cncf-toc/proposals/jaeger.adoc

170 lines
8.4 KiB
Plaintext

== Jaeger
*Name of project:* Jaeger
*Description:*
Jaeger, inspired by Dapper (https://research.google.com/pubs/pub36356.html) and OpenZipkin (http://zipkin.io/),
is a distributed tracing system that can be used for monitoring microservice-based architectures, including
distributed context propagation, distributed transaction monitoring, root cause analysis, service dependency
analysis and performance / latency optimization.
*Sponsor / Advisor from TOC:* Bryan Cantrill
*Preferred maturity level:* incubating
*Unique identifier:* jaeger
*License:* Apache 2.0 (tentative)
*Source control repositories:*
* https://github.com/uber/jaeger
* https://github.com/jaegertracing
* https://github.com/uber/jaeger-lib
* https://github.com/uber/jaeger-ui
* https://github.com/uber/jaeger-client-go
* https://github.com/uber/jaeger-client-java
* https://github.com/uber/jaeger-client-node
* https://github.com/uber/jaeger-client-python
*Initial Committers:* https://github.com/uber/jaeger/blob/master/CODEOWNERS
*Infrastructure requirements (CI / CNCF Cluster):* https://travis-ci.org/uber/jaeger
*Issue tracker:* https://github.com/uber/jaeger/issues
*Mailing list:* https://groups.google.com/forum/#!forum/jaeger-tracing
*Online chat:* https://gitter.im/jaegertracing/
*Website:* https://uber.github.io/jaeger/ (soon jaegertracing.io)
*Documentation:* http://jaeger.readthedocs.io/
*Release methodology and mechanics:*
At the moment, releases happen in an ad hoc fashion, when the maintainers and core contributors think
there are sufficient number of features or bug fixes to warrant a new release. For the main repo (Jaeger backend),
the release is tagged on GitHub and it triggers a build of docker images that are auto-uploaded to Docker Hub.
The releases for instrumentation libraries follow a similar process, and the artifacts are pushed to respective
repositories for the language, i.e. Maven Central, npm, and pypi.
*Roadmap:* http://jaeger.readthedocs.io/en/latest/roadmap/
*Existing sponsorship:* Uber has 7 full time maintainers and Red Hat has 4 full time maintainers
*Contributor statistics:* 1000+ stars and 30+ contributors https://github.com/uber/jaeger/graphs/contributors
*Adopters:* https://github.com/uber/jaeger/blob/master/ADOPTERS.md
*Community involvement:*
Some features added recently in response to community requests or as community contributions:
* Support Elasticsearch as a storage backend - https://github.com/uber/jaeger/issues/140
* Deployment templates for Kubernetes - https://github.com/jaegertracing/jaeger-kubernetes
* Jaeger as a drop-in replacement for Zipkin - https://github.com/uber/jaeger/milestone/2
* Allow Istio/Envoy to report span data to Jaeger - https://github.com/uber/jaeger/issues/225
* Add health checks for better intergation with Kubernetes - https://github.com/uber/jaeger/pull/280
* Use cpf13/cobra for flexible configuration under Kubernetes - https://github.com/uber/jaeger/issues/233
* Instrumentation libraries in more languages - https://github.com/uber/jaeger/issues/366
*Social media accounts:*
* Blog: https://medium.com/jaegertracing/
* Twitter: https://twitter.com/JaegerTracing
*External Dependencies:*
* https://github.com/uber/jaeger
** https://github.com/pkg/errors - BSD-2
** https://github.com/uber-go/zap - MIT
** https://github.com/apache/thrift - AL2
** https://github.com/opentracing/opentracing-go - MIT
** golang.org/x/net - BSD-3
** https://github.com/uber/tchannel-go - MIT
** https://github.com/gorilla/ - BSD-3
** https://github.com/kr/pretty - MIT
** https://github.com/go-kit/kit - MIT
** http://github.com/olivere/elastic - MIT
* https://github.com/jaegertracing
** Apache-2.0
* https://github.com/uber/jaeger-lib
** https://github.com/codahale/hdrhistogram - MIT
** https://github.com/go-kit/kit - MIT
** https://github.com/uber-go/tally - MIT
** https://github.com/prometheus/client_golang - AL2
* https://github.com/uber/jaeger-ui
** Many (see full list - https://gist.github.com/yurishkuro/273b5c0ed3556dfb84a095002f7dc40c)
** "(GPL-2.0 OR MIT)"
** "Apache-2.0"
** "BSD"
** "BSD*"
** "BSD-2-Clause"
** "BSD-3-Clause"
** "ISC"
** "MIT"
** "MIT*"
* https://github.com/uber/jaeger-client-go
** https://github.com/apache/thrift - AL2
** https://github.com/opentracing/opentracing-go - MIT
** golang.org/x/net - BSD-3
** https://github.com/uber/tchannel-go - MIT
* https://github.com/uber/jaeger-client-java
** https://mvnrepository.com/artifact/org.apache.thrift/libthrift - AL2
** https://mvnrepository.com/artifact/io.opentracing/opentracing-api - AL2
** https://mvnrepository.com/artifact/com.google.code.gson/gson - AL2
** https://mvnrepository.com/artifact/org.slf4j/slf4j-api - MIT
* https://github.com/uber/jaeger-client-node
** https://www.npmjs.com/package/node-int64 - MIT
** https://www.npmjs.com/package/thriftrw - MIT
** https://www.npmjs.com/package/xorshift - MIT
** https://www.npmjs.com/package/opentracing - MIT
* https://github.com/uber/jaeger-client-python
** https://pypi.python.org/pypi/futures - PSF
** https://pypi.python.org/pypi/threadloop - MIT
** https://pypi.python.org/pypi/thrift - AL2
** https://pypi.python.org/pypi/tornado - AL2
** https://pypi.python.org/pypi/opentracing - MIT
** https://pypi.python.org/pypi/future - MIT
*Statement on alignment with CNCF mission:*
A challenging part of cloud native computing is managing microservices and especially debugging the situation
when things go awry. Furthermore, distributed tracing instrumentation has been fragmented traditionally until
CNCF efforts like the OpenTracing specification came to existence to help unify existing tracing implementations
out there.
Jaeger is a battle tested distributing system that takes advantage of OpenTracing and advances the state
of open source distributed tracing.
*Comparison with Zipkin*
Zipkin is another popular open source distributed tracing system originally released by Twitter in 2012
(https://blog.twitter.com/engineering/en_us/a/2012/distributed-systems-tracing-with-zipkin.html).
Jaeger and Zipkin provide roughly similar functionality today; in fact the earlier versions of Jaeger relied on some of Zipkin
components as explained in the Uber Eng blog post (https://eng.uber.com/distributed-tracing/). Listed below are some differences
between Jaeger and Zipkin, along with how they can interoperate:
* *OpenTracing Instrumentation Libraries*. All Jaeger instrumentation libraries are built to support the OpenTracing standard.
Zipkin ecosystem has a couple of OpenTracing compatible libraries, but most of them do not support it, instead requiring the
applications to use bespoke APIs and Zipkin-only semantic annotations.
* *OpenTracing Compatible Backend and UI*. Jaeger is built with OpenTracing standard in mind from the ground up, including the
backend, the data models, and the UI. Zipkin backend does not support all of the features of OpenTracing, specifically
structured k-v span logs and multi-parent spans / DAGs (which are possible in OpenTracing via span references).
* *Instrumentation Libraries Maintenance*. Jaeger instrumentation libraries are officially part of the project and undergo
continuous integration testing against each other and against the backend. Zipkin instrumentation libraries are not officially
part of the project, often not even located in the openzipkin Github org, and rarely integration-tested against each other.
On the other hand, Zipkin libraries exist in more languages. Today Jaeger supports Go, Java, Python, and Javascript,
with Ruby and C++ implementation on the way.
* *Distributed Context Propagation*. General purpose context propagation is extremely useful in microservices based systems.
In addition to tracing it can support many other applications, including resource attribution / chargebacks, capacity planning,
chaos engineering, security. All Jaeger libraries support context propagation via OpenTracing feature called “baggage”.
Most Zipkin libraries do not support propagation of custom context metadata.
* *Storage Backends*. Zipkin supports more storage backends since it has been around a lot longer. Jaeger currently supports
Cassandra and Elasticsearch as production quality storage backends.
* *Interoperability*. Jaeger backend can be used as a replacement for Zipkin backend in organizations that already invested
in Zipkin instrumentation. Jaeger instrumentation libraries can also be configured to interoperate with Zipkin libraries
via Zipkin wire format (B3 headers), that allows new services to be instrumented with OpenTracing.