Import prometheus client to collect metrics

Signed-off-by: Kaiyuan Hu <kaiyuan.hu@zilliz.com>

main.py

@@ -1,13 +1,14 @@
-import argparse
 import os
+import argparse
+import time
 
 import uvicorn
 from fastapi import FastAPI, UploadFile
 from fastapi.encoders import jsonable_encoder
+from prometheus_client import start_http_server, Counter, Histogram, Gauge
 
 from config import TEMP_DIR
 
-
 os.makedirs(TEMP_DIR, exist_ok=True)
 
 # Specify mode
@@ -31,17 +32,52 @@
 app = FastAPI()
 origins = ['*']
 
+# Define metrcis
+latencies = []
+answer_latencies = []
+add_latencies = []
+drop_latencies = []
+
+requests_total = Counter('requests_total', 'Cumulative requests')
+requests_success_total = Counter('requests_success_total', 'Cumulative successful requests')
+requests_failed_total = Counter('requests_failed_total', 'Cumulative failed requests')
+
+endpoint_requests_total = Counter('endpoint_requests_total', 'Cumulative requests of each endpoint', ['endpoint'])
+endpoint_requests_success_total = Counter('endpoint_requests_success_total', 'Cumulative successful requests of each endpoint', ['endpoint'])
+endpoint_requests_failed_total = Counter('endpoint_requests_failed_total', 'Cumulative failed requests of each endpoint', ['endpoint'])
+
+process_latency_seconds = Histogram('process_latency_seconds', 'Request process latency histogram')
+endpoint_process_latency_seconds = Histogram('endpoint_process_latency_seconds', 'Request process latency histogram of each endpoint', ['endpoint'])
+
+average_process_latency_seconds = Gauge('average_process_latency_seconds', 'Average request process latency')
+endpoint_average_process_latency_seconds = Gauge(
+    'endpoint_average_process_latency_seconds', 'Average Request process latency of each endpoint', ['endpoint']
+)
+
+
 @app.get('/')
 def check_api():
-    return jsonable_encoder({'status': True, 'msg': 'ok'}), 200
+    requests_total.inc()
+    endpoint_requests_total.labels('/').inc()
+
+    res = jsonable_encoder({'status': True, 'msg': 'ok'}), 200
+
+    requests_success_total.inc()
+    endpoint_requests_success_total.labels('/').inc()
+
+    return res
 
 @app.get('/answer')
 def do_answer_api(session_id: str, project: str, question: str):
     try:
+        begin = time.time()
+        requests_total.inc()
+        endpoint_requests_total.labels('/answer').inc()
+
         new_question, final_answer = chat(session_id=session_id,
                             project=project, question=question)
         assert isinstance(final_answer, str)
-        return jsonable_encoder({
+        res = jsonable_encoder({
             'status': True,
             'msg': final_answer,
             'debug': {
@@ -50,12 +86,31 @@ def do_answer_api(session_id: str, project: str, question: str):
                 'answer': final_answer,
             }
             }), 200
+
+        requests_success_total.inc()
+        endpoint_requests_success_total.labels('/answer').inc()
+        end = time.time()
+        latencies.append(end - begin)
+        answer_latencies.append(end - begin)
+        process_latency_seconds.observe(end - begin)
+        endpoint_process_latency_seconds.labels('/answer').observe(end - begin)
+        average_process_latency_seconds.set(sum(latencies) / len(latencies))
+        endpoint_average_process_latency_seconds.label('answer').set(sum(answer_latencies) / len(answer_latencies))
+
+        return res
     except Exception as e:  # pylint: disable=W0703
+        requests_failed_total.inc()
+        endpoint_requests_failed_total.labels('/answer').inc()
+
         return jsonable_encoder({'status': False, 'msg': f'Failed to answer question:\n{e}', 'code': 400}), 400
 
 
 @app.post('/project/add')
 def do_project_add_api(project: str, url: str = None, file: UploadFile = None):
+    begin = time.time()
+    requests_total.inc()
+    endpoint_requests_total.labels('/project/add').inc()
+
     assert url or file, 'You need to upload file or enter url of document to add data.'
     try:
         if url:
@@ -66,20 +121,56 @@ def do_project_add_api(project: str, url: str = None, file: UploadFile = None):
                 content = file.file.read()
                 f.write(content)
             num = insert(data_src=temp_file, project=project, source_type='file')
-        return jsonable_encoder({'status': True, 'msg': f'Successfully inserted doc chunks: {num}'}), 200
+        res = jsonable_encoder({'status': True, 'msg': f'Successfully inserted doc chunks: {num}'}), 200
+
+        requests_success_total.inc()
+        endpoint_requests_success_total.labels('/project/add').inc()
+        end = time.time()
+        latencies.append(end - begin)
+        add_latencies.append(end - begin)
+        process_latency_seconds.observe(end - begin)
+        endpoint_process_latency_seconds.labels('/project/add').observe(end - begin)
+        average_process_latency_seconds.set(sum(latencies) / len(latencies))
+        endpoint_average_process_latency_seconds.labels('/project/add').set(sum(add_latencies) / len(add_latencies))
+
+        return res
     except Exception as e:  # pylint: disable=W0703
+        requests_failed_total.inc()
+        endpoint_requests_failed_total.labels('/project/add').inc()
+
         return jsonable_encoder({'status': False, 'msg': f'Failed to load data:\n{e}'}), 400
 
 
 @app.post('/project/drop')
 def do_project_drop_api(project: str):
     # Drop data in vector db
     try:
+        begin = time.time()
+        requests_total.inc()
+        endpoint_requests_total.labels('/project/add').inc()
+
         drop(project=project)
-        return jsonable_encoder({'status': True, 'msg': f'Dropped project: {project}'}), 200
+        res = jsonable_encoder({'status': True, 'msg': f'Dropped project: {project}'}), 200
+
+        requests_success_total.inc()
+        endpoint_requests_success_total.labels('/project/drop').inc()
+        end = time.time()
+        end = time.time()
+        latencies.append(end - begin)
+        drop_latencies.append(end - begin)
+        process_latency_seconds.observe(end - begin)
+        endpoint_process_latency_seconds.labels('/project/drop').observe(end - begin)
+        average_process_latency_seconds.set(sum(latencies) / len(latencies))
+        endpoint_average_process_latency_seconds.labels('/project/drop').set(sum(drop_latencies) / len(drop_latencies))
+
+        return res
     except Exception as e:  # pylint: disable=W0703
+        requests_failed_total.inc()
+        endpoint_requests_failed_total.labels('/project/drop').inc()
+
         return jsonable_encoder({'status': False, 'msg': f'Failed to drop project:\n{e}'}), 400
 
 
 if __name__ == '__main__':
+    start_http_server(8901)
     uvicorn.run(app=app, host='0.0.0.0', port=8900)

requirements.txt

@@ -11,3 +11,4 @@ uvicorn
 towhee>=1.1.0
 pymilvus
 elasticsearch>=8.0.0
+prometheus-client