Merge 1a7078a7a2 into a8eca68438

docs/docs/frigate/installation.md

@@ -78,7 +78,7 @@ Users of the Snapcraft build of Docker cannot use storage locations outside your
 
 Frigate utilizes shared memory to store frames during processing. The default `shm-size` provided by Docker is **64MB**.
 
-The default shm size of **128MB** is fine for setups with **2 cameras** detecting at **720p**. If Frigate is exiting with "Bus error" messages, it is likely because you have too many high resolution cameras and you need to specify a higher shm size, using [`--shm-size`](https://docs.docker.com/engine/reference/run/#runtime-constraints-on-resources) (or [`service.shm_size`](https://docs.docker.com/compose/compose-file/compose-file-v2/#shm_size) in Docker Compose).
+The default shm size of **128MB** is fine for setups with **2 cameras** detecting at **720p**. If Frigate is exiting with "Bus error" messages, it is likely because you have too many high resolution cameras and you need to specify a higher shm size, using [`--shm-size`](https://docs.docker.com/engine/reference/run/#runtime-constraints-on-resources) (or [`service.shm_size`](https://docs.docker.com/compose/compose-file/compose-file-v2/#shm_size) in Docker Compose). If raising the shm size does not help, check your [process and file limits](#process-and-file-limits) as well.
 
 The Frigate container also stores logs in shm, which can take up to **40MB**, so make sure to take this into account in your math as well.
 
@@ -86,6 +86,30 @@ The Frigate container also stores logs in shm, which can take up to **40MB**, so
 
 The shm size cannot be set per container for Home Assistant Apps. However, this is probably not required since by default Home Assistant Supervisor allocates `/dev/shm` with half the size of your total memory. If your machine has 8GB of memory, chances are that Frigate will have access to up to 4GB without any additional configuration.
 
+### Process and file limits
+
+Frigate runs many processes and opens a number of shared memory files. Installs with a large number of cameras can exceed the default limits your container runtime applies.
+
+Hitting the PID limit logs `RuntimeError: can't start new thread`, often followed by a "Bus error" that makes it look like an shm sizing problem. Compare the current count against the max from inside the container:
+
+```bash
+cat /sys/fs/cgroup/pids.current
+cat /sys/fs/cgroup/pids.max
+```
+
+If these are close, raise the limit with [`--pids-limit`](https://docs.docker.com/engine/containers/resource_constraints/) (or `service.pids_limit` in Docker Compose).
+
+Running out of file descriptors logs `OSError: [Errno 24] Too many open files`. Raise the limit in Docker Compose:
+
+```yaml
+services:
+  frigate:
+    ulimits:
+      nofile:
+        soft: 65535
+        hard: 65535
+```
+
 ## Extra Steps for Specific Hardware
 
 The following sections contain additional setup steps that are only required if you are using specific hardware. If you are not using any of these hardware types, you can skip to the [Docker](#docker) installation section.
@@ -484,7 +508,6 @@ Generate a Frigate Docker Compose configuration based on your hardware and requi
 
 <DockerComposeGenerator/>
 
-
   </TabItem>
   <TabItem value="original" label="Example Docker Compose File">
 ```yaml

frigate/api/app.py

@@ -971,7 +971,11 @@ def config_set(request: Request, body: AppConfigSetBody):
                 content=(
                     {
                         "success": True,
-                        "message": "Config successfully updated, restart to apply",
+                        "message": (
+                            "Config successfully updated"
+                            if body.requires_restart == 0
+                            else "Config successfully updated, restart to apply"
+                        ),
                     }
                 ),
                 status_code=200,

frigate/output/birdseye.py

@@ -335,6 +335,7 @@ class BirdsEyeFrameManager:
 
         self.camera_layout: list[Any] = []
         self.active_cameras: set[str] = set()
+        self.layout_camera_order: list[str] = []
         self.last_output_time = 0.0
 
     def add_camera(self, cam: str) -> None:
@@ -372,6 +373,13 @@ class BirdsEyeFrameManager:
         if cam in self.cameras:
             del self.cameras[cam]
 
+    def sort_cameras(self, cameras: set[str]) -> list[str]:
+        """Sort cameras by birdseye order, falling back to name when tied."""
