feat(telemetry): simple metrics endpoint for prometheus

Author: NoOPeEKS

web/monitoring/telemetry.py

@@ -0,0 +1,44 @@
+from prometheus_client import Counter, Gauge
+
+REQUEST_COUNTER = Counter(
+    "app_num_requests",
+    "Total number of requests received",
+    ["endpoint", "method"],
+)
+
+MEM_GAUGE = Gauge(
+    "app_current_memory",
+    "Total RAM consumed by application",
+)
+
+UPTIME_GAUGE = Gauge(
+    "app_uptime",
+    "Uptime of the application",
+)
+
+SEARCH_CACHE_HITS_GAUGE = Gauge(
+    "app_search_cache_hits",
+    "Total hits from search cache",
+)
+SEARCH_CACHE_MISSES_GAUGE = Gauge(
+    "app_search_cache_misses",
+    "Total misses from search cache",
+)
+
+INDEX_CACHE_HITS_GAUGE = Gauge(
+    "app_index_cache_hits",
+    "Total hits from index cache",
+)
+INDEX_CACHE_MISSES_GAUGE = Gauge(
+    "app_index_cache_misses",
+    "Total misses from index cache",
+)
+
+AUTOCOMPLETE_CACHE_HITS_GAUGE = Gauge(
+    "app_autocomplete_cache_hits",
+    "Total hits from autocomplete cache",
+)
+AUTOCOMPLETE_CACHE_MISSES_GAUGE = Gauge(
+    "app_autocomplete_cache_misses",
+    "Total misses from autocomplete cache",
+)

web/routes/metrics.py

@@ -0,0 +1,47 @@
+import datetime
+import os
+
+import psutil
+from fastapi import APIRouter, Request, Response
+from prometheus_client import CONTENT_TYPE_LATEST, generate_latest
+
+from web.monitoring.telemetry import (
+    AUTOCOMPLETE_CACHE_HITS_GAUGE,
+    AUTOCOMPLETE_CACHE_MISSES_GAUGE,
+    INDEX_CACHE_HITS_GAUGE,
+    INDEX_CACHE_MISSES_GAUGE,
+    MEM_GAUGE,
+    SEARCH_CACHE_HITS_GAUGE,
+    SEARCH_CACHE_MISSES_GAUGE,
+    UPTIME_GAUGE,
+)
+from web.routes.autocomplete import _get_autocomplete
+from web.routes.index import _get_letter_index
+from web.routes.search import _get_search
+
+router = APIRouter(prefix="/metrics")
+
+
+@router.get(
+    path="",
+    summary="Returns metrics collected throughout the app for Prometheus.",
+)
+def metrics(request: Request) -> Response:
+    """
+    TODO: Docstring this.
+    """
+    rss = int(psutil.Process(os.getpid()).memory_info().rss // 1024**2)
+    MEM_GAUGE.set(rss)
+    startup_time = request.app.state.start_time
+    uptime_seconds = (datetime.datetime.now() - startup_time).total_seconds()
+    UPTIME_GAUGE.set(uptime_seconds)
+    search_cache = _get_search.cache_info()
+    index_cache = _get_letter_index.cache_info()
+    autocomplete_cache = _get_autocomplete.cache_info()
+    SEARCH_CACHE_HITS_GAUGE.set(search_cache.hits)
+    SEARCH_CACHE_MISSES_GAUGE.set(search_cache.misses)
+    INDEX_CACHE_HITS_GAUGE.set(index_cache.hits)
+    INDEX_CACHE_MISSES_GAUGE.set(index_cache.misses)
+    AUTOCOMPLETE_CACHE_HITS_GAUGE.set(autocomplete_cache.hits)
+    AUTOCOMPLETE_CACHE_MISSES_GAUGE.set(autocomplete_cache.misses)
+    return Response(content=generate_latest(), media_type=CONTENT_TYPE_LATEST)