In February we introduced Jaeger in a blog post Evolving Distributed Tracing at Uber Engineering. Two months later Jaeger backend has been released as an open source project. So many things have happened since then. We built a great community of users who deploy Jaeger in their organizations and contribute to the project (special kudos to RedHat’s Hawkular team). We added many new features. We joined the Cloud Native Computing Foundation as its 12th project. And now we are proud to announce the official 1.0 release of Jaeger backend.
Even though Jaeger has been integrated with thousands of applications at Uber and running in production for almost two years, we started the open source journey with modest 0.x versions. We knew we had some work ahead of us before we could make Jaeger easily deployable in the cloud environments, flexible in its infrastructure dependencies, backwards compatible with existing solutions, and integrated with other CNCF projects. We think we reached a milestone where we are ready for a 1.0 version.
Storage backends
The first release of Jaeger only supported Cassandra as a storage backend for traces. We chose Cassandra mostly because our team had operational experience with it, but it was obvious that not every team felt that way. Over the summer we implemented another backend store based on ElasticSearch. ES is a lot more popular as a hosted service, which makes deploying a high throughput Jaeger installation even easier. Later on our community made additional improvements to these backends, including support for various authentication methods for securing db connections.
We were also considering support for MySQL backend, but could not prioritize it above other development. Interestingly, we’ve seen our community experimenting with other backends, such as ScyllaDB and InfluxDB.
Improvements in Jaeger UI
Jaeger 1.0 brings a significant number of improvements in its Web UI. Aside from performance optimizations to make it more efficient, we refactored the trace view using a virtual viewport technique that allows it to smoothly handle large traces containing as many as 50,000 spans. Other UI improvements include better navigation through large traces using the zoom-in feature of the minimap and keyboard shortcuts. The top level menu of the UI can now be customized via configuration, which is very useful when deploying Jaeger inside a company and providing links to internal resources such as help pages or opening a ticket for the tracing team.
Prometheus, Docker, and Kubernetes
Jaeger backend itself is a moderately complex distributed system that needs proper production monitoring. In the 1.0 release we made Prometheus the default metrics system integration in all Jaeger backend components. Deploying Jaeger in the cloud manually, “the hard way”, is no fun, even though we provide ready to run Docker images. Fortunately, fellow contributors at RedHat have created a GitHub project jaeger-kubernetes that contains templates for running Jaeger on Kubernetes. Then the Helm community picked up the torch and built a Helm chart that makes the whole process as simple as helm install incubator/jaeger –name myjaeger.
Instrumentation libraries
Even though Jaeger instrumentation libraries are versioned separately from the backend, it is worth mentioning that they are an active area of development, and the 1.0 milestone coincides with the release of the early version of C++ client. Having a C++ client in the toolbelt allows not only an easier path for other scripting languages, but also integration with high performance load balancers like nginx and haproxy and service meshes like Envoy.
Our community is also working on client libraries in other languages, including Objective-C, Ruby, PHP. We are looking forward to bringing these libraries into the Jaeger project as officially supported.
Backwards compatibility with Zipkin
Another active area of development across the Jaeger ecosystem was backwards compatibility with Zipkin. Jaeger backend was enhanced to be a drop-in replacement for Zipkin backend by accepting several Zipkin span formats (Thrift and JSON). Jaeger client libraries include configuration options to make them use Zipkin in-band wire format for trace context propagation.
Why? For many years Zipkin was the only game in town if you wanted to use an open source tracing system. Many organizations have already invested in instrumenting their applications with Zipkin APIs, instead of, say, vendor-neutral OpenTracing APIs. Jaeger’s interop features allow those organizations to switch to Jaeger backend with minimum cost, and to continue using Zipkin in-band wire format while instrumenting new applications with OpenTracing and Jaeger.
Roadmap
Even though we are proud of the functionality we are releasing in v1.0, we are even more excited about the next generation features we are currently working on. Jaeger is a great tool if you want to look at individual traces and investigate performance issues, but individual traces are a tiny portion of the of the overall knowledge that can be gained from tracing data. For example, Jaeger installations at Uber are ingesting over 10 billion spans a day. Thus our top priority are the features that support aggregations, analytics, and data mining, tools on top of Jaeger platform that allow gathering insights about the whole architecture at large. Some of those features are described on the Roadmap page.
We are also looking at tighter integrations with other CNCF projects like Envoy, Linkerd, and Istio, emerging standards like the Trace-Context HTTP header, and alternative instrumentation APIs like OpenCensus.
Try it out!
You can try out Jaeger 1.0 by following our Getting Started guide. Join us on Twitter (@JaegerTracing), Gitter chat room, or the mailing list.
Finally, we would like to thank all our users who extensively tested the pre-releases and helped us in debugging issues. This huge milestone would not have been possible without you!
Connect at KubeCon + CloudNativeCon
If you are attending KubeCon + CloudNativeCon North America this week, please join us December 8th for a Jaeger Salon to discuss the project and basic tracing concepts, as well as more advanced topics like adaptive sampling, dependency graphs and tracing with Envoy proxy.
Additionally, do not miss the December 6th session by Jaeger founder Yuri Shkuro, “Would You Like Some Tracing With Your Monitoring?” and the December 7th sessions on SIG Jaeger Update and SIG Jaeger Deep Dive Session.