+        return sorted(
+            cameras,
+            key=lambda camera: (self.config.cameras[camera].birdseye.order, camera),
+        )
+
     def clear_frame(self) -> None:
         logger.debug("Clearing the birdseye frame")
         self.frame[:] = self.blank_frame
@@ -482,6 +490,7 @@ class BirdsEyeFrameManager:
             # if the layout needs to be cleared
             self.camera_layout = []
             self.active_cameras = set()
+            self.layout_camera_order = []
             self.clear_frame()
             frame_changed = True
             layout_changed = True
@@ -500,21 +509,21 @@ class BirdsEyeFrameManager:
             else:
                 reset_layout = True
 
+            sorted_active_cameras = self.sort_cameras(active_cameras)
+
+            if not reset_layout and sorted_active_cameras != self.layout_camera_order:
+                logger.debug("Birdseye camera order changed")
+                reset_layout = True
+
             if reset_layout:
                 logger.debug("Resetting Birdseye layout...")
                 self.clear_frame()
                 self.active_cameras = active_cameras
+                self.layout_camera_order = sorted_active_cameras
                 layout_changed = True  # Layout is changing due to reset
                 # this also converts added_cameras from a set to a list since we need
                 # to pop elements in order
-                active_cameras_to_add = sorted(
-                    active_cameras,
-                    # sort cameras by order and by name if the order is the same
-                    key=lambda active_camera: (
-                        self.config.cameras[active_camera].birdseye.order,
-                        active_camera,
-                    ),
-                )
+                active_cameras_to_add = sorted_active_cameras
                 if len(active_cameras) == 1:
                     # show single camera as fullscreen
                     camera = active_cameras_to_add[0]
@@ -780,6 +789,7 @@ class BirdsEyeFrameManager:
             frame_changed, layout_changed = False, False
             self.active_cameras = set()
             self.camera_layout = []
+            self.layout_camera_order = []
             print(traceback.format_exc())
 
         # if the frame was updated or the fps is too low, send frame

frigate/test/test_birdseye.py

@@ -1,8 +1,10 @@
 """Test camera user and password cleanup."""
 
+import multiprocessing as mp
 import unittest
 
-from frigate.output.birdseye import get_canvas_shape
+from frigate.config import FrigateConfig
+from frigate.output.birdseye import BirdsEyeFrameManager, get_canvas_shape
 
 
 class TestBirdseye(unittest.TestCase):
@@ -45,3 +47,70 @@ class TestBirdseye(unittest.TestCase):
         canvas_width, canvas_height = get_canvas_shape(width, height)
         assert canvas_width == width  # width will be the same
         assert canvas_height != height
+
+
+class TestBirdseyeCameraOrder(unittest.TestCase):
+    """Test that birdseye reacts to camera order changes without a restart."""
+
+    def setUp(self):
+        config = {
+            "mqtt": {"enabled": False},
+            "birdseye": {"enabled": True, "mode": "continuous"},
+            "cameras": {
+                camera: {
+                    "ffmpeg": {
+                        "inputs": [
+                            {"path": "rtsp://10.0.0.1:554/video", "roles": ["detect"]}
+                        ]
+                    },
+                    "detect": {"height": 1080, "width": 1920, "fps": 5},
+                }
+                for camera in ("back", "front", "side")
+            },
+        }
+        self.config = FrigateConfig(**config)
+        self.manager = BirdsEyeFrameManager(self.config, mp.Event())
+
+        # mark every camera as continuously active with no frame to draw, which
+        # exercises the layout without needing real yuv frames
+        for camera_data in self.manager.cameras.values():
+            camera_data["current_frame"] = None
+            camera_data["current_frame_time"] = 1.0
+            camera_data["last_active_frame"] = 1.0
+
+    def layout_order(self) -> list[str]:
+        """Return the cameras in the order the current layout renders them."""
+        return [position[0] for row in self.manager.camera_layout for position in row]
+
+    def test_layout_uses_configured_order(self):
+        """Test the layout is sorted by order, then by name when tied."""
+        self.config.cameras["side"].birdseye.order = 0
+        self.config.cameras["back"].birdseye.order = 10
+        self.config.cameras["front"].birdseye.order = 20
+
+        self.manager.update_frame()
+
+        assert self.layout_order() == ["side", "back", "front"]
+
+    def test_order_change_rebuilds_layout(self):
+        """Test a reorder relayouts even though the active cameras are unchanged."""
