The highlight.io SDKs are powered by OpenTelemetry under the hood, and therefore report data to our deployed opentelemetry collector. For a better understanding of the architecture, take a look at the architecture page for a diagram of how data is sent to the collector and the public graph.

In our SDKs, we instantiate the following constructs to exports data over OTLP HTTPS to https://otel.highlight.io:4318/v1/traces and https://otel.highlight.io:4318/v1/logs respectively.

The SDK provides common methods for recording exceptions or logging, but this may depend on the language. For example, in Golang, a logger hook API is provided to be configured by the application, but in Python, we automatically ingest a hook into the built in logging package.

Recording an Error

Data we send over the OpenTelemetry specification is as a Trace with attributes set per the semantic conventions. When we create a Trace, we set three additional SpanAttributes to carry the Highlight context:

Reporting an Error as an OTEL Trace

An exception is represented in OpenTelemetry as a Trace Event, per the semantic convention for exceptions.

Many OpenTelemetry SDK implementations offer a span.record_exception(exc) method that automatically populates the semantic convention attributes with the correct values.

# create a trace for the current invocation
with self.tracer.start_as_current_span("highlight-ctx") as span:
    span.set_attributes({"highlight.project_id": _project_id})
    span.set_attributes({"highlight.session_id": session_id})
    span.set_attributes({"highlight.trace_id": request_id})
    try:
        # contextmanager yields execution to the code using the contextmanager
        yield
    except Exception as e:
        # if an exception is raised, record it on the current span
        span.record_exception(e)
        raise

Reporting a Log as an OTEL Trace

If a language's OpenTelemetry SDK does not support sending logs natively, we choose to send the message data as a Trace Event.

To associate the highlight context with a log, we use the LogRecord Attributes with the following convention:

package main

import "github.com/highlight/highlight/sdk/highlight-go"

func RecordLog(log string) {
	span, _ := highlight.StartTrace(context.TODO(), "highlight-go/logrus")
	defer highlight.EndTrace(span)

	attrs := []attribute.KeyValue{
		LogSeverityKey.String("ERROR"),
		LogMessageKey.String(entry.Message),
	}
	span.AddEvent(highlight.LogEvent, trace.WithAttributes(attrs...))
}

Recording a Log

If an SDK supports the experimental logs ingest endpoint (v1/logs), prefer using that. Otherwise, see above for reporting the log as a trace event.

A LogRecord is exported with an associated trace. Specific attributes for the file logging, line number, and more are set based on the logging semantic convention keys.

Here's an example of the interception of python logging calls in our Python SDK to emit an OTEL LogRecord.

attributes = span.attributes.copy()
attributes["code.function"] = record.funcName
attributes["code.namespace"] = record.module
attributes["code.filepath"] = record.pathname
attributes["code.lineno"] = record.lineno
r = LogRecord(
    timestamp=int(record.created * 1000.0 * 1000.0 * 1000.0),
    trace_id=ctx.trace_id,
    span_id=ctx.span_id,
    trace_flags=ctx.trace_flags,
    severity_text=record.levelname,
    severity_number=std_to_otel(record.levelno),
    body=record.getMessage(),
    resource=span.resource,
    attributes=attributes,
)