`LocalSpan::enter_with_parent` missing? about minitrace-rust HOT 9 CLOSED

Because concurrency is not supported in local span, it's not possible to start a local span whose parent are not the current local parent.

I assume you're interested in tracing a single thread within an asynchronous runtime, correct? Could you describe what kind of results or data you expect to collect from this process?

from minitrace-rust.

@dotansimha, I've devised an alternative solution. Can you please check if it works for you? Here's the link: #201

from minitrace-rust.

@dotansimha, I've devised an alternative solution. Can you please check if it works for you? Here's the link: #201

Interesting, thanks @andylokandy !

So I guess I should use Span now just like before? with the new single-thread feature enabled? (instead or with enabled?)

from minitrace-rust.

You may set both enable and single-thread, and use minitrace::flush() to manually report.

from minitrace-rust.

I assume you're interested in tracing a single thread within an asynchronous runtime, correct? Could you describe what kind of results or data you expect to collect from this process?

Yeah, that's exactly what I'm building! A single-threaded (WASM, CF Worker) with async runtime (wasm-bindgen).

The main issue is that some parts of the codebase are fully agnostic to the runtime, so it's using Span and in_span for the Futures. The main issue is that these are fully ignored when LocalSpan and the LocalCollector are used. I only get SpanRecords for traces created by #[trace] or LocalSpan::enter_with_local_parent.

from minitrace-rust.

Local span can support concurrency but may not be easy... Currently, when a local span exits(drops), the parent of it will become the local parent. We need to add a logical parent to local span that are used to be the parent in SpanRecord, while keeping the real parent unchanged.

from minitrace-rust.

Local span can support concurrency but may not be easy... Currently, when a local span exits(drops), the parent of it will become the local parent. We need to add a logical parent to local span that are used to be the parent in SpanRecord, while keeping the real parent unchanged.

I see. Any insight on why #[trace] does work fine in my case? I'm putting it on async functions and it seems to work fine so far 🤔 My current workaround is to split my code to functions and add #[trace].

from minitrace-rust.

Any insight on why #[trace] does work fine in my case?

could you give a minimum reprodicible example？

from minitrace-rust.

@andylokandy see https://github.com/dotansimha/minitrace-localspan-reproduce

Basically, I'm facing several issues here (all mentioned in code comments in the repo):

1. I can't use in_span (FutureExt) with LocalSpan for async functions.
2. I can't create LocalSpan with an explicit parent (missing LocalSpan::enter_with_parent)

UPDATE 1:
Just noticed a weird behavior, where I have an async function with #[trace] and it doesn't have any traces reported 🤔 Adding enter_on_poll = true seems to partially fix it. Now I have more than one span created for each function call 🤔

UPDATE 2:
I think the main issue is the async runtime as you mentioned @andylokandy . Seems like the async function with the #[trace] is eventually using the minitrace::future::FutureExt::in_span wrapper, so it's using Span which is not being collected by the LocalCollector.

sync functions with the macro are wrapped with LocalSpan::enter_with_local_parent so I assume that's why it's working.

@andylokandy so I guess the issue is the missing async handling for LocalSpan? When I'm creating the spans manually with LocalSpan::enter_with_local_parent, it seems to work pretty well with the async functions.

from minitrace-rust.