+        self.manager.update_frame()
+        assert self.layout_order() == ["back", "front", "side"]
+
+        # a stable active set means only an order change can reset the layout,
+        # which is what a settings reorder publishes to this process
+        self.config.cameras["side"].birdseye.order = -10
+
+        _, layout_changed = self.manager.update_frame()
+
+        assert layout_changed
+        assert self.layout_order() == ["side", "back", "front"]
+
+    def test_unchanged_order_keeps_layout(self):
+        """Test a repeat update with no order change doesn't reset the layout."""
+        self.manager.update_frame()
+
+        _, layout_changed = self.manager.update_frame()
+
+        assert not layout_changed
+        assert self.layout_order() == ["back", "front", "side"]

frigate/test/test_gpu_stats.py

@@ -21,8 +21,12 @@ class TestGpuStats(unittest.TestCase):
     @patch("frigate.util.services.time.sleep")
     @patch("frigate.util.services.time.monotonic")
     @patch("frigate.util.services._read_intel_drm_fdinfo")
-    def test_intel_gpu_stats_fdinfo(self, read_fdinfo, monotonic, sleep, get_names):
+    @patch("frigate.util.services._enumerate_drm_devices")
+    def test_intel_gpu_stats_fdinfo(
+        self, drm_devices, read_fdinfo, monotonic, sleep, get_names
+    ):
         # 1 second of wall clock between snapshots
+        drm_devices.return_value = {"0000:00:02.0": "i915"}
         monotonic.side_effect = [0.0, 1.0]
         get_names.return_value = {"0000:00:02.0": "Intel Graphics"}
 
@@ -96,13 +100,15 @@ class TestGpuStats(unittest.TestCase):
     @patch("frigate.util.services.time.sleep")
     @patch("frigate.util.services.time.monotonic")
     @patch("frigate.util.services._read_intel_drm_fdinfo")
+    @patch("frigate.util.services._enumerate_drm_devices")
     def test_intel_gpu_stats_xe_capacity(
-        self, read_fdinfo, monotonic, sleep, get_names
+        self, drm_devices, read_fdinfo, monotonic, sleep, get_names
     ):
         # Xe engines report cumulative cycles paired with total cycles, plus a
         # per-class capacity. drm-cycles-* is summed across every instance of a
         # class, so on Battlemage (capacity 2 for vcs/vecs) busy/total must be
         # divided by capacity to land in 0-100%.
+        drm_devices.return_value = {"0000:03:00.0": "xe"}
         monotonic.side_effect = [0.0, 1.0]
         get_names.return_value = {"0000:03:00.0": "Intel Arc"}
 
@@ -150,7 +156,145 @@ class TestGpuStats(unittest.TestCase):
             },
         }
 
+    @patch("frigate.stats.intel_gpu_info.intel_gpu_name_resolver.get_names")
+    @patch("frigate.util.services.time.sleep")
     @patch("frigate.util.services._read_intel_drm_fdinfo")
-    def test_intel_gpu_stats_no_clients(self, read_fdinfo):
+    @patch("frigate.util.services._enumerate_drm_devices")
+    def test_intel_gpu_stats_no_clients_reports_idle(
+        self, drm_devices, read_fdinfo, sleep, get_names
+    ):
+        # The device exists but nothing holds it open, e.g. while camera
+        # processes are restarting. This is an idle state, not an error:
+        # returning None here would latch the hwaccel error cooldown and
+        # blank GPU stats for an hour over a momentary gap.
+        drm_devices.return_value = {"0000:00:02.0": "i915"}
         read_fdinfo.return_value = {}
+        get_names.return_value = {"0000:00:02.0": "Intel Graphics"}
+
+        assert get_intel_gpu_stats(None) == {
+            "0000:00:02.0": {
+                "name": "Intel Graphics",
+                "vendor": "intel",
+                "gpu": "0.0%",
+                "mem": "-%",
+                "compute": "0.0%",
+                "dec": "0.0%",
+            },
+        }
+        # Idle short-circuits before spending the sample window
+        sleep.assert_not_called()
+        read_fdinfo.assert_called_once()
+
+    @patch("frigate.util.services.time.sleep")
+    @patch("frigate.util.services._read_intel_drm_fdinfo")
+    @patch("frigate.util.services._enumerate_drm_devices")
+    def test_intel_gpu_stats_clients_without_engine_counters(
+        self, drm_devices, read_fdinfo, sleep
+    ):
+        # i915 publishes drm-driver/drm-pdev/drm-client-id but no drm-engine-*
+        # lines while GuC submission is active on kernels older than 6.5, so
+        # clients are found with nothing to sample. Reporting idle here would
+        # be a lie, and sampling a second time cannot help.
