Should ending a span force its children to end by default?

[open-telemetry]

Should ending a span force its children to end by default?

[tedsuo]

Sounds like there is agreement that we should not close the child spans (I agree as well).

Also, I feel this is a spec (cross-language) issue, and it should work the same in all languages.

[mayurkale22]

This looks like a legacy implementation and I am not sure about the logic. I think we should not close child spans forcefully (by default), and like the idea of passing an option to force child spans to end.

Few questions floated in my mind,

• Does it cause memory leakage if anyone failed to end the span, due to bad code?
• How does it get visualized in Zipkin, Jaeger etc exporters? Does it create separate orphan span?

[SergeyKanzhelev]

it is a good point that cross languages spec needs to document this decision. As an experiment, I've added spec-needed label and corresponding issue: open-telemetry/opentelemetry-specification#18

[rochdev]

There are tons of valid use cases for a child span to finish after its parent, and in some cases even start after its parent is finished. This is especially true in Node where everything happens asynchronously.

What is the use case for the configuration option? When would you want to auto finish the children?

[hekike]

+1 on not closing child spans automatically. For the same reason, I find it hard to get span ending casualty right around closing an operation.

[mayurkale22]

Closed via open-telemetry/opentelemetry-specification#18