Semantic Conventions for Exceptions on Spans

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 {
  span.end();
}

Attributes

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

The event name MUST be exception.

Attribute	Type	Description	Examples	Requirement Level
`exception.escaped`	boolean	SHOULD 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. [1]		Recommended
`exception.message`	string	The exception message.	`Division by zero`; `Can't convert 'int' object to str implicitly`	See below
`exception.stacktrace`	string	A 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(GenerateTrace.java:13)\n at com.example.GenerateTrace.methodA(GenerateTrace.java:9)\n at com.example.GenerateTrace.main(GenerateTrace.java:5)`	Recommended
`exception.type`	string	The 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 it.	`java.net.ConnectException`; `OSError`	See below

[1]: 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 above.

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.

Additional attribute requirements: At least one of the following sets of attributes is required:

exception.type
exception.message

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.

Language	Format
C#	the return value of Exception.ToString()
Elixir	the return value of Exception.format/3
Erlang	the return value of `erl_error:format`
Go	the return value of runtime.Stack
Java	the contents of Throwable.printStackTrace()
Javascript	the return value of error.stack as returned by V8
Python	the return value of traceback.format_exc()
Ruby	the 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.