };
-        }
-        return { name, ok: true, message: result };
+        return ok(name, result === true ? 'OK' : result);
     }
     catch (error) {
-        return { name, ok: false, message: error instanceof Error ? error.message : String(error) };
+        return fail(name, error instanceof Error ? error.message : String(error));
     }
 }
 function checkNodeVersion() {
@@ -24,7 +30,9 @@ function checkNodeVersion() {
         return `${version} (>= v18)`;
     });
 }
-function checkTeamVibeDir() {
+function checkTeamVibeDir(container) {
+    if (container)
+        return skipped('TeamVibe directory', 'container');
     return check('TeamVibe directory', () => {
         if (!fs.existsSync(TEAMVIBE_DIR)) {
             throw new Error(`${TEAMVIBE_DIR} does not exist. Run 'poller install' first.`);
@@ -32,7 +40,16 @@ function checkTeamVibeDir() {
         return TEAMVIBE_DIR;
     });
 }
-function checkEnvFile() {
+function checkEnvFile(container) {
+    if (container) {
+        // In containers, check process.env instead of .env file
+        const hasToken = !!process.env['TEAMVIBE_POLLER_TOKEN'];
+        const hasDirect = !!(process.env['SQS_QUEUE_URL'] && process.env['SESSIONS_TABLE']);
+        if (!hasToken && !hasDirect) {
+            return fail('Environment config', 'Missing auth env vars (TEAMVIBE_POLLER_TOKEN or SQS_QUEUE_URL + SESSIONS_TABLE)');
+        }
+        return ok('Environment config', hasToken ? 'Token mode (env vars)' : 'Direct mode (env vars)');
+    }
     return check('Environment config', () => {
         const envPath = path.join(TEAMVIBE_DIR, '.env');
         if (!fs.existsSync(envPath)) {
@@ -48,10 +65,9 @@ function checkEnvFile() {
     });
 }
 function checkLaunchdPlist() {
+    if (process.platform !== 'darwin')
+        return skipped('Launchd plist', 'not macOS');
     return check('Launchd plist', () => {
-        if (process.platform !== 'darwin') {
-            return 'Skipped (not macOS)';
-        }
         if (!fs.existsSync(PLIST_PATH)) {
             throw new Error(`Plist not found at ${PLIST_PATH}. Run 'poller install'.`);
         }
@@ -66,10 +82,9 @@ function checkLaunchdPlist() {
     });
 }
 function checkServiceRunning() {
+    if (process.platform !== 'darwin')
+        return skipped('Service status', 'not macOS');
     return check('Service status', () => {
-        if (process.platform !== 'darwin') {
-            return 'Skipped (not macOS)';
-        }
         try {
             const output = execSync(`launchctl list | grep ${SERVICE_LABEL}`, { encoding: 'utf-8' });
             if (output.trim()) {
@@ -91,7 +106,9 @@ function checkServiceRunning() {
         }
     });
 }
-function checkLogsDir() {
+function checkLogsDir(container) {
+    if (container)
+        return skipped('Logs directory', 'container uses stdout');
     return check('Logs directory', () => {
         if (!fs.existsSync(LOGS_DIR)) {
             throw new Error(`${LOGS_DIR} does not exist`);
@@ -140,48 +157,55 @@ function checkBrainsDir() {
         return `${brains.length} brain(s) cloned`;
     });
 }
-function checkDiskSpace() {
+function checkDiskSpace(container) {
     return check('Disk space', () => {
-        try {
-            const output = execSync(`df -h "${TEAMVIBE_DIR}" | tail -1`, { encoding: 'utf-8' }).trim();
-            const parts = output.split(/\s+/);
-            const available = parts[3];
-            const capacityStr = parts[4]?.replace('%', '');
-            const capacity = parseInt(capacityStr || '0', 10);
-            if (capacity > 95) {
-                throw new Error(`Disk ${capacity}% full — only ${available} available`);
-            }
-            if (capacity > 85) {
-                return `Warning: ${capacity}% used (${available} available)`;
-            }
-            return `${available} available (${capacity}% used)`;
-        }
-        catch (error) {
-            if (error instanceof Error && error.message.includes('Disk'))
-                throw error;
-            throw new Error('Could not check disk space');
-        }
+        // Use TEAMVIBE_DIR for installed setups, fall back to / for containers or when dir doesn't exist
+        const dfTarget = !container && fs.existsSync(TEAMVIBE_DIR) ? TEAMVIBE_DIR : '/';
+        const output = execSync(`df -h "${dfTarget}" | tail -1`, { encoding: 'utf-8' }).trim();
+        const parts = output.split(/\s+/);
+        const available = parts[3];
+        const capacityStr = parts[4]?.replace('%', '');
+        const capacity = parseInt(capacityStr || '0', 10);
+        if (capacity > 95) {
+            throw new Error(`Disk ${capacity}% full — only ${available} available`);
+        }
+        if (capacity > 85) {
+            return `Warning: ${capacity}% used (${available} available)`;
+        }
+        return `${available} available (${capacity}% used)`;
     });
 }
 export async function doctor() {
-    console.log('TeamVibe Poller - Health Check\n');
+    const container = isContainerRuntime();
+    const mode = container ? 'container' : 'installed';
+    console.log(`TeamVibe Poller - Health Check (${mode})\n`);
     const results = [
         checkNodeVersion(),
-        checkTeamVibeDir(),
-        checkEnvFile(),
+        checkTeamVibeDir(container),
+        checkEnvFile(container),
         checkLaunchdPlist(),
         checkServiceRunning(),
-        checkLogsDir(),
+        checkLogsDir(container),
         checkClaudeCli(),
         checkBrainsDir(),
-        checkDiskSpace(),
+        checkDiskSpace(container),
     ];
     let hasFailure = false;
     for (const result of results) {
-        const icon = result.ok ? '\x1b[32m✓\x1b[0m' : '\x1b[31m✗\x1b[0m';
+        let icon;
+        switch (result.status) {
+            case 'ok':
+                icon = '\x1b[32m✓\x1b[0m';
+                break;
+            case 'skipped':
+                icon = '\x1b[33m–\x1b[0m';
+                break;
+            case 'fail':
+                icon = '\x1b[31m✗\x1b[0m';
+                hasFailure = true;
+                break;
+        }
         console.log(`  ${icon} ${result.name}: ${result.message}`);
-        if (!result.ok)
-            hasFailure = true;
     }
     console.log('');
     if (hasFailure) {