+        drm_devices.return_value = {"0000:00:02.0": "i915"}
+        read_fdinfo.return_value = {
+            ("0000:00:02.0", "48", "1109"): {
+                "driver": "i915",
+                "pid": "1109",
+                "engines": {},
+            },
+            ("0000:00:02.0", "51", "1258"): {
+                "driver": "i915",
+                "pid": "1258",
+                "engines": {},
+            },
+        }
+
         assert get_intel_gpu_stats(None) is None
+        sleep.assert_not_called()
+        read_fdinfo.assert_called_once()
+
+    @patch("frigate.util.services._read_intel_drm_fdinfo")
+    @patch("frigate.util.services._enumerate_drm_devices")
+    def test_intel_gpu_stats_no_intel_device(self, drm_devices, read_fdinfo):
+        # Only a non-Intel GPU is visible in sysfs; /proc is never scanned
+        drm_devices.return_value = {"0000:01:00.0": "nvidia"}
+
+        assert get_intel_gpu_stats(None) is None
+        read_fdinfo.assert_not_called()
+
+    @patch("frigate.util.services._read_intel_drm_fdinfo")
+    @patch("frigate.util.services._enumerate_drm_devices")
+    @patch("frigate.util.services._resolve_intel_gpu_pdev")
+    def test_intel_gpu_stats_unresolvable_device_hint(
+        self, resolve_pdev, drm_devices, read_fdinfo
+    ):
+        # A configured intel_gpu_device that cannot be resolved is a config
+        # error, not a reason to silently fall back to reporting all GPUs
+        resolve_pdev.return_value = None
+
+        assert get_intel_gpu_stats("/dev/dri/renderD999") is None
+        drm_devices.assert_not_called()
+        read_fdinfo.assert_not_called()
+
+    @patch("frigate.util.services._read_intel_drm_fdinfo")
+    @patch("frigate.util.services._enumerate_drm_devices")
+    @patch("frigate.util.services._resolve_intel_gpu_pdev")
+    def test_intel_gpu_stats_hint_resolves_to_non_intel_gpu(
+        self, resolve_pdev, drm_devices, read_fdinfo
+    ):
+        # card numbering can reorder across reboots on multi-GPU hosts, so a
+        # configured hint may point at another vendor's card; call it out
+        # instead of reporting nothing
+        resolve_pdev.return_value = "0000:01:00.0"
+        drm_devices.return_value = {
+            "0000:00:02.0": "i915",
+            "0000:01:00.0": "nvidia",
+        }
+
+        assert get_intel_gpu_stats("/dev/dri/card0") is None
+        read_fdinfo.assert_not_called()
+
+    @patch("frigate.util.services._read_intel_drm_fdinfo")
+    @patch("frigate.util.services._enumerate_drm_devices")
+    def test_intel_gpu_stats_unreadable_proc(self, drm_devices, read_fdinfo):
+        # A scan failure (None) is a different condition than a scan that
+        # finds no clients ({}) and must not report idle
+        drm_devices.return_value = {"0000:00:02.0": "i915"}
+        read_fdinfo.return_value = None
+
+        assert get_intel_gpu_stats(None) is None
+
+    @patch("frigate.stats.intel_gpu_info.intel_gpu_name_resolver.get_names")
+    @patch("frigate.util.services.time.sleep")
+    @patch("frigate.util.services.time.monotonic")
+    @patch("frigate.util.services._read_intel_drm_fdinfo")
+    @patch("frigate.util.services._enumerate_drm_devices")
+    def test_intel_gpu_stats_clients_lost_between_samples(
+        self, drm_devices, read_fdinfo, monotonic, sleep, get_names
+    ):
+        # Clients disappearing during the sample window is transient process
+        # churn, so report idle rather than latching an error
+        drm_devices.return_value = {"0000:00:02.0": "i915"}
+        monotonic.side_effect = [0.0, 1.0]
+        get_names.return_value = {"0000:00:02.0": "Intel Graphics"}
+        read_fdinfo.side_effect = [
+            {
+                ("0000:00:02.0", "1", "100"): {
+                    "driver": "i915",
+                    "pid": "100",
+                    "engines": {"video": (5_000_000_000, 0, 1)},
+                },
+            },
+            {},
+        ]
+
+        assert get_intel_gpu_stats(None) == {
+            "0000:00:02.0": {
+                "name": "Intel Graphics",
