feat(cli/daemon/processmonitor): Add flexible target resolution and search command; improve restart/backoff and error handling

ts/daemon/processmonitor.ts

@@ -18,6 +18,10 @@ export class ProcessMonitor extends EventEmitter {
   private processId?: ProcessId;
   private currentLogMemorySize: number = 0;
   private readonly MAX_LOG_MEMORY_SIZE = 10 * 1024 * 1024; // 10MB
+  private restartTimer: NodeJS.Timeout | null = null;
+  private lastRetryAt: number | null = null;
+  private readonly MAX_RETRIES = 10;
+  private readonly RESET_WINDOW_MS = 60 * 60 * 1000; // 1 hour
 
   constructor(config: IMonitorConfig & { id?: ProcessId }) {
     super();
@@ -132,10 +136,7 @@ export class ProcessMonitor extends EventEmitter {
         this.emit('exit', code, signal);
 
         if (!this.stopped) {
-          this.logger.info('Restarting process...');
-          this.log('Restarting process...');
-          this.restartCount++;
-          this.spawnProcess();
+          this.scheduleRestart('exit');
         } else {
           this.logger.debug(
             'Not restarting process because monitor is stopped',
@@ -164,10 +165,7 @@ export class ProcessMonitor extends EventEmitter {
       }
 
       if (!this.stopped) {
-        this.logger.info('Restarting process due to error...');
-        this.log('Restarting process due to error...');
-        this.restartCount++;
-        this.spawnProcess();
+        this.scheduleRestart('error');
       } else {
         this.logger.debug('Not restarting process because monitor is stopped');
       }
@@ -185,6 +183,49 @@ export class ProcessMonitor extends EventEmitter {
     }
   }
 
+  /**
+   * Schedule a restart with incremental debounce and failure cutoff.
+   */
+  private scheduleRestart(reason: 'exit' | 'error'): void {
+    const now = Date.now();
+    // Reset window: if last retry was more than 1 hour ago, reset counter
+    if (this.lastRetryAt && now - this.lastRetryAt >= this.RESET_WINDOW_MS) {
+      this.logger.info('Resetting retry counter after 1 hour window');
+      this.restartCount = 0;
+    }
+
+    // Already at or above max retries?
+    if (this.restartCount >= this.MAX_RETRIES) {
+      const msg = 'Maximum restart attempts reached. Marking process as failed.';
+      this.logger.warn(msg);
+      this.log(msg);
+      this.stopped = true;
+      // Emit a specific event so manager can set status to errored
+      this.emit('failed');
+      return;
+    }
+
+    // Increment and compute delay (1..10 seconds)
+    this.restartCount++;
+    const delaySec = Math.min(this.restartCount, 10);
+    const msg = `Restarting process in ${delaySec}s (attempt ${this.restartCount}/${this.MAX_RETRIES}) due to ${reason}...`;
+    this.logger.info(msg);
+    this.log(msg);
+
+    // Clear existing timer if any, then schedule
+    if (this.restartTimer) {
+      clearTimeout(this.restartTimer);
+    }
+    this.lastRetryAt = now;
+    this.restartTimer = setTimeout(() => {
+      // If stopped in the meantime, do not spawn
+      if (this.stopped) {
+        return;
+      }
+      this.spawnProcess();
+    }, delaySec * 1000);
+  }
+
   /**
    * Monitor the process group's memory usage. If the total memory exceeds the limit,
    * kill the process group so that the 'exit' handler can restart it.