package/dist/cli/plist.d.ts

@@ -2,4 +2,9 @@ export declare const SERVICE_LABEL = "ai.teamvibe.poller";
 export declare const PLIST_PATH: string;
 export declare const TEAMVIBE_DIR: string;
 export declare const LOGS_DIR: string;
+/**
+ * Detect container runtime (Railway, Docker, etc.) where `poller install` was never run.
+ * In containers, env vars are injected directly — no ~/.teamvibe directory exists.
+ */
+export declare function isContainerRuntime(): boolean;
 export declare function generatePlist(nodePath: string, pollerPath: string, envVars?: Record<string, string>): string;

package/dist/cli/plist.js

@@ -1,9 +1,27 @@
+import * as fs from 'fs';
 import * as path from 'path';
 import * as os from 'os';
 export const SERVICE_LABEL = 'ai.teamvibe.poller';
 export const PLIST_PATH = path.join(os.homedir(), 'Library', 'LaunchAgents', `${SERVICE_LABEL}.plist`);
 export const TEAMVIBE_DIR = path.join(os.homedir(), '.teamvibe');
 export const LOGS_DIR = path.join(TEAMVIBE_DIR, 'logs');
+/**
+ * Detect container runtime (Railway, Docker, etc.) where `poller install` was never run.
+ * In containers, env vars are injected directly — no ~/.teamvibe directory exists.
+ */
+export function isContainerRuntime() {
+    // Railway-specific env vars
+    if (process.env['RAILWAY_ENVIRONMENT'] || process.env['RAILWAY_SERVICE_ID'])
+        return true;
+    // Generic container signals
+    if (process.env['CONTAINER'] === 'true' || process.env['DOCKER'] === 'true')
+        return true;
+    // Fallback heuristic: has auth env vars but no ~/.teamvibe directory
+    const hasEnvAuth = !!(process.env['TEAMVIBE_POLLER_TOKEN'] || (process.env['SQS_QUEUE_URL'] && process.env['SESSIONS_TABLE']));
+    if (hasEnvAuth && !fs.existsSync(TEAMVIBE_DIR))
+        return true;
+    return false;
+}
 export function generatePlist(nodePath, pollerPath, envVars) {
     let envSection = '';
     if (envVars && Object.keys(envVars).length > 0) {

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": "0.2.2",
+  "version": "0.2.4",
   "type": "module",
   "main": "./dist/index.js",
   "bin": {