libdd-data-pipeline

Data pipeline for processing and exporting APM data to Datadog.

Overview

libdd-data-pipeline provides a high-performance pipeline for processing distributed tracing data, applying normalization and obfuscation, computing statistics, and exporting to Datadog backends.

Status

Currently the project is a proof of concept so the API is not definitive and possible frequent changes should be expected.

Modules

• TraceExporter: Main trace export functionality. Provides a minimum viable product (MVP) to send traces to agents. The aim of the project at this state is to provide a basic API in order to test its viability and integration in different languages.

Requirements

The current implementation assumes the following requisites must be met by the tracer:

• The protocol used is v0.4.
• All initialization must come from the tracer. The module won't try to infer any configuration.
• The trace must be serialized in msgpack before passing it to the libdd-data-pipeline module.
• Sending process is synchronous.
• The agent's response will be handled by the tracer.

Dataflow

  +--------------+                 +--------------+                  +--------------+
  |              |  msgpack(v0.4)  |              |   msgpack(v0.x)  |              |
  |              +---------------->|              +----------------->|              |
  |    Tracer    |                 |   Exporter   |                  |    Agent     |
  |              |                 |              |                  |              |
  |              |    Response     |              |   Response       |              |
  |              |<----------------+              |<-----------------+              |
  +--------------+                 +--------------+                  +--------------+

API

Currently the libdd-data-pipeline crate exports just one module TraceExporter which follows a builder pattern to configure the communication.

See trace_exporter.rs.

Integrating the TraceExporter in the tracers

[WIP]Importing a binary into an existing project

In case you want to use a binary to hook C-like functions in your language we will provide a crate to build such binary. This crate is located in: libdatadog
|
+--libdd-data-pipeline-ffi

Build

cargo build --debug/release

Artifacts

The build will produce two artifacts:

• libdata-pipeline-ffi.so
• libdata-pipeline.h They will be located in libdatadog/target/\[debug|release\]/.

Building the bindings directly in libdatadog

In case of using a Rust framework in order to build the bindings for you language there is the possibility to create a new crate in libdatadog workspace and use the libdd-data-pipeline crate as a dependency.

Create new crate

cargo new libdd-data-pipeline-nodejs --lib

Set up dependencies

In order to use the TraceExporter in your project the libdd-data-pipeline crate needs to be added in the bindings dependency list.

[package]
# Package attributes

[lib]
crate-type = ["cdylib"]

[dependencies]
# Language bindings framework dependencies.

libdd-data-pipeline = { path = ../libdd-data-pipeline }

[build-dependencies]
# Build framework dependencies

Build and artifact generation

The building and artifact generation process will depend heavily on the framework selected.

Expected workflow from teams integrating libdd-data-pipeline

• Create your own language bindings.
• Hook it in the tracer.
• Feedback about difficulties about integrating the solution, performance and package size.

Future work

• Asynchronous interface.
• Handle transformations between different protocol versions.
• Agent API discovery.