add monitoring with grafana and prometheus

config.yaml

@@ -8,6 +8,7 @@ service:
   batch_size: $BATCH_SIZE|2  # Number of images in memory simultaneously
   verbose: $VERBOSE|True  # Service prints document processing progress to stdout
   run_id: $RUN_ID|fabfb1f192c745369b88cab34471aba7  # The ID of the mlflow run to load the model from
+  monitoring_enabled: $MONITORING_ENABLED|True # if app is doing monitoring or not
 
 
 # These variables control filters that are applied to either images, image metadata or model predictions. The filter

requirements.txt

@@ -19,3 +19,5 @@ PDFNetPython3~=9.1.0
 Pillow~=8.3.2
 PyYAML~=5.4.1
 scikit_learn~=0.24.2
+prometheus-client==0.13.1
+prometheus_flask_exporter==0.19.0

src/serve.py

@@ -1,5 +1,6 @@
 import logging
 import tempfile
+import tracemalloc
 
 from flask import Flask, request, jsonify
 from waitress import serve
@@ -7,14 +8,21 @@ from waitress import serve
 from image_prediction.config import CONFIG
 from image_prediction.predictor import Predictor, extract_image_metadata_pairs, classify_images
 from image_prediction.response import build_response
+from prometheus_client import Gauge, Counter
+from prometheus_flask_exporter import PrometheusMetrics
 
 
 def main():
 
     predictor = Predictor()
     logging.info("Predictor ready.")
+    tracemalloc.start()
 
     app = Flask(__name__)
+    metrics = PrometheusMetrics(app=app, path='/prometheus')
+
+    file_counter = Counter("image_prediction_file_counter", "count processed files")
+    ram_metric = Gauge("image_prediction_memory_usage", "Memory usage in Mb")
 
     @app.route("/ready", methods=["GET"])
     def ready():
@@ -29,7 +37,13 @@ def main():
         return resp
 
     @app.route("/", methods=["POST"])
+    @metrics.summary('image_prediction_request_time_seconds', 'Time spent processing request')
     def predict():
+        def do_monitoring():
+            file_counter.inc()
+            _, peak = tracemalloc.get_traced_memory()
+            ram_metric.set(peak / 10 ** 6)
+
         pdf = request.data
 
         logging.debug("Running predictor on document...")
@@ -38,6 +52,8 @@ def main():
             image_metadata_pairs = extract_image_metadata_pairs(tmp_file.name)
             try:
                 predictions, metadata = classify_images(predictor, image_metadata_pairs)
+                if CONFIG.service.monitoring_enabled:
+                    do_monitoring()
             except Exception as err:
                 logging.warning("Analysis failed.")
                 logging.exception(err)
@@ -59,6 +75,7 @@ def run_prediction_server(app, mode="development"):
         app.run(host=CONFIG.webserver.host, port=CONFIG.webserver.port, debug=True)
     elif mode == "production":
         serve(app, host=CONFIG.webserver.host, port=CONFIG.webserver.port)
+    tracemalloc.stop()
 
 
 if __name__ == "__main__":