+                "vendor": "intel",
+                "gpu": "0.0%",
+                "mem": "-%",
+                "compute": "0.0%",
+                "dec": "0.0%",
+            },
+        }

frigate/util/services.py

@@ -285,6 +285,10 @@ _XE_ENGINE_KEYS = {
     "vecs": "video-enhance",
     "ccs": "compute",
 }
+_INTEL_DRM_DRIVERS = ("i915", "xe")
+_PCI_ADDRESS_RE = re.compile(
+    r"^[0-9a-fA-F]{4}:[0-9a-fA-F]{2}:[0-9a-fA-F]{2}\.[0-9a-fA-F]$"
+)
 
 
 def _resolve_intel_gpu_pdev(device: str | None) -> str | None:
@@ -294,29 +298,70 @@ def _resolve_intel_gpu_pdev(device: str | None) -> str | None:
     if not device:
         return None
 
-    if re.match(r"^[0-9a-fA-F]{4}:[0-9a-fA-F]{2}:[0-9a-fA-F]{2}\.[0-9a-fA-F]$", device):
+    if _PCI_ADDRESS_RE.match(device):
         return device
 
     name = os.path.basename(device.rstrip("/"))
     try:
-        return os.path.basename(os.path.realpath(f"/sys/class/drm/{name}/device"))
+        pdev = os.path.basename(os.path.realpath(f"/sys/class/drm/{name}/device"))
     except OSError:
         return None
 
+    # realpath does not raise on a nonexistent node; it returns the input
+    # path unchanged, so validate the result actually looks like a PCI
+    # address before trusting it.
+    return pdev if _PCI_ADDRESS_RE.match(pdev) else None
 
-def _read_intel_drm_fdinfo(target_pdev: str | None) -> dict:
+
+def _enumerate_drm_devices() -> dict[str, str]:
+    """Map each PCI-attached DRM device to its bound kernel driver.
+
+    Reads /sys/class/drm, which reflects every GPU on the host even when only
+    some render nodes are mapped into the container, so device presence can be
+    verified without /dev access. Returns {pdev: driver}, e.g.
+    {"0000:00:02.0": "i915"}.
+    """
+    devices: dict[str, str] = {}
+
+    try:
+        entries = os.listdir("/sys/class/drm")
+    except OSError:
+        return devices
+
+    for entry in entries:
+        device_dir = f"/sys/class/drm/{entry}/device"
+        pdev = os.path.basename(os.path.realpath(device_dir))
+
+        if not _PCI_ADDRESS_RE.match(pdev):
+            continue
+
+        try:
+            driver = os.path.basename(os.readlink(f"{device_dir}/driver"))
+        except OSError:
+            continue
+
+        devices[pdev] = driver
+
+    return devices
+
+
+def _read_intel_drm_fdinfo(target_pdev: str | None) -> dict | None:
     """Snapshot DRM fdinfo for every Intel client visible in /proc.
 
     Returns a dict keyed by (pdev, drm-client-id, pid) so the same context
     seen via multiple file descriptors on a single process collapses to one
-    entry.
+    entry. Clients whose fdinfo carries no engine counters are still included
+    with an empty "engines" dict so the caller can distinguish "clients exist
+    but the kernel publishes no busyness" from "no clients at all". Returns
+    None when /proc itself cannot be scanned, which is a different failure
+    than a scan that finds nothing.
     """
     snapshot: dict = {}
 
     try:
         proc_entries = os.listdir("/proc")
     except OSError:
-        return snapshot
+        return None
 
     for entry in proc_entries:
         if not entry.isdigit():
@@ -400,41 +445,124 @@ def _read_intel_drm_fdinfo(target_pdev: str | None) -> dict:
                     except (ValueError, IndexError):
                         continue
 
-            if not engines:
-                continue
-
             snapshot[key] = {"driver": driver, "pid": entry, "engines": engines}
 
     return snapshot
 
 
+def _idle_intel_gpu_stats(
+    target_pdev: str | None, intel_pdevs: dict[str, str]
+) -> dict[str, dict[str, Any]]:
+    """Build a 0% reading for the configured (or every) Intel GPU.
+
+    Used when the device is confirmed present but no DRM client is currently
+    attached, e.g. while camera processes are restarting. That is an idle
+    state, not a collection failure, so it must produce a valid reading:
+    returning None would latch the hwaccel error cooldown and blank GPU stats
+    for an hour over a momentary gap.
