Add Prometheus metrics for Chord-Bot

metrics.py

@@ -0,0 +1,115 @@
+# metrics.py
+import os
+import time
+
+from prometheus_client import (
+    Counter,
+    Histogram,
+    Gauge,
+    Summary,
+    start_http_server,
+)
+
+SERVICE_NAME = os.getenv("SERVICE_NAME", "chord-bot-local")
+
+#tot requests by status
+REQUEST_COUNT = Counter(
+    "chordbot_requests_total",
+    "Total number of Chord-Bot requests",
+    ["service", "endpoint", "status"],  # status: success | invalid_input | error
+)
+
+#latency
+REQUEST_LATENCY = Histogram(
+    "chordbot_request_latency_seconds",
+    "Latency of Chord-Bot requests in seconds",
+    ["service", "endpoint"],
+)
+
+# Number of note tokens per request
+NOTES_PER_REQUEST = Histogram(
+    "chordbot_notes_per_request",
+    "Number of note tokens received per request",
+    ["service", "endpoint"],
+    
+    buckets=(1, 2, 3, 4, 5, 6, 8, 12, 16),
+)
+
+INVALID_REQUESTS = Counter(
+    "chordbot_invalid_requests_total",
+    "Number of Chord-Bot requests with invalid input",
+    ["service", "endpoint"],
+)
+
+#counts of predicted chords by label
+CHORD_PREDICTIONS = Counter(
+    "chordbot_chord_predictions_total",
+    "Number of times each chord label is predicted",
+    ["service", "endpoint", "chord_label"],
+)
+
+#how many requests
+ACTIVE_REQUESTS = Gauge(
+    "chordbot_active_requests",
+    "Number of active Chord-Bot requests being processed",
+    ["service", "endpoint"],
+)
+
+
+def init_metrics(port: int = 8000, service_name: str | None = None) -> None:
+    """
+    Start the Prometheus metrics HTTP server on `port`.
+
+    Call this once at app startup (before serving requests).
+    """
+    global SERVICE_NAME
+
+    if service_name is not None:
+        SERVICE_NAME = service_name
+
+    #expose on http://0.0.0.0:port/metrics
+    start_http_server(port)
+    print(f"[metrics] Prometheus metrics server started on :{port} for service={SERVICE_NAME}")
+
+
+def instrument_chord_request(endpoint: str, note_tokens, chord_label: str | None, status: str, elapsed_seconds: float) -> None:
+    """
+    Helper you can call from your endpoint / handler *after* processing a request.
+
+    Parameters:
+        endpoint       - logical endpoint name, e.g. "analyze" or "predict"
+        note_tokens    - list of parsed notes (e.g., ["C", "E", "G"]) or None
+        chord_label    - predicted chord label (e.g., "Cmaj") or None if invalid/failed
+        status         - "success", "invalid_input", or "error"
+        elapsed_seconds - wall-clock latency for the request
+    """
+
+    REQUEST_COUNT.labels(
+        service=SERVICE_NAME,
+        endpoint=endpoint,
+        status=status,
+    ).inc()
+
+    REQUEST_LATENCY.labels(
+        service=SERVICE_NAME,
+        endpoint=endpoint,
+    ).observe(elapsed_seconds)
+
+    if note_tokens is not None:
+        NOTES_PER_REQUEST.labels(
+            service=SERVICE_NAME,
+            endpoint=endpoint,
+        ).observe(len(note_tokens))
+
+    if status == "invalid_input":
+        INVALID_REQUESTS.labels(
+            service=SERVICE_NAME,
+            endpoint=endpoint,
+        ).inc()
+
+    if status == "success" and chord_label is not None:
+        CHORD_PREDICTIONS.labels(
+            service=SERVICE_NAME,
+            endpoint=endpoint,
+            chord_label=chord_label,
+        ).inc()