Datadog opentracing doesn't support agent URL directive

[Yahavna14]

Seems like the Nginx ingress opentracing third party add-on (https://kubernetes.github.io/ingress-nginx/user-guide/third-party-addons/opentracing/) doesn't support the dd-opentracing directive of agent URL (https://github.com/DataDog/dd-opentracing-cpp/blob/master/doc/configuration.md#agent-url) and support only agent HOST directive.
I ran into it while I changed my Datadog agent to work over UDS (https://docs.datadoghq.com/developers/dogstatsd/unix_socket/?tab=host), and now my Nginx ingress controller in my k8s cluster doesn't success to send traces to the Datadog agent.

[longwuyuan]

• There is a issue where a DataDog developer and a ingress-nginx developer have discussed current state of the related code. Please search for more details. If I find I will ink here
• Opentracing.io website declares that opentracing project is archived and ingress-nginx project is removing support for opentracing (if not already removed)

[longwuyuan]

found it #10082

[longwuyuan]

/triage accepted
/priority important-longterm

[longwuyuan]

@dgoffredo any comments on this one welcome

[dgoffredo]

@Yahavna14 There are a few ways around this. I recommend the last option (OpenTelemetry).

1. Datadog OpenTracing over TCP

Use the Datadog Agent's TCP interface on port 8126 instead of Unix domain sockets. This requires that the Datadog Agent be configured to listen on port 8126. See the relevant instructions for the Datadog Operator or for Helm. In either case, you might have to change a "port" related boolean that defaults to false to true instead. Then arrange the ingress controller side configuration to be as described in relevant Datadog documentation.

This is the setup that is currently recommended for existing installations.

2. Datadog OpenTracing over UDS

Alter the ingress controller's Deployment so that its env contains a DD_TRACE_AGENT_URL of the form "unix://<full path to domain socket file>", and mount the Datadog Agent's socket from the host node into the ingress controller pods.

I've never tried this, but it's probably possible.

3. Use OpenTelemetry Instead

This is the new way to do things. The Datadog Agent can be configured to listen on an OTLP interface.

Then use the Datadog Agent's fully qualified kubernetes service name as the otlp-collector-host when configuring OpenTelemetry in the ingress controller.

[github-actions]

This is stale, but we won't close it automatically, just bare in mind the maintainers may be busy with other tasks and will reach your issue ASAP. If you have any question or request to prioritize this, please reach #ingress-nginx-dev on Kubernetes Slack.