+    """
+    from frigate.stats.intel_gpu_info import intel_gpu_name_resolver
+
+    names = intel_gpu_name_resolver.get_names()
+    pdevs = [target_pdev] if target_pdev else sorted(intel_pdevs)
+
+    return {
+        pdev: {
+            "name": names.get(pdev) or "Intel iGPU",
+            "vendor": "intel",
+            "gpu": "0.0%",
+            "mem": "-%",
+            "compute": "0.0%",
+            "dec": "0.0%",
+        }
+        for pdev in pdevs
+    }
+
+
 def get_intel_gpu_stats(
     intel_gpu_device: str | None,
 ) -> dict[str, dict[str, Any]] | None:
     """Get stats by reading DRM fdinfo files, bucketed per-pdev.
 
     Each DRM client FD exposes monotonic per-engine busy counters via
-    /proc/<pid>/fdinfo/<fd> (i915 since kernel 5.19, Xe since first release).
-    We sample twice and divide busy-time deltas by wall-clock to derive
-    utilization. Render/3D and Compute are pooled into "compute"; Video and
-    VideoEnhance into "dec". Overall "gpu" is the sum of those pools (clamped
-    to 100%).
+    /proc/<pid>/fdinfo/<fd>. For i915 this requires kernel 6.5 or newer:
+    earlier kernels omit the per-engine counters whenever GuC submission is
+    active, which is the default on 12th gen and newer. Xe has exposed them
+    since its first release. We sample twice and divide busy-time deltas by
+    wall-clock to derive utilization. Render/3D and Compute are pooled into
+    "compute"; Video and VideoEnhance into "dec". Overall "gpu" is the sum of
+    those pools (clamped to 100%).
 
     The return value is keyed by the GPU's drm-pdev string so multiple Intel
     GPUs in the same system are reported separately. Each entry carries a
     "name" populated from OpenVINO (falling back to the pdev) so callers can
     surface a real device name in the UI.
+
+    A device that exists but has no attached DRM clients reports an idle 0%
+    reading. None is returned only for durable failures (no Intel GPU, a bad
+    intel_gpu_device config, unreadable /proc, or a kernel that publishes no
+    counters), each of which logs a distinct warning, and the caller latches
+    it against retries for an hour.
     """
     from frigate.stats.intel_gpu_info import intel_gpu_name_resolver
 
     target_pdev = _resolve_intel_gpu_pdev(intel_gpu_device)
+    if intel_gpu_device and not target_pdev:
+        logger.warning(
+            "Unable to collect Intel GPU stats: configured intel_gpu_device %s "
+            "does not exist or could not be resolved to a PCI device",
+            intel_gpu_device,
+        )
+        return None
+
+    drm_devices = _enumerate_drm_devices()
+    intel_pdevs = {
+        pdev: driver
+        for pdev, driver in drm_devices.items()
+        if driver in _INTEL_DRM_DRIVERS
+    }
+
+    if not intel_pdevs:
+        logger.warning(
+            "Unable to collect Intel GPU stats: no Intel GPU (i915/xe) found in "
+            "/sys/class/drm. Check that the driver is loaded on the host"
+        )
+        return None
+
+    if target_pdev and target_pdev not in intel_pdevs:
+        logger.warning(
+            "Unable to collect Intel GPU stats: configured intel_gpu_device %s "
+            "resolved to %s (driver: %s), which is not an Intel GPU",
+            intel_gpu_device,
+            target_pdev,
+            drm_devices.get(target_pdev, "unknown"),
+        )
+        return None
 
     snapshot_a = _read_intel_drm_fdinfo(target_pdev)
+    if snapshot_a is None:
+        logger.warning("Unable to collect Intel GPU stats: /proc could not be read")
+        return None
+
     if not snapshot_a:
+        # No process currently holds the GPU open, e.g. while camera processes
+        # are restarting. The device is confirmed present, so report idle
+        # rather than an error; the next stats cycle re-samples normally.
+        logger.debug("No active DRM clients for Intel GPU, reporting idle")
+        return _idle_intel_gpu_stats(target_pdev, intel_pdevs)
+
+    if not any(client["engines"] for client in snapshot_a.values()):
+        # Clients exist but the kernel published no busyness for them, so
+        # there is nothing to sample and a second snapshot would not help.
