This crate bridges between gstreamer and tracing ecosystems.

Last update: Jun 7, 2022

Related tags

Overview

This crate provides a bridge between gstreamer and the tracing ecosystem.

The goal is to allow Rust applications utilizing GStreamer to better integrate into application that otherwise use the tracing crate for their observability needs.

Examples

Events

To output gstreamer log messages as tracing events, call the [integrate_events] function. Calling it before the call to any other gstreamer call (especially before the gstreamer::init) is most likely to correctly forward all of the messages:

// Set up the tracing subscriber.
//
// e.g. tracing_subscriber::fmt::init();

tracing_gstreamer::integrate_events();
gstreamer::debug_remove_default_log_function();
gstreamer::init();

Keep in mind that both GST_DEBUG and tracing filters are in effect. The gstreamer side of filters can be relaxed from code via:

gstreamer::debug_set_default_threshold(gstreamer::DebugLevel::Count);

Similarly you can use tracing APIs to adjust the filters on the tracing side.

Spans

To provide tracing with more contextual information for some of the events, you can also enable support for generating spans via gstreamer's own tracing infrastructure.

This functionality can be enabled by calling the [integrate_spans] function. It must be called after gstreamer::init.

gstreamer::init();
tracing_gstreamer::integrate_spans();

Subscriber showcase

This section demonstrates the results obtained with different kinds of subscribers.

`tracing_subscriber::fmt`

This subscriber is a great replacement for the built-in gstreamer log handler. Here's an example of what the output might look like when using this subscriber:

$ env RUST_LOG=info cargo run --example videoenc
    Finished dev [unoptimized + debuginfo] target(s) in 0.04s
     Running `target/debug/examples/videoenc`
Jan 01 00:00:00.000  INFO gstreamer::GST_INIT: Initializing GStreamer Core Library version 1.18.4
<snip>
Jan 01 00:00:00.000  INFO gstreamer::GST_INIT: initialized GStreamer successfully
Jan 01 00:00:00.000  INFO gstreamer::GST_PIPELINE: parsing pipeline description '
        videotestsrc num-buffers=120
        ! vp9enc
        ! webmmux name=mux
        ! fakesink sync=false

        audiotestsrc num-buffers=120
        ! opusenc
        ! mux.
    '

Certain messages may also provide more information than the built-in logger. While builtin logger does present you with the type and address of the object being logged, tracing-gstreamer may provide more readily useful information such as the element name:

Jan 01 00:00:00.000  INFO gstreamer::GST_STATES: completed state change to READY gobject.address=94331150660528 gobject.type="GstAudioTestSrc" gstobject.name="audiotestsrc0" gstelement.state="ready" gstelement.pending_state="void-pending"

or provide additional context via spans, which may help to figure out which element is logging the message when there is no other way to tell this otherwise, such as in this example:

Jan 01 00:00:00.000  INFO pad_push{gstpad.state={NEED_PARENT} gstpad.parent.name="audiotestsrc0"}: gstreamer::structure: Expected field 'channel-mask' in structure: audio/x-raw, rate=(int)48000, channels=(int)1, format=(string)S16LE, layout=(string)interleaved;

`tracing-tracy`

Tracy is a profiler focused primarily on game development workloads, but works fairly well for general purpose code as well. Tracy features a sampling profiler, but works best with applications that have manually instrumented points of interest. tracing is a great source of such manual instrumentation points and tracing-tracy is the bridge between the two. The following video demonstrates the videoenc example from this repository adapted to utilize the tracing-tracy subscriber.

gstreamer-in-tracy.mp4

In this video there are a couple of highlights

We can quickly see the amount of concurrency our pipeline enables (2 threads; perhaps adding some queues would help?)
Overall thread utilization (low for the audio portion and high for the video portion);
Investigate the performance of the specific elements and quickly find out why some of them are slow. For example the opusenc0 element sometimes takes an unusually long time because the downstream muxer already has a buffer queued at the time.

Similar results can be achieved with some other subscribers as well.

A patch to fix tracing LocalTime problem.

tracing-local-time A patch to fix tracing LocalTime problem. Tracing-subscriber now has a bug in LocalTime, so build ourselves' to fix it. In this pat

2 Dec 27, 2021

A dynamic binary tracing tool

Backlight Backlight is a dynamic binary tracing tool. Install $ git clone [email protected]:JoshMcguigan/backlight.git $ cd backlight $ cargo install-b

42 Dec 3, 2022

tracing - a framework for instrumenting Rust programs to collect structured, event-based diagnostic information

tracing-appender Writers for logging events and spans Documentation | Chat Overview tracing is a framework for instrumenting Rust programs to collect

1 Mar 9, 2022

A rust `tracing` compatible framework inspired by log4rs.

trace4rs This crate allows users to configure output from tracing in the same way as you would configure the output of log4rs. Overview For a usage ex

5 Oct 24, 2022

AWS Cloudwatch layer for tracing-subscriber

tracing-cloudwatch tracing-cloudwatch is a custom tracing-subscriber layer that sends your application's tracing events(logs) to AWS CloudWatch Logs.

