Prometheus support

README.md

@@ -196,7 +196,9 @@ docker-compose -f docker-compose.cuda.yml up -d --build
 | --suggestions               | Allow user suggestions                                                                                      | `false`    | LT_SUGGESTIONS               |
 | --disable-files-translation | Disable files translation                                                                                   | `false`    | LT_DISABLE_FILES_TRANSLATION |
 | --disable-web-ui            | Disable web ui                                                                                              | `false`    | LT_DISABLE_WEB_UI            |
-| --update-models             | Update language models at startup                                                                                    | `false`    | LT_UPDATE_MODELS            |
+| --update-models             | Update language models at startup                                                                           | `false`    | LT_UPDATE_MODELS            |
+| --metrics                   | Enable the /metrics endpoint for exporting [Prometheus](https://prometheus.io/) usage metrics               | `false`    | LT_METRICS            |
+| --metrics-auth-token        | Protect the /metrics endpoint by allowing only clients that have a valid Authorization Bearer token         | `No auth`    | LT_METRICS_AUTH_TOKEN            |
 
 Note that each argument has an equivalent environment variable that can be used instead. The env. variables overwrite the default values but have lower priority than the command arguments and are particularly useful if used with Docker. The environment variable names are the upper-snake-case of the equivalent command argument's name with a `LT` prefix.
 