+        # i915 suppresses per-client engine counters while GuC submission is
+        # active on kernels older than 6.5 (kernel commit 1324680a80eb lifted
+        # this), which covers stock Debian 12 and Ubuntu 22.04 on 12th gen
+        # and newer.
         logger.warning(
-            "Unable to collect Intel GPU stats: no DRM fdinfo entries found"
-            "%s. Check that /proc is readable and the i915/xe driver is loaded",
-            f" for pdev {target_pdev}" if target_pdev else "",
+            "Unable to collect Intel GPU stats: found %d DRM client(s) for %s but "
+            "no per-engine counters. Kernel 6.5 or newer is required.",
+            len(snapshot_a),
+            "/".join(sorted({client["driver"] for client in snapshot_a.values()})),
         )
         return None
 
@@ -443,12 +571,18 @@ def get_intel_gpu_stats(
     elapsed_ns = (time.monotonic() - start) * 1e9
 
     snapshot_b = _read_intel_drm_fdinfo(target_pdev)
-    if not snapshot_b or elapsed_ns <= 0:
-        logger.warning(
-            "Unable to collect Intel GPU stats: second DRM fdinfo sample was empty"
-        )
+    if snapshot_b is None:
+        logger.warning("Unable to collect Intel GPU stats: /proc could not be read")
         return None
 
+    if not snapshot_b or elapsed_ns <= 0:
+        # Every client disappeared during the sample window; transient by
+        # definition, so report idle instead of latching an error.
+        logger.debug(
+            "No DRM clients persisted across Intel GPU samples, reporting idle"
+        )
+        return _idle_intel_gpu_stats(target_pdev, intel_pdevs)
+
     def _new_engine_pct() -> dict[str, float]:
         return {"render": 0.0, "video": 0.0, "video-enhance": 0.0, "compute": 0.0}
 
@@ -460,6 +594,11 @@ def get_intel_gpu_stats(
         if not data_a or data_a["driver"] != data_b["driver"]:
             continue
 
+        # Skip before the setdefault below so a counter-less client cannot
+        # register its pdev on its own.
+        if not data_b["engines"]:
+            continue
+
         pdev = key[0]
         engine_pct = per_pdev_engine_pct.setdefault(pdev, _new_engine_pct())
         pid_pct = per_pdev_pid_pct.setdefault(pdev, {})
@@ -491,11 +630,12 @@ def get_intel_gpu_stats(
         pid_pct[data_b["pid"]] = pid_pct.get(data_b["pid"], 0.0) + client_total
 
     if not per_pdev_engine_pct:
-        logger.warning(
-            "Unable to collect Intel GPU stats: no per-engine counters available "
-            "(i915 requires kernel >= 5.19)"
+        # Clients were seen in both snapshots but none persisted as the same
+        # (pdev, client-id, pid); process churn, so report idle.
+        logger.debug(
+            "No DRM clients persisted across Intel GPU samples, reporting idle"
         )
-        return None
+        return _idle_intel_gpu_stats(target_pdev, intel_pdevs)
 
     names = intel_gpu_name_resolver.get_names()
     results: dict[str, dict[str, Any]] = {}

web/public/locales/en/views/system.json

@@ -107,12 +107,7 @@
       },
       "npuUsage": "NPU Usage",
       "npuMemory": "NPU Memory",
-      "npuTemperature": "NPU Temperature",
-      "intelGpuWarning": {
-        "title": "Intel GPU Stats Warning",
-        "message": "GPU stats unavailable",
-        "description": "This is a known bug in Intel's GPU stats reporting tools (intel_gpu_top) where it will break and repeatedly return a GPU usage of 0% even in cases where hardware acceleration and object detection are correctly running on the (i)GPU. This is not a Frigate bug. You can restart the host to temporarily fix the issue and confirm that the GPU is working correctly. This does not affect performance."
