The idea here is to set the parent span for incoming federation requests to the
*outgoing* span on the other end. That means that you can see (most of) the
full end-to-end flow when you have a process that includes federation requests.
However, in order not to lose information, we still want a link to the
`incoming-federation-request` span from the servlet, so we have to create
another span to do exactly that.
`start_active_span` was inconsistent as to whether it would activate the span
immediately, or wait for `scope.__enter__` to happen (it depended on whether
the current logcontext already had an associated scope). The inconsistency was
rather confusing if you were hoping to set up a couple of separate spans before
activating either.
Looking at the other implementations of opentracing `ScopeManager`s, the
intention is that it *should* be activated immediately, as the name
implies. Indeed, the idea is that you don't have to use the scope as a
contextmanager at all - you can just call `.close` on the result. Hence, our
cleanup has to happen in `.close` rather than `.__exit__`.
So, the main change here is to ensure that `start_active_span` does activate
the span, and that `scope.close()` does close the scope.
We also add some tests, which requires a `tracer` param so that we don't have
to rely on the global variable in unit tests.
I've never found this terribly useful. I think it was added in the early days
of Synapse, without much thought as to what would actually be useful to log,
and has just been cargo-culted ever since.
Rather, it tends to clutter up debug logs with useless information.
* remove `start_active_span_from_request`
Instead, pull out a separate function, `span_context_from_request`, to extract
the parent span, which we can then pass into `start_active_span` as
normal. This seems to be clearer all round.
* Remove redundant tags from `incoming-federation-request`
These are all wrapped up inside a parent span generated in AsyncResource, so
there's no point duplicating all the tags that are set there.
* Leave request spans open until the request completes
It may take some time for the response to be encoded into JSON, and that JSON
to be streamed back to the client, and really we want that inside the top-level
span, so let's hand responsibility for closure to the SynapseRequest.
* opentracing logs for HTTP request events
* changelog
* Teach MyPy that the sentinel context is False
This means that if `ctx: LoggingContextOrSentinel`
then `bool(ctx)` narrows us to `ctx:LoggingContext`, which is a really
neat find!
* Annotate RequestMetrics
- Raise errors for sentry if we use the sentinel context
- Ensure we don't raise an error and carry on, but not recording stats
- Include stack trace in the error case to lower Sean's blood pressure
* Make mypy pass for synapse.http.request_metrics
* Make synapse.http.connectproxyclient pass mypy
Co-authored-by: reivilibre <oliverw@matrix.org>
The following scenarios would halt the user directory updater:
- user joins room
- user leaves room
- user present in room which switches from private to public, or vice versa.
for two classes of users:
- appservice senders
- users missing from the user table.
If this happened, the user directory would be stuck, unable to make forward progress.
Exclude both cases from the user directory, so that we ignore them.
Co-authored-by: Eric Eastwood <erice@element.io>
Co-authored-by: reivilibre <oliverw@matrix.org>
Co-authored-by: Sean Quah <8349537+squahtx@users.noreply.github.com>
Co-authored-by: Brendan Abolivier <babolivier@matrix.org>
This commit fixes two bugs to do with decorators not instrumenting
`ReplicationEndpoint`'s `send_request` correctly. There are two
decorators on `send_request`: Prometheus' `Gauge.track_inprogress()`
and Synapse's `opentracing.trace`.
`Gauge.track_inprogress()` does not have any support for async
functions when used as a decorator. Since async functions behave like
regular functions that return coroutines, only the creation of the
coroutine was covered by the metric and none of the actual body of
`send_request`.
`Gauge.track_inprogress()` returns a regular, non-async function
wrapping `send_request`, which is the source of the next bug.
The `opentracing.trace` decorator would normally handle async functions
correctly, but since the wrapped `send_request` is a non-async function,
the decorator ends up suffering from the same issue as
`Gauge.track_inprogress()`: the opentracing span only measures the
creation of the coroutine and none of the actual function body.
Using `Gauge.track_inprogress()` as a context manager instead of a
decorator resolves both bugs.
Updating mypy past version 0.9 means that third-party stubs are no-longer distributed with typeshed. See http://mypy-lang.blogspot.com/2021/06/mypy-0900-released.html for details.
We therefore pull in stub packages in setup.py
Additionally, some modules that we were previously ignoring import failures for now have stubs. So let's use them.
The rest of this change consists of fixups to make the newer mypy + stubs pass CI.
Co-authored-by: Patrick Cloke <clokep@users.noreply.github.com>
* switch from `types.CoroutineType` to `typing.Coroutine`
these should be identical semantically, and since `defer.ensureDeferred` is
defined to take a `typing.Coroutine`, will keep mypy happy
* Fix some annotations on inlineCallbacks functions
* changelog
* Trace event persistence
When we persist a batch of events, set the parent opentracing span to the that
from the request, so that we can trace all the way in.
* changelog
* When we force tracing, set a baggage item
... so that we can check again later.
* Link in both directions between persist_events spans
This adds quite a lot of OpenTracing decoration for database activity. Specifically it adds tracing at four different levels:
* emit a span for each "interaction" - ie, the top level database function that we tend to call "transaction", but isn't really, because it can end up as multiple transactions.
* emit a span while we hold a database connection open
* emit a span for each database transaction - actual actual transaction.
* emit a span for each database query.
I'm aware this might be quite a lot of overhead, but even just running it on a local Synapse it looks really interesting, and I hope the overhead can be offset just by turning down the sampling frequency and finding other ways of tracing requests of interest (eg, the `force_tracing_for_users` setting).
As far as I can tell our logging contexts are meant to log the request ID, or sometimes the request ID followed by a suffix (this is generally stored in the name field of LoggingContext). There's also code to log the name@memory location, but I'm not sure this is ever used.
This simplifies the code paths to require every logging context to have a name and use that in logging. For sub-contexts (created via nested_logging_contexts, defer_to_threadpool, Measure) we use the current context's str (which becomes their name or the string "sentinel") and then potentially modify that (e.g. add a suffix).
Part of #9744
Removes all redundant `# -*- coding: utf-8 -*-` lines from files, as python 3 automatically reads source code as utf-8 now.
`Signed-off-by: Jonathan de Jong <jonathan@automatia.nl>`
Running `dmypy run` will do a `mypy` check while spinning up a daemon
that makes rerunning `dmypy run` a lot faster.
`dmypy` doesn't support `follow_imports = silent` and has
`local_partial_types` enabled, so this PR enables those options and
fixes the issues that were newly raised. Note that `local_partial_types`
will be enabled by default in upcoming mypy releases.
Patrick Cloke