Add Prometheus metrics for HTTPX requests (#130)

Add a new prometheus metrics module and hook it into HTTPX event handlers. Created src/om1_utils/prometheus/__init__.py which defines Histograms and Gauges for request duration, upstream total, upstream TTFB, and proxy total (including `_last_seconds` gauges). Updated src/om1_utils/httpx/__init__.py to import these metrics, handle missing request start_time (log a warning and skip metrics), compute elapsed time, and record observe/set metrics by parsing relevant response headers (convert ms to seconds and ignore invalid values).

Author: openminddev

src/om1_utils/httpx/__init__.py

@@ -3,6 +3,17 @@
 
 import httpx
 
+from ..prometheus import (
+    om1_http_proxy_total_last_seconds,
+    om1_http_proxy_total_seconds,
+    om1_http_request_duration_last_seconds,
+    om1_http_request_duration_seconds,
+    om1_http_upstream_total_last_seconds,
+    om1_http_upstream_total_seconds,
+    om1_http_upstream_ttfb_last_seconds,
+    om1_http_upstream_ttfb_seconds,
+)
+
 
 def get_httpx_event_hooks() -> dict[str, list]:
     """
@@ -34,7 +45,14 @@ def log_response(response: httpx.Response):
         response : httpx.Response
             The HTTP response object to log.
         """
-        start_time = response.request.extensions.get("start_time", 0)
+        start_time = response.request.extensions.get("start_time", None)
+        if start_time is None:
+            logging.warning(
+                f"HTTP {response.request.method} {response.request.url} - "
+                "No start_time recorded, skipping metrics"
+            )
+            return
+
         elapsed = (time.perf_counter() - start_time) * 1000
         http_version = response.http_version
         proxy_parse_total_time = response.headers.get("x-proxy-parse-ms", "?")
@@ -52,6 +70,55 @@ def log_response(response: httpx.Response):
             f"Proxy Total Time: {proxy_total_time} ms"
         )
 
+        method = response.request.method
+        status_code = str(response.status_code)
+        host = str(response.request.url.host)
+        path = str(response.request.url.path)
+        elapsed_s = elapsed / 1000.0
+
+        om1_http_request_duration_seconds.labels(
+            host=host, path=path, method=method, status_code=status_code
+        ).observe(elapsed_s)
+        om1_http_request_duration_last_seconds.labels(
+            host=host, path=path, method=method, status_code=status_code
+        ).set(elapsed_s)
+
+        if upstream_total_time != "?":
+            try:
+                val = float(upstream_total_time) / 1000.0
+                om1_http_upstream_total_seconds.labels(
+                    host=host, path=path, method=method, status_code=status_code
+                ).observe(val)
+                om1_http_upstream_total_last_seconds.labels(
+                    host=host, path=path, method=method, status_code=status_code
+                ).set(val)
+            except ValueError:
+                pass
+
+        if upstream_ttfb_time != "?":
+            try:
+                val = float(upstream_ttfb_time) / 1000.0
+                om1_http_upstream_ttfb_seconds.labels(
+                    host=host, path=path, method=method, status_code=status_code
+                ).observe(val)
+                om1_http_upstream_ttfb_last_seconds.labels(
+                    host=host, path=path, method=method, status_code=status_code
+                ).set(val)
+            except ValueError:
+                pass
+
+        if proxy_total_time != "?":
+            try:
+                val = float(proxy_total_time) / 1000.0
+                om1_http_proxy_total_seconds.labels(
+                    host=host, path=path, method=method, status_code=status_code
+                ).observe(val)
+                om1_http_proxy_total_last_seconds.labels(
+                    host=host, path=path, method=method, status_code=status_code
+                ).set(val)
+            except ValueError:
+                pass
+
     return {
         "request": [log_request],
         "response": [log_response],

src/om1_utils/prometheus/__init__.py

@@ -0,0 +1,49 @@
+from prometheus_client import Gauge, Histogram
+
+om1_http_request_duration_seconds = Histogram(
+    "om1_http_request_duration_seconds",
+    "Total HTTP request duration (client-side) in seconds",
+    ["host", "path", "method", "status_code"],
+)
+
+om1_http_upstream_total_seconds = Histogram(
+    "om1_http_upstream_total_seconds",
+    "Upstream total time in seconds (from x-upstream-total-ms header)",
+    ["host", "path", "method", "status_code"],
+)
+
+om1_http_upstream_ttfb_seconds = Histogram(
+    "om1_http_upstream_ttfb_seconds",
+    "Upstream TTFB in seconds (from x-upstream-ttfb-ms header)",
+    ["host", "path", "method", "status_code"],
+)
+
+om1_http_proxy_total_seconds = Histogram(
+    "om1_http_proxy_total_seconds",
+    "Proxy total time in seconds (from x-proxy-total-ms header)",
+    ["host", "path", "method", "status_code"],
+)
+
+om1_http_request_duration_last_seconds = Gauge(
+    "om1_http_request_duration_last_seconds",
+    "Most recent HTTP request duration (client-side) in seconds",
+    ["host", "path", "method", "status_code"],
+)
+
+om1_http_upstream_total_last_seconds = Gauge(
+    "om1_http_upstream_total_last_seconds",
+    "Most recent upstream total time in seconds",
+    ["host", "path", "method", "status_code"],
+)
+
+om1_http_upstream_ttfb_last_seconds = Gauge(
+    "om1_http_upstream_ttfb_last_seconds",
+    "Most recent upstream TTFB in seconds",
+    ["host", "path", "method", "status_code"],
+)
+
+om1_http_proxy_total_last_seconds = Gauge(
+    "om1_http_proxy_total_last_seconds",
+    "Most recent proxy total time in seconds",
+    ["host", "path", "method", "status_code"],
+)