Distributed systems are complex and have many moving parts, much of which are asynchronous and run in parallel. When building complex systems, it is best to consider design in small chunks that are composable. Instrumenting complex systems is no different. Lightbend Telemetry breaks capture down into composable parts that will provide better insight into your system.
Lightbend Telemetry provides insight into applications built with Lightbend technologies. It does so by instrumenting frameworks and toolkits such as Akka. The instrumentation is done by a Java agent that runs when your application is starting up. Lightbend Telemetry (a.k.a. Cinnamon) collects information, in runtime, about your application based on a configuration that you must provide. As you can see below, Cinnamon is running in the same JVM as your application.
Based on configuration, Telemetry will send data to a backend of your choice. Lightbend Console provides a set of built in monitors and dashboards for Telemetry-instrumented applications. Telemetry also provides integrations with Elasticsearch, StatsD, Datadog, JMX, etc. It is also possible to provide a custom integration should the backend of your choice not be available.
If you run a cluster or multiple nodes in general, Cinnamon will run on each node. Each individual node will report to the backend you have configured:
By using configuration, you can instruct how Cinnamon should report the information it is collecting. Out of the box, Cinnamon provides several plugins.
Below is an example of what this may look like for integration with Elasticsearch. In this example, we also use Kibana and Grafana to retrieve and display the information that gets published into Elasticsearch. This also happens to be the setup of the Cinnamon developer sandbox environment: an easy way to bootstrap and try Cinnamon out.
Instrumentations are the enablers of our stack that hook into the underlying toolkit or framework for our telemetry solution. Currently, we support
Lagom with the following feature sets:
- Akka: captures telemetry (metrics, events, or traces) for Akka Actors, Akka Remoting, Akka Cluster, and Akka Persistence.
- Akka Streams: captures telemetry (metrics, events, or traces) for Akka Streams.
- Akka HTTP: captures server, endpoint, and client telemetry (metrics or traces) for Akka HTTP applications.
- Play: captures server, endpoint, and client telemetry (metrics or traces) for Play applications.
- Lagom: captures server, endpoint, and client telemetry (metrics or traces), as well as circuit breaker metrics, for Lagom services.
- Scala Futures: captures telemetry (metrics or traces) for explicitly named Futures.
- Java Futures: captures telemetry (metrics or traces) for explicitly named CompletableFutures.
Instruments are the nitty gritty of our stack. Keeping composable design in mind, we classify our instruments into one of three categories: metrics, events, or traces. Our metrics represent a unit of measure within a time constraint, whereas our events embody historical behavior.
- Metrics include counters, gauges, and rates.
- Events include errors, unhandled messages, and dead letters.
- Traces follow asynchronous or distributed message flows.
Asynchronous boundaries are one of the primary challenges behind instrumenting distributed systems. It is difficult to reason about behavior when stuff does not happen in the order we think it should. To manage this, Lightbend Telemetry provides context propagation in the form of
Mapped Diagnostic Context (MDC), and the
Stopwatch extension. You can think of them as buckets designed to capture data of a particular type or path regardless of when or where it occurs.
- SLF4J MDC
- Custom Events
- Custom Metrics
- JMX Importer
- JVM Metrics producer
Our telemetry solution is designed to support pluggable backends for
trace data. Lightbend Telemetry provides the following backend plugins:
- Prometheus (metrics and event rates)
- Datadog (metrics and event rates)
- Coda Hale Metrics (metrics and event rates)
- StatsD (metrics and event rates)
- Elasticsearch (metrics and events)
- SLF4J events (events)
- Jaeger (traces)
- Zipkin (traces)
- Datadog (traces)
It is possible to use multiple backends simultaneously.
At the end of the day, we have to reason about the data we capture, and as they say, a picture is worth a thousand words. In this vein, we provide plugins for the following visualization suites:
Lightbend Telemetry provides a developer sandbox environment that you can use to quickly get started. Unless you already have your monitoring infrastructure set up, using the developer sandbox is the fastest way to test your application with Lightbend Telemetry. The developer sandbox comes prepackaged with Elasticsearch, Kibana and Grafana all configured to be used in together. The developer sandbox is only for testing purposes and is not intended for production.