gridfleet-agent 0.2.2__tar.gz → 0.2.4__tar.gz

{gridfleet_agent-0.2.2 → gridfleet_agent-0.2.4}/CHANGELOG.md

@@ -2,6 +2,20 @@
 
 All notable changes to the GridFleet host agent (`gridfleet-agent` on PyPI) are documented here.
 
+## [0.2.4](https://github.com/quidow/gridfleet/compare/gridfleet-agent-v0.2.3...gridfleet-agent-v0.2.4) (2026-05-03)
+
+
+### Bug Fixes
+
+* **agent:** trigger release for port conflict cleanup ([6a561ca](https://github.com/quidow/gridfleet/commit/6a561ca480c62b9abb2d5141fa98fc4e1a7696b6))
+
+## [0.2.3](https://github.com/quidow/gridfleet/compare/gridfleet-agent-v0.2.2...gridfleet-agent-v0.2.3) (2026-05-03)
+
+
+### Bug Fixes
+
+* **agent:** prioritize node in service path ([c1d2b72](https://github.com/quidow/gridfleet/commit/c1d2b728d5d01a4a0b76f53ca48be8740de17918))
+
 ## [0.2.2](https://github.com/quidow/gridfleet/compare/gridfleet-agent-v0.2.1...gridfleet-agent-v0.2.2) (2026-05-03)
 
 

{gridfleet_agent-0.2.2 → gridfleet_agent-0.2.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gridfleet-agent
-Version: 0.2.2
+Version: 0.2.4
 Summary: GridFleet agent — runs on each host
 Project-URL: Homepage, https://github.com/quidow/gridfleet
 Project-URL: Repository, https://github.com/quidow/gridfleet

{gridfleet_agent-0.2.2 → gridfleet_agent-0.2.4}/agent_app/appium_process.py

@@ -88,6 +88,10 @@ class RuntimeNotInstalledError(RuntimeError):
     """Raised when no runtime is installed for the requested pack."""
 
 
+class AppiumPortOccupiedError(RuntimeError):
+    """Raised when a managed Appium port is now owned by another listener."""
+
+
 def resolve_appium_invocation_for_pack(
     pack_id: str,
     registry: RuntimeRegistry | None,
@@ -617,6 +621,24 @@ class AppiumProcessManager:
             )
             try:
                 restarted = await self._restart_from_launch_spec(port)
+            except AppiumPortOccupiedError:
+                self._record_restart_event(
+                    process="appium",
+                    kind="port_conflict",
+                    port=port,
+                    pid=info.pid if info is not None else None,
+                    attempt=attempt_number,
+                    delay_sec=None,
+                    exit_code=last_exit_code,
+                    will_retry=False,
+                )
+                logger.error(
+                    "Appium auto-restart stopped for connection_target=%s port=%d because the port is occupied",
+                    info.connection_target if info is not None else "unknown",
+                    port,
+                )
+                await self._drop_failed_managed_port(port)
+                return
             except Exception:
                 self._advance_restart_backoff(self._appium_restart_backoff_steps, port)
                 logger.exception("Appium auto-restart failed for port %d on attempt %d", port, attempt_number)
@@ -859,7 +881,7 @@ class AppiumProcessManager:
             appium_cmd.extend(["--allow-insecure", ",".join(spec.insecure_features)])
 
         if await self._can_connect_to_appium(spec.port):
-            raise RuntimeError(
+            raise AppiumPortOccupiedError(
                 f"Port {spec.port} is already in use by another Appium listener; "
                 "stop the existing process before starting a new managed node"
             )
@@ -1128,6 +1150,19 @@ class AppiumProcessManager:
             with contextlib.suppress(OSError):
                 os.unlink(toml_path)
 
+    async def _drop_failed_managed_port(self, port: int) -> None:
+        """Forget stale ownership for a crashed Appium process without touching an unmanaged listener."""
+        self._cancel_task(self._appium_restart_tasks, port)
+        self._cancel_task(self._appium_watch_tasks, port)
+        await self._stop_grid_node_process(port)
+        self._appium_procs.pop(port, None)
+        self._info.pop(port, None)
+        self._launch_specs.pop(port, None)
+        self._appium_restart_attempts.pop(port, None)
+        self._grid_node_restart_attempts.pop(port, None)
+        self._appium_restart_backoff_steps.pop(port, None)
+        self._grid_node_restart_backoff_steps.pop(port, None)
+
     async def stop(self, port: int) -> None:
         async with self._start_lock:
             self._intentional_stop_ports.add(port)

{gridfleet_agent-0.2.2 → gridfleet_agent-0.2.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "gridfleet-agent"
-version = "0.2.2"
+version = "0.2.4"
 description = "GridFleet agent — runs on each host"
 readme = "README.md"
 license = "Apache-2.0"