@@ -267,6 +269,47 @@ ltmanage keys remove <api-key>
 ltmanage keys
 ```
 
+## Prometheus Metrics
+
+LibreTranslate has Prometheus [exporter](https://prometheus.io/docs/instrumenting/exporters/) capabilities when you pass the `--metrics` argument at startup (disabled by default). When metrics are enabled, a `/metrics` endpoint is mounted on the instance:
+
+http://localhost:5000/metrics
+
+```
+# HELP request_inprogress Multiprocess metric
+# TYPE request_inprogress gauge
+request_inprogress{api_key="",endpoint="/translate",request_ip="127.0.0.1"} 0.0
+# HELP request_seconds Multiprocess metric
+# TYPE request_seconds summary
+request_seconds_count{api_key="",endpoint="/translate",request_ip="127.0.0.1",status="200"} 0.0
+request_seconds_sum{api_key="",endpoint="/translate",request_ip="127.0.0.1",status="200"} 0.0
+```
+
+You can then configure `prometheus.yml` to read the metrics:
+
+```
+scrape_configs:
+  - job_name: "libretranslate"
+    
+    # Needed only if you use --metrics-auth-token
+    #authorization:
+      #credentials: "mytoken"
+    
+    static_configs:
+      - targets: ["localhost:5000"]
+```
+
+To secure the `/metrics` endpoint you can also use `--metrics-auth-token mytoken`.
+
+If you use Gunicorn, make sure to create a directory for storing multiprocess data metrics and set `PROMETHEUS_MULTIPROC_DIR`:
+
+```
+mkdir -p /tmp/prometheus_data
+rm /tmp/prometheus_data/*
+export PROMETHEUS_MULTIPROC_DIR=/tmp/prometheus_data 
+gunicorn -c gunicorn_conf.py --bind 0.0.0.0:5000 'wsgi:app(metrics=True)' 
+```
+
 ## Language Bindings
 
 You can use the LibreTranslate API using the following bindings:

VERSION

@@ -1 +1 @@
-1.3.4
+1.3.5

app/app.py

@@ -4,15 +4,17 @@ import tempfile
 import uuid
 from functools import wraps
 from html import unescape
+from timeit import default_timer
 
 import argostranslatefiles
 from argostranslatefiles import get_supported_formats
 from flask import (Flask, abort, jsonify, render_template, request, send_file,
-                   url_for)
+                   url_for, Response)
 from flask_swagger import swagger
 from flask_swagger_ui import get_swaggerui_blueprint
 from translatehtml import translate_html
 from werkzeug.utils import secure_filename
+from werkzeug.exceptions import HTTPException
 
 from app import flood, remove_translated_files, security
 from app.language import detect_languages, improve_translation_formatting
@@ -174,6 +176,28 @@ def create_app(args):
     if args.req_flood_threshold > 0:
         flood.setup(args.req_flood_threshold)
 
+    measure_request = None
+    gauge_request = None
+    if args.metrics:
+      from prometheus_client import CONTENT_TYPE_LATEST, Summary, Gauge, CollectorRegistry, multiprocess, generate_latest
+
+      @app.route("/metrics")
+      def prometheus_metrics():
+        if args.metrics_auth_token:
+          authorization = request.headers.get('Authorization')
+          if authorization != "Bearer " + args.metrics_auth_token:
+            abort(401, description="Unauthorized")
+
+        registry = CollectorRegistry()
+        multiprocess.MultiProcessCollector(registry)
+        return Response(generate_latest(registry), mimetype=CONTENT_TYPE_LATEST)
+
+      measure_request = Summary('request_seconds', 'Time spent on request', ['endpoint', 'status', 'request_ip', 'api_key'])
+      measure_request.labels('/translate', 200, '127.0.0.1', '')
+
+      gauge_request = Gauge('request_inprogress', 'Active requests', ['endpoint', 'request_ip', 'api_key'], multiprocess_mode='livesum')
+      gauge_request.labels('/translate', '127.0.0.1', '')
+
     def access_check(f):
         @wraps(f)
         def func(*a, **kw):
@@ -203,11 +227,30 @@ def create_app(args):
                         403,
                         description=description,
                     )
-
             return f(*a, **kw)
-
-        return func
-
+        
+        if args.metrics:
+          @wraps(func)
+          def measure_func(*a, **kw):
+              start_t = default_timer()
+              status = 200
+              ip = get_remote_address()
+              ak = get_req_api_key() or ''
+              g = gauge_request.labels(request.path, ip, ak)
+              try:
+                g.inc()
+                return func(*a, **kw)
+              except HTTPException as e:
+                status = e.code
+                raise e
+              finally:
+                duration = max(default_timer() - start_t, 0)
+                measure_request.labels(request.path, status, ip, ak).observe(duration)
+                g.dec()
+          return measure_func
+        else:
+          return func
+    
     @app.errorhandler(400)
     def invalid_api(e):
         return jsonify({"error": str(e.description)}), 400

app/default_values.py

@@ -160,6 +160,16 @@ _default_options_objects = [
         'name': 'UPDATE_MODELS',
         'default_value': False,
         'value_type': 'bool'
+    },    
+    {
+        'name': 'METRICS',
+        'default_value': False,
+        'value_type': 'bool'
+    },
+    {
+        'name': 'METRICS_AUTH_TOKEN',
+        'default_value': '',
+        'value_type': 'str'
     },
 ]
 

app/main.py

@@ -147,6 +147,18 @@ def get_args():
     parser.add_argument(
         "--update-models", default=DEFARGS['UPDATE_MODELS'], action="store_true", help="Update language models at startup"
     )
+    parser.add_argument(
+        "--metrics",
+        default=DEFARGS['METRICS'],
+        action="store_true",
+        help="Enable the /metrics endpoint for exporting Prometheus usage metrics",
+    )
+    parser.add_argument(
+        "--metrics-auth-token",
+        default=DEFARGS['METRICS_AUTH_TOKEN'],
+        type=str,
+        help="Protect the /metrics endpoint by allowing only clients that have a valid Authorization Bearer token (%(default)s)",
+    )
     return parser.parse_args()
 
 

gunicorn_conf.py

@@ -0,0 +1,4 @@
+from prometheus_client import multiprocess
+
+def child_exit(server, worker):
+    multiprocess.mark_process_dead(worker.pid)

requirements.txt

@@ -15,3 +15,4 @@ itsdangerous==2.1.2
 Werkzeug==2.2.2
 requests==2.28.1
 redis==4.3.4
+prometheus-client==0.15.0