Merge remote-tracking branch 'origin/dev' into permissions-system-update

.gitignore

@@ -61,3 +61,6 @@ web/src/__generated__
 
 # pulumi config files
 Pulumi*.yaml
+
+# performance profile files
+profiles

api/server.py

@@ -1,3 +1,4 @@
+import datetime
 import os
 import time
 import traceback
@@ -12,7 +13,12 @@
 
 from api import routes
 from api.graphql.schema import MetamistGraphQLRouter  # type: ignore
-from api.settings import PROFILE_REQUESTS, SKIP_DATABASE_CONNECTION, SM_ENVIRONMENT
+from api.settings import (
+    PROFILE_REQUESTS,
+    PROFILE_REQUESTS_OUTPUT,
+    SKIP_DATABASE_CONNECTION,
+    SM_ENVIRONMENT,
+)
 from api.utils.exceptions import determine_code_from_error
 from api.utils.openapi import get_openapi_schema_func
 from db.python.connect import SMConnections
@@ -49,10 +55,42 @@ async def app_lifespan(_: FastAPI):
 
 
 if PROFILE_REQUESTS:
+    from pyinstrument import Profiler
+    from pyinstrument.renderers.speedscope import SpeedscopeRenderer
 
-    from fastapi_profiler.profiler import PyInstrumentProfilerMiddleware
+    @app.middleware('http')
+    async def profile_request(request: Request, call_next):
+        """optional profiling for http requests"""
+        profiler = Profiler(async_mode='enabled')
+        profiler.start()
+        resp = await call_next(request)
+        profiler.stop()
+
+        if 'text' in PROFILE_REQUESTS_OUTPUT:
+            text_output = profiler.output_text()
+            print(text_output)
+
+        timestamp = (
+            datetime.datetime.now().replace(microsecond=0).isoformat().replace(':', '-')
+        )
+
+        if 'json' in PROFILE_REQUESTS_OUTPUT:
+            os.makedirs('profiles', exist_ok=True)
+            json = profiler.output(renderer=SpeedscopeRenderer())
+
+            with open(f'profiles/{timestamp}.json', 'w') as file:
+                file.write(json)
+                file.close()
+
+        if 'html' in PROFILE_REQUESTS_OUTPUT:
+            os.makedirs('profiles', exist_ok=True)
+            html = profiler.output_html()
+            with open(f'profiles/{timestamp}.html', 'w') as file:
+                file.write(html)
+                file.close()
+
+        return resp
 
-    app.add_middleware(PyInstrumentProfilerMiddleware)  # type: ignore
 
 if SM_ENVIRONMENT == 'local':
     app.add_middleware(
@@ -185,6 +223,5 @@ async def exception_handler(request: Request, e: Exception):
         'api.server:app',
         host='0.0.0.0',
         port=int(os.getenv('PORT', '8000')),
-        # debug=True,
         reload=True,
     )

api/settings.py

@@ -13,6 +13,7 @@
 SM_ENVIRONMENT = os.getenv('SM_ENVIRONMENT', 'production').lower()
 SKIP_DATABASE_CONNECTION = bool(os.getenv('SM_SKIP_DATABASE_CONNECTION'))
 PROFILE_REQUESTS = os.getenv('SM_PROFILE_REQUESTS', 'false').lower() in TRUTH_SET
+PROFILE_REQUESTS_OUTPUT = os.getenv('SM_PROFILE_REQUESTS_OUTPUT', 'text').lower()
 IGNORE_GCP_CREDENTIALS_ERROR = os.getenv('SM_IGNORE_GCP_CREDENTIALS_ERROR') in TRUTH_SET
 MEMBERS_CACHE_LOCATION = os.getenv('SM_MEMBERS_CACHE_LOCATION')
 

docs/installation.md

@@ -427,3 +427,12 @@ The deploy github action builds the container, and is deployed.
 Additionally you can access metamist through the identity-aware proxy (IAP),
 which handles the authentication through OAuth, allowing you to access the
 front-end.
+
+
+## Performance Profiling
+
+If you are working on performance issues it can be handy to generate a report that shows which bits of code are taking most of the time. The api server has pyinstrument profiling support that can be turned on by setting  the environment variable `SM_PROFILE_REQUESTS` to `true`.
+
+There are a few different options for outputting profiles which can be specified in the `SM_PROFILE_REQUESTS_OUTPUT` environment variable. The possible values are `text` which will print the profiling results to stdout, `html` which will generate an interactive pyinstrument report, or `json` which will generate a json profiling report which can be dropped into [speedscope](https://www.speedscope.app/) to explore the profile.
+
+You can output multiple report types by specifying the types in a list like: `export SM_PROFILE_REQUESTS_OUTPUT=json,text,html`