@litmers/cursorflow-orchestrator 0.1.40 → 0.2.3

package/src/core/failure-policy.ts

@@ -23,7 +23,6 @@ export enum FailureType {
   DEPENDENCY_BLOCK = 'DEPENDENCY_BLOCK',
   DEPENDENCY_FAILED = 'DEPENDENCY_FAILED',
   DEPENDENCY_TIMEOUT = 'DEPENDENCY_TIMEOUT',
-  REVIEW_FAIL = 'REVIEW_FAIL',
   GIT_ERROR = 'GIT_ERROR',
   GIT_PUSH_REJECTED = 'GIT_PUSH_REJECTED',
   MERGE_CONFLICT = 'MERGE_CONFLICT',
@@ -60,60 +59,6 @@ export interface FailureAnalysis {
 /**
  * Multi-layer stall detection configuration
  */
-export interface StallDetectionConfig {
-  /** Time without stdout activity before sending continue signal */
-  idleTimeoutMs: number;
-  /** Time without state file update before considering stalled */
-  progressTimeoutMs: number;
-  /** Maximum time for a single task */
-  taskTimeoutMs: number;
-  /** Grace period for known long operations (e.g., npm install) */
-  longOperationGraceMs: number;
-  /** Patterns that indicate long operations */
-  longOperationPatterns: RegExp[];
-  /** Maximum restarts before aborting */
-  maxRestarts: number;
-}
-
-export const DEFAULT_STALL_CONFIG: StallDetectionConfig = {
-  idleTimeoutMs: 2 * 60 * 1000,       // 2 minutes without output (idle detection)
-  progressTimeoutMs: 10 * 60 * 1000,  // 10 minutes without progress
-  taskTimeoutMs: 30 * 60 * 1000,      // 30 minutes max per task
-  longOperationGraceMs: 10 * 60 * 1000, // 10 minute grace for long ops
-  longOperationPatterns: [
-    /Installing dependencies/i,
-    /npm install/i,
-    /pnpm install/i,
-    /yarn install/i,
-    /Building/i,
-    /Compiling/i,
-    /Downloading/i,
-    /Fetching/i,
-    /Cloning/i,
-    /Bundling/i,
-  ],
-  maxRestarts: 2,
-};
-
-export interface StallContext {
-  /** Current stall phase (0: normal, 1: continued, 2: stronger_prompt, 3: restarted) */
-  stallPhase: number;
-  /** Time since last activity */
-  idleTimeMs: number;
-  /** Time since last state update */
-  progressTimeMs?: number;
-  /** Last output line (for long operation detection) */
-  lastOutput?: string;
-  /** Number of restarts */
-  restartCount?: number;
-  /** Task start time */
-  taskStartTimeMs?: number;
-  /** Bytes received since last check (0 = no response at all) */
-  bytesReceived?: number;
-  /** Number of continue signals already sent */
-  continueSignalsSent?: number;
-}
-
 export interface FailureContext {
   exitCode?: number;
   stallPhase?: number;
@@ -126,167 +71,6 @@ export interface FailureContext {
   circuitBreakerName?: string;
 }
 
-/**
- * Analyze stall condition with multi-layer detection and escalating recovery
- * 
- * @deprecated Use StallDetectionService from './stall-detection' instead.
- * This function is kept for backward compatibility but will be removed in a future version.
- * 
- * The new unified StallDetectionService provides:
- * - Single source of truth for stall state
- * - Automatic recovery action execution
- * - Better heartbeat filtering
- * - Consistent state management
- * 
- * Recovery escalation stages:
- * 1. Phase 0 → Phase 1: Send continue signal (after 2 min idle)
- * 2. Phase 1 → Phase 2: Send stronger prompt (after 2 min grace)
- * 3. Phase 2 → Phase 3: Kill and restart process (after 2 min grace)
- * 4. Phase 3+: Abort after max restarts exceeded
- */
-export function analyzeStall(context: StallContext, config: StallDetectionConfig = DEFAULT_STALL_CONFIG): FailureAnalysis {
-  const { 
-    stallPhase, 
-    idleTimeMs, 
-    progressTimeMs, 
-    lastOutput, 
-    restartCount = 0, 
-    taskStartTimeMs,
-    bytesReceived = -1, // -1 means not tracked
-    continueSignalsSent = 0,
-  } = context;
-  
-  // Check if this might be a long operation
-  const isLongOperation = lastOutput && config.longOperationPatterns.some(p => p.test(lastOutput));
-  
-  // If it's a long operation but we've received 0 real bytes for a while, 
-  // reduce the grace period to avoid waiting forever for a hung process.
