Extra AWS X-Ray instrumentation to use distributed tracing with asyncio Python libraries that are not (yet) supported by the official aws_xray_sdk library.
aws_xray_sdk is the standard library to collect trace data from your Python
code and send the trace data to the
AWS X-Ray distributed tracing tool. However,
if you have asyncio Python code, then there are some gaps and occasional
bugs in the functionality provided by that library. xraysink plugs those gaps.
It can be a bit confusing using two libraries together, so here's a high-level breakdown of which library will help you do what:
- Add tracing to HTTP requests handled by FastAPI (or another async Python
web framework):
xraysink(via middleware) - Add tracing to background (non-HTTP-request) functions written as async
Python functions:
xraysink(via xray_task_async decorator) - Everything else: aws_xray_sdk
- Generic ASGI-compatible tracing middleware for any ASGI-compliant web framework. This has been tested with:
- asyncio Task's
- Background jobs/tasks
xraysink is distributed as a standard python package through pypi, so you can install it with your favourite Python package manager. For example:
pip install xraysink
xraysink augments the functionality provided by aws_xray_sdk. Before
using the tools in xraysink, you first need to configure aws_xray_sdk -
this will probably involve calling xray_recorder.configure() when your
process starts, and optionally aws_xray_sdk.core.patch().
Extra instrumentation provided by xraysink is described below.
Instrument incoming requests in your FastAPI web server by adding the
xray_middleware to your app. For example, you could do:
from starlette.middleware.base import BaseHTTPMiddleware
from xraysink.asgi.middleware import xray_middleware
# Standard asyncio X-Ray configuration, customise as you choose
xray_recorder.configure(context=AsyncContext(), service="my-cute-little-service")
# Create a FastAPI app with various middleware
app = FastAPI()
app.add_middleware(MyTracingDependentMiddleware) # Any middleware that is added earlier will have the X-Ray tracing context available to it
app.add_middleware(BaseHTTPMiddleware, dispatch=xray_middleware)
If you start asyncio Task's from a standard request handler, then the AWS X-Ray SDK will not correctly instrument any outgoing requests made inside those Tasks.
Use the fixed AsyncContext from xraysink as a drop-in replacement, like so:
from aws_xray_sdk.core import xray_recorder
from xraysink.context import AsyncContext # NB: Use the AsyncContext from xraysink
# Use the fixed AsyncContext when configuring X-Ray,
# and customise other configuration as you choose.
xray_recorder.configure(context=AsyncContext(use_task_factory=True))
If your process starts background tasks that make network calls (eg. to the
database or an API in another service), then each execution of one of those
tasks should be treated as a new X-Ray trace. Indeed, if you don't do so then
you will likely get context_missing errors.
An async function that implements a background task can be easily instrumented
using the @xray_task_async() decorator, like so:
from aws_xray_sdk.core import xray_recorder
from xraysink.tasks import xray_task_async
# Standard asyncio X-Ray configuration, customise as you choose
xray_recorder.configure(context=AsyncContext(), service="my-cute-little-service")
# Any call to this function will start a new X-Ray trace
@xray_task_async()
async def cleanup_stale_tokens():
await database.get_table("tokens").delete(age__gt=1)
# Start your background task using your scheduling system of choice :)
schedule_recurring_task(cleanup_stale_tokens)
If your background task functions are called from a function that is already
instrumented (eg. send an email immediately after handling a request), then
the background task will appear as a child segment of that trace. In this case,
you must ensure you use the non-buggy AsyncContext when configuring the recorder
(ie. from xraysink.context import AsyncContext)
You can link your X-Ray traces to your CloudWatch Logs log records, which enhances the integration with AWS CloudWatch ServiceLens. Take the following steps:
-
Put the X-Ray trace ID into every log message. There is no convention for how to do this (it just has to appear verbatim in the log message somewhere), but if you are using structured logging then the convention is to use a field called
traceId. Here's an exampletrace_id = xray_recorder.get_trace_entity().trace_id logging.getLogger("example").info("Hello World!", extra={"traceId": trace_id}) -
Explicitly set the name of the CloudWatch Logs log group associated with your process. There is no general way to detect the Log Group from inside the process, hence it requires manual configuration as part of your process initialisation (eg. in the same place where you call
xray_recorder.configure).set_xray_log_group("/example/service-name")
Note that this feature relies on undocumented functionality, and is not yet supported by the official Python SDK.
This project uses the Apache 2.0 licence, to make it compatible with aws_xray_sdk, the primary library for integrating with AWS X-Ray.
![github-actions[bot] avatar](https://avatars.githubusercontent.com/in/15368?v=4 "github-actions[bot]")
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent