Hi everyone,
Following up on the public feature request
<https://github.com/apache/beam/issues/33176> and discussions I've decided
to try to implement open telemetry tracing into Beam SDK, starting with
Java SDK and two IOs - Kafka and PubSub.

I've been working on this and the results can be reviewed in this PR
<https://github.com/apache/beam/pull/34544>.

Where it is helpful - user can track latency for sampled messages across
multiple services e.g:
pubsub publisher --> dataflow with bigtable enrichment --> kafkaIO -->
kafka consumer

Support in this PR is limited to:
- Trace context propagation from/to KafkaIO and from PubSubIO and in the
future to PubSubIO as well as long as both IOs are running in the same
stage.
- Cloud Bigtable's client and other libraries with opencensus tracing are
automatically supported
- Default support for Cloud Trace when Dataflow is used

What is not working:
- Trace context is not propagated yet beyond the stage/thread in which it
was initiated
- Other SDKs are not yet supported

Looking for your feedback, let me know if you like it or you find this
implementation limiting your use cases.

Radek

Reply via email to