-      }
+      "npuTemperature": "NPU Temperature"
     },
     "otherProcesses": {
       "title": "Other Processes",

web/src/components/config-form/sectionExtras/BirdseyeCameraReorder.tsx

@@ -91,6 +91,7 @@ export default function BirdseyeCameraReorder({
     try {
       await axios.put("config/set", {
         requires_restart: 0,
+        update_topic: "config/cameras/*/birdseye",
         config_data: { cameras: cameraUpdates },
       });
       await updateConfig();

web/src/components/icons/IconPicker.tsx

@@ -153,11 +153,15 @@ export default function IconPicker({
 }
 
 type IconRendererProps = {
-  icon: IconType;
+  icon: IconType | undefined;
   size?: number;
   className?: string;
 };
 
 export function IconRenderer({ icon, size, className }: IconRendererProps) {
+  if (!icon) {
+    return null;
+  }
+
   return <>{React.createElement(icon, { size, className })}</>;
 }

web/src/components/overlay/ExportDialog.tsx

@@ -65,6 +65,7 @@ import { Textarea } from "../ui/textarea";
 import { useNavigate } from "react-router-dom";
 import { useIsAdmin } from "@/hooks/use-is-admin";
 import { isReplayCamera } from "@/utils/cameraUtil";
+import { isValidIconName } from "@/utils/iconUtil";
 
 const EXPORT_OPTIONS = [
   "1",
@@ -1066,7 +1067,11 @@ export function ExportContent({
                                 }
                               >
                                 <IconRenderer
-                                  icon={LuIcons[group.icon]}
+                                  icon={
+                                    isValidIconName(group.icon)
+                                      ? LuIcons[group.icon]
+                                      : LuIcons.LuFolder
+                                  }
                                   className="mr-2 size-4 text-secondary-foreground"
                                 />
                                 <span className="truncate">{group.name}</span>

web/src/views/system/GeneralMetrics.tsx

@@ -470,50 +470,6 @@ export default function GeneralMetrics({
     return Object.keys(series).length > 0 ? Object.values(series) : undefined;
   }, [statsHistory]);
 
-  // Check if Intel GPU has all 0% usage values (known bug)
-  const showIntelGpuWarning = useMemo(() => {
-    if (!statsHistory || statsHistory.length < 3) {
-      return false;
-    }
-
-    const hasIntelGpu = Object.values(statsHistory[0]?.gpu_usages ?? {}).some(
-      (stats) => stats.vendor === "intel",
-    );
-
-    if (!hasIntelGpu) {
-      return false;
-    }
-
-    // Check if all GPU usage values are 0% across all stats
-    let allZero = true;
-    let hasDataPoints = false;
-
-    for (const stats of statsHistory) {
-      if (!stats) {
-        continue;
-      }
-
-      Object.values(stats.gpu_usages || {}).forEach((gpuStats) => {
-        if (gpuStats.vendor !== "intel") {
-          return;
-        }
-        if (gpuStats.gpu) {
-          hasDataPoints = true;
-          const gpuValue = parseFloat(gpuStats.gpu.slice(0, -1));
-          if (!isNaN(gpuValue) && gpuValue > 0) {
-            allZero = false;
-          }
-        }
-      });
-
-      if (!allZero) {
-        break;
-      }
-    }
-
-    return hasDataPoints && allZero;
-  }, [statsHistory]);
-
   // npu stats
 
   const npuSeries = useMemo(() => {
@@ -819,46 +775,8 @@ export default function GeneralMetrics({
                 <>
                   {statsHistory.length != 0 ? (
                     <div className="rounded-lg bg-background_alt p-2.5 md:rounded-2xl">
-                      <div className="mb-5 flex flex-row items-center justify-between">
+                      <div className="mb-5">
                         {t("general.hardwareInfo.gpuUsage")}
-                        {showIntelGpuWarning && (
-                          <Popover>
-                            <PopoverTrigger asChild>
-                              <button
-                                className="flex flex-row items-center gap-1.5 text-yellow-600 focus:outline-none dark:text-yellow-500"
-                                aria-label={t(
-                                  "general.hardwareInfo.intelGpuWarning.title",
-                                )}
-                              >
-                                <CiCircleAlert
-                                  className="size-5"
-                                  aria-label={t(
-                                    "general.hardwareInfo.intelGpuWarning.title",
-                                  )}
-                                />
-                                <span className="text-sm">
-                                  {t(
-                                    "general.hardwareInfo.intelGpuWarning.message",
-                                  )}
-                                </span>
-                              </button>
-                            </PopoverTrigger>
-                            <PopoverContent className="w-80">
-                              <div className="space-y-2">
-                                <div className="font-semibold">
-                                  {t(
-                                    "general.hardwareInfo.intelGpuWarning.title",
-                                  )}
-                                </div>
-                                <div>
-                                  {t(
-                                    "general.hardwareInfo.intelGpuWarning.description",
-                                  )}
-                                </div>
-                              </div>
-                            </PopoverContent>
-                          </Popover>
-                        )}
                       </div>
                       {gpuSeries.map((series) => (
                         <ThresholdBarGraph