-  // We use 2x the normal idle timeout as a "sanity check" for silent long operations.
-  const silentLongOpCappedTimeout = config.idleTimeoutMs * 2;
-  const effectiveIdleTimeout = isLongOperation 
-    ? (bytesReceived === 0 ? Math.min(config.longOperationGraceMs, silentLongOpCappedTimeout) : config.longOperationGraceMs)
-    : config.idleTimeoutMs;
-  
-  // Check for task timeout
-  if (taskStartTimeMs && (Date.now() - taskStartTimeMs) > config.taskTimeoutMs) {
-    return {
-      type: FailureType.AGENT_TIMEOUT,
-      action: restartCount < config.maxRestarts ? RecoveryAction.KILL_AND_RESTART : RecoveryAction.RUN_DOCTOR,
-      message: `Task exceeded maximum timeout of ${Math.round(config.taskTimeoutMs / 60000)} minutes`,
-      isTransient: restartCount < config.maxRestarts,
-      details: { taskDurationMs: Date.now() - taskStartTimeMs, restartCount },
-    };
-  }
-  
-  // Check for zero bytes received (agent completely unresponsive)
-  if (bytesReceived === 0 && idleTimeMs > effectiveIdleTimeout) {
-    return {
-      type: FailureType.AGENT_NO_RESPONSE,
-      action: stallPhase < 2 ? RecoveryAction.CONTINUE_SIGNAL : RecoveryAction.KILL_AND_RESTART,
-      message: `Agent produced 0 bytes for ${Math.round(idleTimeMs / 1000)}s - possible API issue`,
-      isTransient: true,
-      details: { idleTimeMs, bytesReceived, stallPhase },
-    };
-  }
-  
-  // Check for no progress (state file not updating)
-  if (progressTimeMs && progressTimeMs > config.progressTimeoutMs) {
-    return {
-      type: FailureType.STALL_NO_PROGRESS,
-      action: stallPhase === 0 ? RecoveryAction.CONTINUE_SIGNAL : 
-              stallPhase === 1 ? RecoveryAction.STRONGER_PROMPT :
-              RecoveryAction.KILL_AND_RESTART,
-      message: `No progress for ${Math.round(progressTimeMs / 60000)} minutes`,
-      isTransient: true,
-      details: { progressTimeMs, stallPhase },
-    };
-  }
-  
-  // Phase 0: Normal operation, check for initial idle
-  if (stallPhase === 0 && idleTimeMs > effectiveIdleTimeout) {
-    return {
-      type: FailureType.STALL_IDLE,
-      action: RecoveryAction.CONTINUE_SIGNAL,
-      message: `Lane idle for ${Math.round(idleTimeMs / 1000)}s. Sending continue signal...`,
-      isTransient: true,
-      details: { idleTimeMs, isLongOperation, phase: 0 },
-    };
-  }
-  
-  // Phase 1: Continue signal sent, wait for response
-  if (stallPhase === 1) {
-    const graceTimeout = 2 * 60 * 1000; // 2 minutes grace after continue
-    
-    if (idleTimeMs > graceTimeout) {
-      return {
-        type: FailureType.STALL_IDLE,
-        action: RecoveryAction.STRONGER_PROMPT,
-        message: `Still idle after continue signal. Sending stronger prompt...`,
-        isTransient: true,
-        details: { idleTimeMs, continueSignalsSent, phase: 1 },
-      };
-    }
-  }
-  
-  // Phase 2: Stronger prompt sent, wait or escalate
-  if (stallPhase === 2) {
-    const strongerGraceTimeout = 2 * 60 * 1000; // 2 minutes grace after stronger prompt
-    
-    if (idleTimeMs > strongerGraceTimeout) {
-      if (restartCount < config.maxRestarts) {
-        return {
-          type: FailureType.STALL_IDLE,
-          action: RecoveryAction.KILL_AND_RESTART,
-          message: `No response after stronger prompt. Killing and restarting process...`,