7 May 14, 2023

Utilities for integrating Datadog with opentelemetry + tracing in rust

Non-official datadog tracing and log correlation for Rust services. This crate contains the necessary glue to bridge the gap between OpenTelemetry, tr

5 Oct 31, 2023

A lean, minimal, and stable set of types for color interoperation between crates in Rust.

This library provides a lean, minimal, and stable set of types for color interoperation between crates in Rust. Its goal is to serve the same function that mint provides for (linear algebra) math types.

16 Sep 21, 2022

Easy switch between AWS Profiles and Regions

AWSP - CLI To Manage your AWS Profiles! AWSP provides an interactive terminal to interact with your AWS Profiles. The aim of this project is to make i

14 Dec 25, 2022

Simple interoperability between C++ coroutines and asynchronous Rust

cxx-async Overview cxx-async is a Rust crate that extends the cxx library to provide seamless interoperability between asynchronous Rust code using as

180 Dec 16, 2022

Comments

non-exhaustive patterns/ does not compile

Сompile time error :

➜  new-rtsp-service git:(main) ✗ cargo run
   Compiling tracing-gstreamer v0.3.0
   Compiling gstreamer-rtsp-server v0.18.0
error[E0004]: non-exhaustive patterns: `(&Kind(0_u8), _)` and `(&Kind(3_u8..=u8::MAX), _)` not covered
   --> /home/***/.cargo/registry/src/github.com-1ecc6299db9ec823/tracing-gstreamer-0.3.0/src/callsite.rs:152:38
    |
152 |             .then_with(move || match (&self.kind, &other.kind) {
    |                                      ^^^^^^^^^^^^^^^^^^^^^^^^^ patterns `(&Kind(0_u8), _)` and `(&Kind(3_u8..=u8::MAX), _)` not covered
    |
    = help: ensure that all possible cases are being handled, possibly by adding wildcards or more match arms
    = note: the matched value is of type `(&Kind, &Kind)`

Cargo.toml:

[dependencies]
tracing = "0.1"
tracing-subscriber = "0.3"
tracing-gstreamer = "0.3"
log-panics = "2"


gstreamer-rtsp-server = "0.18"

url = "2"

tokio = {version = "1", features = ["rt-multi-thread", "macros", "sync"]}

opened by RustPanda 5

Address breaking changes in tracing_core

tracing-core 0.1.22 introduced an unfortunate breaking change. One of the public types was intended to be non-exhaustive, but turns out rustc was smart enough to make it exhaustive anyway.

The hope was that there wouldn't be any way to obtain the Kind from a Metadata and even less of a reason to actually match on one. Alas, it turned out to be not true for us, where we were trying to open code an implementation of PartialEq for Kind.

In retrospective we should've been using our own defined type anyway and deriving PartialEq as that's just cleaner (we don't need to think about any future variants added to Kind.)

Fixes #6

opened by nagisa 0
Implement handling of event interests

This was pretty fun a headscratcher. What happens here is that any time a new Callsite is created (conceptually a location in code which contains a log statement) we must register ourselves with tracing runtime and obtain subscribers' interest in the event, and filter accordingly.

This avoids unnecessarily spending timme on invoking all of the subscriber machineries among other things. As a future optimisation on our side, we can also check the interest before allocating/leaking the Metadata and Callsite, at the cost of checking the interest twice when we do actually register the callsite.

opened by nagisa 0
More configurable span tracing categories

Today this library registers a single tracer which in turn unconditionally registers all the hooks it is interested at once:

https://github.com/standard-ai/tracing-gstreamer/blob/9b7d80b43b794499a40f5252634427ad362494e8/src/tracer.rs#L108-L117

If we wanted to extend the scope of the tracing this library does (which I would very much like to,) we sort-of need to figure out a way to make the trace interest more configurable. One option would be to actually have different Tracer objects for each different trace categories. We could alternatively expose the Tracer as a public API and handle the registration of different events outside of the object creation, giving users granular control to do the same for themselves.

If nothing else, this gives a great way for people to mitigate locally issues that might occur as a result of bug in bindings or this library in tracers for a specific event.

opened by nagisa 0