Merge pull request #721 from roboflow/revert-717-additional-prometheu…

…s-stats

Revert "Add Prometheus CPU/GPU custom metrics and latency"

inference/core/interfaces/http/http_api.py

@@ -13,7 +13,6 @@
 from fastapi.staticfiles import StaticFiles
 from fastapi_cprofile.profiler import CProfileMiddleware
 from prometheus_fastapi_instrumentator import Instrumentator
-from prometheus_fastapi_instrumentator import metrics as prom_metrics
 from starlette.convertors import StringConvertor, register_url_convertor
 from starlette.middleware.base import BaseHTTPMiddleware
 
@@ -184,10 +183,6 @@
     MessageToBigError,
 )
 from inference.core.managers.base import ModelManager
-from inference.core.managers.metrics import (
-    prom_cpu_utilization_total,
-    prom_gpu_utilization_total,
-)
 from inference.core.roboflow_api import (
     get_roboflow_dataset_type,
     get_roboflow_workspace,
@@ -497,15 +492,7 @@ def __init__(
         )
 
         if ENABLE_PROMETHEUS:
-            instrumentator = Instrumentator().instrument(app)
-            instrumentator.add(prom_cpu_utilization_total())
-            instrumentator.add(prom_gpu_utilization_total())
-            instrumentator.add(
-                prom_metrics.latency(
-                    buckets=(1,),
-                )
-            )
-            instrumentator.expose(app, endpoint="/metrics")
+            Instrumentator().expose(app, endpoint="/metrics")
 
         if METLO_KEY:
             app.add_middleware(

inference/core/managers/metrics.py

@@ -1,23 +1,13 @@
-import os
 import platform
 import re
 import socket
 import time
 import uuid
-from typing import Callable
-
-import GPUtil
-from prometheus_client import REGISTRY, Gauge
-from prometheus_fastapi_instrumentator.metrics import Info
 
 from inference.core.cache import cache
 from inference.core.logger import logger
 from inference.core.version import __version__
 
-previous_cpu_total = None
-previous_time = None
-NUM_CPU_CORES = os.cpu_count()
-
 
 def get_model_metrics(
     inference_server_id: str, model_id: str, min: float = -1, max: float = float("inf")
@@ -109,48 +99,3 @@ def get_inference_results_for_model(
         inference_results.append({"request_time": score, "inference": result})
 
     return inference_results
-
-
-def prom_cpu_utilization_total() -> Callable[[Info], None]:
-    cpu_utilization_gauge = Gauge(
-        "process_cpu_utilization_total", "Total CPU utilization"
-    )
-
-    def instrumentation(info: Info) -> None:
-        global previous_cpu_total, previous_time
-        cpu_metric = REGISTRY.get_sample_value("process_cpu_seconds_total")
-        if cpu_metric is None:
-            return
-        current_time = time.time()
-        if previous_cpu_total is None:
-            previous_time = current_time
-            previous_cpu_total = cpu_metric
-        else:
-            cpu_delta = cpu_metric - previous_cpu_total
-            time_delta = current_time - previous_time
-            if time_delta > 0:
-                cpu_utilization_percent = 100 * (cpu_delta / time_delta) / NUM_CPU_CORES
-                cpu_utilization_gauge.set(cpu_utilization_percent)
-            previous_cpu_total = cpu_metric
-            previous_time = current_time
-
-    return instrumentation
-
-
-def prom_gpu_utilization_total() -> Callable[[Info], None]:
-    gpu_load_gauge = Gauge("gpu_load_percentage", "GPU Load", ["gpu_id"])
-    gpu_memory_gauge = Gauge(
-        "gpu_memory_utilization", "GPU Memory Utilization", ["gpu_id"]
-    )
-    gpu_temp_gauge = Gauge("gpu_temperature_celsius", "GPU Temperature", ["gpu_id"])
-
-    def instrumentation(info: Info) -> None:
-        gpus = GPUtil.getGPUs()
-        if not gpus:
-            return
-        for gpu in gpus:
-            gpu_load_gauge.labels(gpu_id=gpu.id).set(gpu.load * 100)
-            gpu_memory_gauge.labels(gpu_id=gpu.id).set(gpu.memoryUtil * 100)
-            gpu_temp_gauge.labels(gpu_id=gpu.id).set(gpu.temperature)
-
-    return instrumentation