-          isTransient: true,
-          details: { idleTimeMs, restartCount, maxRestarts: config.maxRestarts, phase: 2 },
-        };
-      } else {
-        return {
-          type: FailureType.STALL_IDLE,
-          action: RecoveryAction.RUN_DOCTOR,
-          message: `Lane failed after ${restartCount} restarts. Running diagnostics...`,
-          isTransient: false,
-          details: { restartCount, phase: 2 },
-        };
-      }
-    }
-  }
-  
-  // Phase 3+: After restart, monitor with shorter timeout
-  if (stallPhase >= 3) {
-    const postRestartTimeout = config.idleTimeoutMs * 0.75; // Shorter timeout after restart
-    
-    if (idleTimeMs > postRestartTimeout) {
-      if (restartCount < config.maxRestarts) {
-        return {
-          type: FailureType.STALL_IDLE,
-          action: RecoveryAction.CONTINUE_SIGNAL,
-          message: `Lane idle after restart. Retrying continue signal...`,
-          isTransient: true,
-          details: { idleTimeMs, restartCount, phase: stallPhase },
-        };
-      } else {
-        return {
-          type: FailureType.STALL_IDLE,
-          action: RecoveryAction.RUN_DOCTOR,
-          message: `Lane repeatedly stalled. Running diagnostics for root cause...`,
-          isTransient: false,
-          details: { stallPhase, restartCount },
-        };
-      }
-    }
-  }
-  
-  // No action needed yet
-  return {
-    type: FailureType.STALL_IDLE,
-    action: RecoveryAction.NONE,
-    message: 'Monitoring for stall',
-    isTransient: true,
-  };
-}
-
 /**
  * Analyze an error message or state to determine the failure type and recovery action
  */
@@ -311,11 +95,12 @@ export function analyzeFailure(error: string | null | undefined, context?: Failu
   // 1. Network errors
   if (msg.includes('econnreset') || msg.includes('econnrefused') || 
       msg.includes('etimedout') || msg.includes('enotfound') ||
-      msg.includes('socket hang up') || msg.includes('network')) {
+      msg.includes('socket hang up') || msg.includes('network') ||
+      msg.includes('canceled') || msg.includes('http/2') || msg.includes('stream closed')) {
     return {
       type: FailureType.NETWORK_ERROR,
       action: (context?.retryCount || 0) < 3 ? RecoveryAction.RETRY_TASK : RecoveryAction.RESTART_LANE,
-      message: 'Network error. Retrying...',
+      message: 'Network error or connection lost. Retrying...',
       isTransient: true,
       suggestedDelayMs: 5000 * Math.pow(2, context?.retryCount || 0),
     };
@@ -426,17 +211,10 @@ export function analyzeFailure(error: string | null | undefined, context?: Failu
     };
   }
 
-  // 10. Stalls (handled by phase)
-  if (context?.stallPhase !== undefined && context.stallPhase >= 0) {
-    return analyzeStall({
-      stallPhase: context.stallPhase,
-      idleTimeMs: context.idleTimeMs || 0,
-      progressTimeMs: context.progressTimeMs,
-      restartCount: context.restartCount,
-      taskStartTimeMs: context.taskStartTimeMs,
-    });
-  }
-
+  // 10. Stalls
+  // Deprecated: analyzeStall call removed. Orchestrator now uses StallDetectionService
+  // for all stall-related monitoring and recovery.
+  
   // 11. Default fallback
   return {
     type: FailureType.UNKNOWN_CRASH,