Semantic Conventions for Exceptions on Spans

Status: Stable

This document defines semantic conventions for recording application exceptions associated with spans.

Recording an Exception

An exception SHOULD be recorded as an Event on the span during which it occurred. The name of the event MUST be "exception".

A typical template for an auto-instrumentation implementing this semantic convention using an API-provided recordException method could look like this (pseudo-Java):

Span span = myTracer.startSpan(/*...*/);
try {
  // Code that does the actual work which the Span represents
} catch (Throwable e) {
  span.recordException(e, Attributes.of("exception.escaped", true));
  throw e;
} finally {


The table below indicates which attributes should be added to the Event and their types.

The event name MUST be exception.

AttributeTypeDescriptionExamplesRequirement LevelStability
exception.messagestringThe exception message.Division by zero; Can't convert 'int' object to str implicitlyConditionally Required [1]Stable
exception.typestringThe type of the exception (its fully-qualified class name, if applicable). The dynamic type of the exception should be preferred over the static type in languages that support; OSErrorConditionally Required [2]Stable
exception.escapedbooleanSHOULD be set to true if the exception event is recorded at a point where it is known that the exception is escaping the scope of the span. [3]RecommendedStable
exception.stacktracestringA stacktrace as a string in the natural representation for the language runtime. The representation is to be determined and documented by each language SIG.Exception in thread "main" java.lang.RuntimeException: Test exception\n at com.example.GenerateTrace.methodB(\n at com.example.GenerateTrace.methodA(\n at com.example.GenerateTrace.main(

[1]: Required if exception.type is not set, recommended otherwise.

[2]: Required if exception.message is not set, recommended otherwise.

[3]: An exception is considered to have escaped (or left) the scope of a span, if that span is ended while the exception is still logically “in flight”. This may be actually “in flight” in some languages (e.g. if the exception is passed to a Context manager’s __exit__ method in Python) but will usually be caught at the point of recording the exception in most languages.

It is usually not possible to determine at the point where an exception is thrown whether it will escape the scope of a span. However, it is trivial to know that an exception will escape, if one checks for an active exception just before ending the span, as done in the example for recording span exceptions.

It follows that an exception may still escape the scope of the span even if the exception.escaped attribute was not set or set to false, since the event might have been recorded at a time where it was not clear whether the exception will escape.

Stacktrace Representation

The table below, adapted from Google Cloud, includes possible representations of stacktraces in various languages. The table is not meant to be a recommendation for any particular language, although SIGs are free to adopt them if they see fit.

C#the return value of Exception.ToString()
Elixirthe return value of Exception.format/3
Erlangthe return value of erl_error:format
Gothe return value of runtime.Stack
Javathe contents of Throwable.printStackTrace()
Javascriptthe return value of error.stack as returned by V8
Pythonthe return value of traceback.format_exc()
Rubythe return value of Exception.full_message

Backends can use the language specified methodology for generating a stacktrace combined with platform information from the telemetry sdk resource in order to extract more fine grained information from a stacktrace, if necessary.