kOps is experimenting with initial support for OpenTelemetry, starting with tracing.
The support should be considered experimental; the trace file format and schema will likely change, and these initial experiments might be removed entirely.
kOps supports a "serverless" mode of operation, where it logs the OpenTracing output to a file. We do this because our e2e test runner (prow) doesn't yet have a destination for OpenTelemetry data.
To try this out:
OTEL_EXPORTER_OTLP_TRACES_FILE=/tmp/trace go run ./cmd/kops get cluster
You should now see that the /tmp/trace file is created.
Then we have an experimental tool to serve the trace file to jaeger. First ensure jaeger-query
is in your PATH, available here
cd tools/otel/traceserver
go run . --src /tmp/trace --run jaeger
Alternatively you can use the jaeger docker image with:
go run . --src /tmp/trace --run docker-jaeger
This will open your local jaeger-query's frontend in a web browser:
Not everything is instrumented yet, and not all the traces are fully joined up (we need to thread more contexts through more methods), but you should be able to start to explore the operations that we run and their performance.
Tracing is enabled in the kops prow jobs. .otel
files are created during job execution and included as job artifacts in the otlp
subdirectory.
Example:
https://gcsweb.k8s.io/gcs/kubernetes-jenkins/pr-logs/pull/kops/16204/presubmit-kops-aws-scale-amazonvpc-using-cl2/1742962895290896384/artifacts/otlp/
Download these files to a local directory, either with gsutil
or through a web browser
mkdir /tmp/job-traces
gsutil cp -r gs://kubernetes-jenkins/pr-logs/pull/kops/16204/presubmit-kops-aws-scale-amazonvpc-using-cl2/1742962895290896384/artifacts/otlp/ /tmp/job-traces
Then run the trace server as normal:
cd tools/otel/traceserver
go run . --src /tmp/traces --run jaeger