Merge branch 'add_metrics_counter' into 'main'

Prepare utils for metrics counters

See merge request oss/imalive!11

CONTRIBUTING.md

@@ -13,6 +13,7 @@ docker-compose -f docker-compose-local.yml up --build --abort-on-container-exit
 ```
 
 Then you can try:
+* the api endpoint: http://localhost:8080
 * jaegger UI here for the traces: http://localhost:16686
 * the opentelemetry metrics exporter endpoint: http://localhost:8889/metrics
 * prometheus: http://localhost:9090

VERSION

@@ -1 +1 @@
-3.6.1
+3.6.2

requirements.txt

@@ -2,6 +2,7 @@ fastapi[all]
 requests
 uvicorn[standard]
 fastapi-utils
+asgi-correlation-id
 psutil
 prometheus-fastapi-instrumentator
 opentelemetry-api

src/exception/ImaliveHTTPException.py

@@ -0,0 +1,4 @@
+class ImaliveHTTPException(Exception):
+    def __init__(self, status_code: int, message: dict):
+        self.status_code = status_code
+        self.message = message

src/main.py

@@ -1,10 +1,16 @@
-from fastapi import FastAPI
+from uuid import uuid4
+from fastapi import FastAPI, Request, HTTPException
+from fastapi.responses import JSONResponse
+from fastapi.exception_handlers import http_exception_handler
+from asgi_correlation_id import CorrelationIdMiddleware, correlation_id
 from prometheus_fastapi_instrumentator import Instrumentator
 from opentelemetry.instrumentation.fastapi import FastAPIInstrumentor
 
 from restful_ressources import import_ressources
+from exception.ImaliveHTTPException import ImaliveHTTPException
 
 from utils.common import is_not_empty
+from utils.cid import get_current_cid
 from utils.manifests import get_manifest_as_dict
 from utils.heartbit import heartbit
 from utils.otel import init_otel_tracer, init_otel_metrics
@@ -22,6 +28,12 @@
     description="Official Imalive API Swagger documentation"
 )
 
+app.add_middleware(
+    CorrelationIdMiddleware,
+    header_name = 'x-imalive-cid',
+    generator = lambda: "{}".format(uuid4())
+)
+
 instrumentator = Instrumentator()
 
 init_otel_tracer()
@@ -35,4 +47,12 @@
 
 FastAPIInstrumentor.instrument_app(app)
 
+@app.exception_handler(Exception)
+async def unhandled_exception_handler(request: Request, exc: Exception) -> JSONResponse:
+    headers = {'x-imalive-cid': get_current_cid()}
+
+    if isinstance(exc, ImaliveHTTPException):
+        return JSONResponse(content = exc.message, status_code = exc.status_code, headers = headers)
+    return await http_exception_handler(request, HTTPException(500, 'Internal server error', headers = headers))
+
 import_ressources(app)

src/routes/api_health.py

@@ -1,16 +1,21 @@
 from fastapi import APIRouter
 
+from utils.cid import get_current_cid
+from utils.counter import create_counter, increment_counter
 from utils.otel import get_otel_tracer
 from utils.health import health
 
 router = APIRouter()
+_counter = create_counter("health_api_counter", "Health API counter")
 
 @router.get("")
 def get_health():
     with get_otel_tracer().start_as_current_span("imalive-health-get-route"):
+        increment_counter(_counter, get_current_cid())
         return health()
 
 @router.post("")
 def post_health():
     with get_otel_tracer().start_as_current_span("imalive-health-post-route"):
+        increment_counter(_counter, get_current_cid())
         return health()

src/routes/api_manifest.py

@@ -1,16 +1,19 @@
 from fastapi import APIRouter
 from fastapi.responses import JSONResponse
 
+from utils.cid import get_current_cid
+from utils.counter import create_counter, increment_counter
 from utils.otel import get_otel_tracer
 from utils.manifests import get_manifest_as_dict
 
 router = APIRouter()
+_counter = create_counter("manifest_api_counter", "Health API counter")
 
 @router.get("")
 def get_manifest():
     with get_otel_tracer().start_as_current_span("imalive-manifest-route"):
         manifest = get_manifest_as_dict()
-
+        increment_counter(_counter, get_current_cid())
         if manifest['status'] == 'error':
             return JSONResponse(content=manifest, status_code=500)
         else:

src/routes/api_metrics.py

@@ -1,11 +1,15 @@
 from fastapi import APIRouter
 
+from utils.cid import get_current_cid
+from utils.counter import create_counter, increment_counter
 from utils.otel import get_otel_tracer
 from utils.metrics import all_metrics
 
 router = APIRouter()
+_counter = create_counter("metrics_api_counter", "Health API counter")
 
 @router.get("")
 def get_metrics():
     with get_otel_tracer().start_as_current_span("imalive-metrics-route"):
+        increment_counter(_counter, get_current_cid())
         return all_metrics()

src/routes/api_root.py

@@ -2,10 +2,14 @@
 
 from utils.otel import get_otel_tracer
 from utils.health import health
+from utils.counter import create_counter, increment_counter
+from utils.cid import get_current_cid
 
 router = APIRouter()
+_counter = create_counter("root_api_counter", "Root API counter")
 
 @router.get("/")
 def get_root():
     with get_otel_tracer().start_as_current_span("imalive-root-route"):
+        increment_counter(_counter, get_current_cid())
         return health()

src/utils/cid.py

@@ -0,0 +1,13 @@
+from uuid import uuid4
+from asgi_correlation_id import correlation_id
+
+from utils.common import is_empty
+
+def get_current_cid():
+    try:
+        cid = correlation_id.get()
+        if is_empty(cid):
+            cid = "{}".format(uuid4())
+        return cid
+    except Exception:
+        return "{}".format(uuid4())

src/utils/counter.py

@@ -0,0 +1,11 @@
+from utils.otel import get_otel_meter
+
+def create_counter(name, description):
+    return get_otel_meter().create_counter(
+        name = name,
+        description = description,
+        unit = "1"
+    )
+
+def increment_counter(counter, tid):
+    counter.add(1, {"tid": tid})

src/utils/heartbit.py

@@ -4,7 +4,6 @@
 
 from time import sleep
 
-
 from utils.common import is_enabled
 from utils.otel import get_otel_tracer
 from utils.gauge import create_gauge, set_gauge

src/utils/logger.py

@@ -5,6 +5,7 @@
 import sys
 
 from datetime import datetime
+from utils.cid import get_current_cid
 
 from utils.common import is_disabled, is_enabled, is_true
 
@@ -93,15 +94,17 @@ def get_int_value_level(level):
 
 def quiet_log_msg (log_level, message):
     vdate = datetime.now().isoformat()
-    formatted_log = "[{}][{}][{}] {}".format(log_level, vdate, NODE_NAME, message)
+    cid = get_current_cid()
+    formatted_log = "[{}][{}][{}][{}] {}".format(log_level, vdate, NODE_NAME, cid, message)
     if is_enabled(LOG_FORMAT) and LOG_FORMAT == "json":
         if isinstance(message, dict):
             message['level'] = log_level
             message['time'] = vdate
             message['node'] = NODE_NAME
+            message['cid'] = cid
             formatted_log = json.dumps(message)
         else:
-            formatted_log = json.dumps({"body": message, "level": log_level, "time": vdate, "node": NODE_NAME})
+            formatted_log = json.dumps({"body": message, "level": log_level, "time": vdate, "node": NODE_NAME, "cid": cid})
 
     if is_debug(log_level):
         logging.debug(formatted_log)