fix: Added circuit breaker

.claude/settings.local.json

@@ -27,7 +27,15 @@
       "Bash(git push:*)",
       "Bash(git remote:*)",
       "Bash(git config:*)",
-      "Bash(git fetch:*)"
+      "Bash(git fetch:*)",
+      "mcp__serena__initial_instructions",
+      "mcp__serena__check_onboarding_performed",
+      "mcp__serena__list_dir",
+      "mcp__serena__think_about_collected_information",
+      "Bash(.venvScriptspython.exe -m pytest tests/executors/test_nextcloud_executor.py -v)",
+      "Bash(set PYTHONPATH=%cd%)",
+      "Bash(python:*)",
+      "Bash(pip install:*)"
     ],
     "deny": [],
     "ask": []

app/clients/orchestrator_client.py

@@ -707,6 +707,21 @@ class OrchestratorClient:
 
         self._invalidate_client()
 
+    def reset_circuit_breaker(self) -> None:
+        """Manually reset the circuit breaker.
+
+        Useful when retrying registration after a long wait period,
+        to give the orchestrator a fresh chance to respond.
+        """
+        if self._circuit_open_until is not None or self._consecutive_failures > 0:
+            logger.info(
+                "circuit_breaker_manual_reset",
+                was_open=self._circuit_open_until is not None,
+                previous_failures=self._consecutive_failures,
+            )
+        self._circuit_open_until = None
+        self._consecutive_failures = 0
+
     async def close(self) -> None:
         """Close the HTTP client."""
         if self._client and not self._client.is_closed:

app/main.py

@@ -103,10 +103,47 @@ async def main() -> int:
         pass
 
     try:
-        # Register with orchestrator
-        if not await agent.register():
-            logger.error("registration_failed_exit")
-            return 1
+        # Register with orchestrator - retry indefinitely until success or shutdown
+        # This ensures the agent survives orchestrator restarts/updates
+        registration_attempt = 0
+        while not shutdown_event.is_set():
+            registration_attempt += 1
+
+            # Reset circuit breaker before each attempt to give orchestrator a fresh chance
+            # This is important after waiting - orchestrator may have come back up
+            client.reset_circuit_breaker()
+
+            logger.info(
+                "registration_attempt",
+                attempt=registration_attempt,
+                message="Attempting to register with orchestrator...",
+            )
+
+            if await agent.register():
+                break  # Registration successful
+
+            # Wait before retrying, with exponential backoff up to 60 seconds
+            wait_time = min(30 * (1.5 ** min(registration_attempt - 1, 4)), 60)
+            logger.warning(
+                "registration_retry_wait",
+                attempt=registration_attempt,
+                wait_seconds=wait_time,
+                message="Orchestrator unavailable, will retry...",
+            )
+
+            # Wait but allow shutdown to interrupt
+            try:
+                await asyncio.wait_for(shutdown_event.wait(), timeout=wait_time)
+                # If we get here, shutdown was requested
+                logger.info("shutdown_during_registration")
+                return 0
+            except asyncio.TimeoutError:
+                # Normal timeout, continue to retry
+                pass
+
+        if shutdown_event.is_set():
+            logger.info("shutdown_before_registration_complete")
+            return 0
 
         # Start background tasks
         heartbeat_task = asyncio.create_task(