{gridfleet_agent-0.2.2 → gridfleet_agent-0.2.4}/tests/installer/test_plan.py

@@ -54,7 +54,7 @@ def test_render_config_env_includes_detected_paths_and_optional_auth() -> None:
     assert "AGENT_MANAGER_AUTH_PASSWORD=secret" in rendered
     assert "AGENT_ENABLE_WEB_TERMINAL=true" in rendered
     assert "AGENT_TERMINAL_TOKEN=terminal-token" in rendered
-    assert "PATH=/usr/bin:/opt/node/bin:" in rendered
+    assert "PATH=/opt/node/bin:/usr/bin:" in rendered
 
 
 def test_load_installed_config_reads_persisted_agent_env(tmp_path: Path) -> None:
@@ -175,10 +175,10 @@ def test_dry_run_output_names_generated_artifacts_and_warnings() -> None:
     assert "ExecStart=/opt/gridfleet-agent/venv/bin/gridfleet-agent serve" in output
 
 
-def test_build_service_path_prepends_discovered_tool_dirs() -> None:
+def test_build_service_path_prioritizes_node_before_system_dirs() -> None:
     discovery = ToolDiscovery(java_bin="/usr/lib/jvm/bin/java", node_bin_dir="/opt/node/bin", android_home="/opt/sdk")
 
-    assert build_service_path(discovery).startswith("/usr/lib/jvm/bin:/opt/node/bin:/opt/sdk/platform-tools:")
+    assert build_service_path(discovery).startswith("/opt/node/bin:/usr/lib/jvm/bin:/opt/sdk/platform-tools:")
 
 
 def test_find_node_bin_dir_prefers_home_nvm_over_system_node(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:

{gridfleet_agent-0.2.2 → gridfleet_agent-0.2.4}/tests/test_appium_process.py

@@ -705,6 +705,60 @@ async def test_auto_restart_cap_stops_retrying_after_threshold() -> None:
     assert snapshot["recent_restart_events"][0]["will_retry"] is False
 
 
+async def test_auto_restart_drops_managed_state_when_port_is_taken_by_unmanaged_listener() -> None:
+    manager = AppiumProcessManager()
+    old_appium_proc = FakeProcess(pid=1111, returncode=1)
+    old_grid_proc = FakeProcess(pid=2222)
+    manager._appium_procs[4723] = cast("asyncio.subprocess.Process", old_appium_proc)
+    manager._node_procs[4723] = cast("asyncio.subprocess.Process", old_grid_proc)
+    manager._launch_specs[4723] = AppiumLaunchSpec(
+        connection_target="device-conflict",
+        port=4723,
+        plugins=None,
+        extra_caps=None,
+        stereotype_caps=None,
+        session_override=True,
+        device_type=None,
+        ip_address=None,
+        manage_grid_node=True,
+        pack_id="appium-uiautomator2",
+        platform_id="android_mobile",
+    )
+    manager._info[4723] = AppiumProcessInfo(
+        port=4723,
+        pid=1111,
+        connection_target="device-conflict",
+        platform_id="android_mobile",
+    )
+
+    real_sleep = asyncio.sleep
+
+    async def fake_sleep(_delay: float) -> None:
+        await real_sleep(0)
+
+    with (
+        patch("agent_app.appium_process.resolve_appium_invocation_for_pack", return_value=_STUB_INVOCATION),
+        patch("agent_app.appium_process._build_env", return_value={"PATH": "/usr/bin"}),
+        patch.object(manager, "_can_connect_to_appium", new_callable=AsyncMock, return_value=True),
+        patch("agent_app.appium_process.asyncio.create_subprocess_exec", new_callable=AsyncMock) as create_proc,
+        patch("agent_app.appium_process.asyncio.sleep", side_effect=fake_sleep),
+    ):
+        await manager._auto_restart_appium(4723, exit_code=1)
+
+    create_proc.assert_not_awaited()
+    assert manager.list_running() == []
+    assert 4723 not in manager._launch_specs
+    assert 4723 not in manager._info
+    assert 4723 not in manager._node_procs
+    assert old_grid_proc.sent_signals
+    snapshot = manager.process_snapshot()
+    assert [event["kind"] for event in snapshot["recent_restart_events"]] == [
+        "crash_detected",
+        "port_conflict",
+    ]
+    assert snapshot["recent_restart_events"][-1]["process"] == "appium"
+
+
 async def test_successful_restart_resets_backoff_step_for_next_crash() -> None:
     manager = AppiumProcessManager()
     first_proc = FakeProcess(pid=1001)

{gridfleet_agent-0.2.2 → gridfleet_agent-0.2.4}/uv.lock

@@ -169,7 +169,7 @@ wheels = [
 
 [[package]]
 name = "gridfleet-agent"
-version = "0.2.2"
+version = "0.2.4"
 source = { editable = "." }
 dependencies = [
     { name = "fastapi" },

{gridfleet_agent-0.2.2 → gridfleet_agent-0.2.4}/agent_app/installer/plan.py

@@ -221,10 +221,10 @@ def discover_tools(
 
 def build_service_path(discovery: ToolDiscovery) -> str:
     prefixes: list[str] = []
-    if discovery.java_bin:
-        prefixes.append(str(Path(discovery.java_bin).parent))
     if discovery.node_bin_dir:
         prefixes.append(discovery.node_bin_dir)
+    if discovery.java_bin:
+        prefixes.append(str(Path(discovery.java_bin).parent))
     if discovery.android_home:
         prefixes.append(f"{discovery.android_home}/platform-tools")
     return ":".join([*prefixes, DEFAULT_SERVICE_PATH])