ContainerImages
/
prometheus-frigate-exporter
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

add labels to cpu and memory process metrics allowing to filter process by camera or detector

main
Rhys Bailey 2023-03-05 23:46:30 +11:00
parent 1deecf4797
commit 3943b72088
1 changed files with 68 additions and 55 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

123
prometheus_frigate_exporter.py
View File

@ -8,9 +8,9 @@ import sys
import logging import logging
def add_metric(metric, label, data, key, multiplier=1.0): def add_metric(metric, label, stats, key, multiplier=1.0):
try: try:
string = str(data[key]) string = str(stats[key])
value = float(re.findall(r'\d+', string)[0]) value = float(re.findall(r'\d+', string)[0])
metric.add_metric([label], value * multiplier) metric.add_metric([label], value * multiplier)
except (KeyError, TypeError, IndexError): except (KeyError, TypeError, IndexError):
@ -21,40 +21,54 @@ class CustomCollector(object):
def __init__(self, _url): def __init__(self, _url):
self.url = _url self.url = _url
self.process_stats = {}
def add_metric_process(self, metric, stats, camera_name, pid_name, process_name, cpu_or_memory, process_type):
try:
pid = str(stats[camera_name][pid_name])
label_values = [pid, camera_name, process_name, process_type]
metric.add_metric(label_values, self.process_stats[pid][cpu_or_memory])
del self.process_stats[pid][cpu_or_memory]
except (KeyError, TypeError, IndexError):
pass
def collect(self): def collect(self):
data = json.loads(urlopen(self.url).read()) stats = json.loads(urlopen(self.url).read())
self.process_stats = stats['cpu_usages']
# process stats for cameras, detectors and other
cpu_usages = GaugeMetricFamily('frigate_cpu_usage_percent', 'Process CPU usage %',
labels=['pid', 'name', 'process', 'type'])
mem_usages = GaugeMetricFamily('frigate_mem_usage_percent', 'Process memory usage %',
labels=['pid', 'name', 'process', 'type'])
# camera stats # camera stats
ffmpeg_pid = GaugeMetricFamily('frigate_ffmpeg_pid', 'PID for ffmpeg process', labels=['camera'])
capture_pid = GaugeMetricFamily('frigate_capture_pid', 'PID for the ffmpeg process that consumes this camera',
labels=['camera'])
detect_pid = GaugeMetricFamily('frigate_detect_pid', 'PID for the process that runs detection for this camera',
labels=['camera'])
camera_fps = GaugeMetricFamily('frigate_camera_fps', 'Frames per second being consumed from your camera.', camera_fps = GaugeMetricFamily('frigate_camera_fps', 'Frames per second being consumed from your camera.',
labels=['camera']) labels=['camera_name'])
detection_fps = GaugeMetricFamily('frigate_detection_fps', 'Number of times detection is run per second.', detection_fps = GaugeMetricFamily('frigate_detection_fps', 'Number of times detection is run per second.',
labels=['camera']) labels=['camera_name'])
process_fps = GaugeMetricFamily('frigate_process_fps', 'Frames per second being processed by frigate.', process_fps = GaugeMetricFamily('frigate_process_fps', 'Frames per second being processed by frigate.',
labels=['camera']) labels=['camera_name'])
skipped_fps = GaugeMetricFamily('frigate_skipped_fps', 'Frames per second skip for processing by frigate.', skipped_fps = GaugeMetricFamily('frigate_skipped_fps', 'Frames per second skip for processing by frigate.',
labels=['camera']) labels=['camera_name'])
detection_enabled = GaugeMetricFamily('frigate_detection_enabled', 'Detection enabled for camera', detection_enabled = GaugeMetricFamily('frigate_detection_enabled', 'Detection enabled for camera',
labels=['camera']) labels=['camera_name'])
for k, d in data.items(): for camera_name, camera_stats in stats.items():
add_metric(ffmpeg_pid, k, d, 'ffmpeg_pid') add_metric(camera_fps, camera_name, camera_stats, 'camera_fps')
add_metric(detect_pid, k, d, 'pid') add_metric(detection_fps, camera_name, camera_stats, 'detection_fps')
add_metric(capture_pid, k, d, 'capture_pid') add_metric(process_fps, camera_name, camera_stats, 'process_fps')
add_metric(camera_fps, k, d, 'camera_fps') add_metric(skipped_fps, camera_name, camera_stats, 'skipped_fps')
add_metric(detection_fps, k, d, 'detection_fps') add_metric(detection_enabled, camera_name, camera_stats, 'detection_enabled')
add_metric(process_fps, k, d, 'process_fps')
add_metric(skipped_fps, k, d, 'skipped_fps') self.add_metric_process(cpu_usages, stats, camera_name, 'ffmpeg_pid', 'ffmpeg', 'cpu', 'Camera')
add_metric(detection_enabled, k, d, 'detection_enabled') self.add_metric_process(cpu_usages, stats, camera_name, 'capture_pid', 'capture', 'cpu', 'Camera')
self.add_metric_process(cpu_usages, stats, camera_name, 'pid', 'detect', 'cpu', 'Camera')
self.add_metric_process(mem_usages, stats, camera_name, 'ffmpeg_pid', 'ffmpeg', 'mem', 'Camera')
self.add_metric_process(mem_usages, stats, camera_name, 'capture_pid', 'capture', 'mem', 'Camera')
self.add_metric_process(mem_usages, stats, camera_name, 'pid', 'detect', 'mem', 'Camera')
yield ffmpeg_pid
yield capture_pid
yield detect_pid
yield camera_fps yield camera_fps
yield detection_fps yield detection_fps
yield process_fps yield process_fps
@ -65,38 +79,37 @@ class CustomCollector(object):
try: try:
yield GaugeMetricFamily('frigate_detection_total_fps', yield GaugeMetricFamily('frigate_detection_total_fps',
'Sum of detection_fps across all cameras and detectors.', 'Sum of detection_fps across all cameras and detectors.',
value=data['detection_fps']) value=stats['detection_fps'])
except KeyError: except KeyError:
pass pass
detector_inference_speed = GaugeMetricFamily('frigate_detector_inference_speed_seconds', detector_inference_speed = GaugeMetricFamily('frigate_detector_inference_speed_seconds',
'Time spent running object detection in seconds.', labels=['name']) 'Time spent running object detection in seconds.', labels=['name'])
detector_pid = GaugeMetricFamily('frigate_detector_pid',
'PID for the shared process that runs object detection on the detector',
labels=['name'])
detector_detection_start = GaugeMetricFamily('frigate_detection_start', detector_detection_start = GaugeMetricFamily('frigate_detection_start',
'Detector start time (unix timestamp)', 'Detector start time (unix timestamp)',
labels=['name']) labels=['name'])
try: try:
for k, d in data['detectors'].items(): for detector_name, detector_stats in stats['detectors'].items():
add_metric(detector_inference_speed, k, d, 'inference_speed', 0.001) # ms to seconds add_metric(detector_inference_speed, detector_name, detector_stats, 'inference_speed',
add_metric(detector_pid, k, d, 'pid') 0.001) # ms to seconds
add_metric(detector_detection_start, k, d, 'detection_start') add_metric(detector_detection_start, detector_name, detector_stats, 'detection_start')
self.add_metric_process(cpu_usages, stats['detectors'], detector_name, 'pid', 'detect', 'cpu',
'Detector')
self.add_metric_process(mem_usages, stats['detectors'], detector_name, 'pid', 'detect', 'mem',
'Detector')
except KeyError: except KeyError:
pass pass
yield detector_inference_speed yield detector_inference_speed
yield detector_pid
yield detector_detection_start yield detector_detection_start
# process stats # remaining process stats
cpu_usages = GaugeMetricFamily('frigate_cpu_usage_percent', 'Process CPU usage %', labels=['pid'])
mem_usages = GaugeMetricFamily('frigate_mem_usage_percent', 'Process memory usage %', labels=['pid'])
try: try:
for k, d in data['cpu_usages'].items(): for process_id, pid_stats in self.process_stats.items():
add_metric(cpu_usages, k, d, 'cpu') add_metric(cpu_usages, process_id, pid_stats, 'cpu')
add_metric(mem_usages, k, d, 'mem') add_metric(mem_usages, process_id, pid_stats, 'mem')
except KeyError: except KeyError:
pass pass
@ -104,13 +117,13 @@ class CustomCollector(object):
yield mem_usages yield mem_usages
# gpu stats # gpu stats
gpu_usages = GaugeMetricFamily('frigate_gpu_usage_percent', 'GPU utilisation %', labels=['gpu']) gpu_usages = GaugeMetricFamily('frigate_gpu_usage_percent', 'GPU utilisation %', labels=['gpu_name'])
gpu_mem_usages = GaugeMetricFamily('frigate_gpu_mem_usage_percent', 'GPU memory usage %', labels=['gpu']) gpu_mem_usages = GaugeMetricFamily('frigate_gpu_mem_usage_percent', 'GPU memory usage %', labels=['gpu_name'])
try: try:
for k, d in data['gpu_usages'].items(): for gpu_name, gpu_stats in stats['gpu_usages'].items():
add_metric(gpu_usages, k, d, 'gpu') add_metric(gpu_usages, gpu_name, gpu_stats, 'gpu')
add_metric(gpu_usages, k, d, 'mem') add_metric(gpu_usages, gpu_name, gpu_stats, 'mem')
except KeyError: except KeyError:
pass pass
@ -123,11 +136,11 @@ class CustomCollector(object):
'Stats recorded time (unix timestamp)') 'Stats recorded time (unix timestamp)')
try: try:
s = data['service'] service_stats = stats['service']
add_metric(uptime_seconds, '', s, 'uptime') add_metric(uptime_seconds, '', service_stats, 'uptime')
add_metric(last_updated_timestamp, '', s, 'last_updated') add_metric(last_updated_timestamp, '', service_stats, 'last_updated')
info = {'latest_version': data['service']['latest_version'], 'version': data['service']['version']} info = {'latest_version': stats['service']['latest_version'], 'version': stats['service']['version']}
yield InfoMetricFamily('frigate_service', 'Frigate version info', value=info) yield InfoMetricFamily('frigate_service', 'Frigate version info', value=info)
except KeyError: except KeyError:
@ -144,11 +157,11 @@ class CustomCollector(object):
storage_used = GaugeMetricFamily('frigate_storage_used_bytes', 'Storage used bytes', labels=['storage']) storage_used = GaugeMetricFamily('frigate_storage_used_bytes', 'Storage used bytes', labels=['storage'])
try: try:
for k, d in data['service']['storage'].items(): for storage_path, storage_stats in stats['service']['storage'].items():
add_metric(storage_free, k, d, 'free', 1e6) # MB to bytes add_metric(storage_free, storage_path, storage_stats, 'free', 1e6) # MB to bytes
add_metric(storage_total, k, d, 'total', 1e6) # MB to bytes add_metric(storage_total, storage_path, storage_stats, 'total', 1e6) # MB to bytes
add_metric(storage_used, k, d, 'used', 1e6) # MB to bytes add_metric(storage_used, storage_path, storage_stats, 'used', 1e6) # MB to bytes
storage_mount_type.add_metric(k, {'mount_type': d['mount_type']}) storage_mount_type.add_metric(storage_path, {'mount_type': storage_stats['mount_type']})
except KeyError: except KeyError:
pass pass