Predicatable trace ids?

[carlpett]

Hi!
I have a use-case where I think I would like predictable trace ids, which I cannot figure out how to do? Very open to better ways of doing it, though!

We have a system which handles queued "tasks". When dequeued, these tasks can succeed, be retried, postponed or failed. Retries and postponements lead to the task being enqueued again, and tasks may not be picked up for many hours, which might also be on a separate instance of our worker pool.

What we'd like is to trace the full flow of these tasks. Naively, setting the trace-id to our internal id would be the easy way to do this. Is this possible, and if not, what is the suggested way to solve this case?

[Flarna]

Maybe using a custom IdGenerator could be used (see here).

Clearly you have to know/define the traceId at the very beginning of the trace as it is shared for all spans.

[carlpett]

Aha, interesting! Didn't notice that being pluggable.
After looking at it for a bit, though, it appears tricky to pass in the "external knowledge" to it? If I read it correctly, it is only used here:

opentelemetry-js/packages/opentelemetry-sdk-trace-base/src/Tracer.ts

Lines 89 to 100 in 3920b15

	if (
	!parentSpanContext ||
	!api.trace.isSpanContextValid(parentSpanContext)
	) {
	// New root span.
	traceId = this._idGenerator.generateTraceId();
	} else {
	// New child span.
	traceId = parentSpanContext.traceId;
	traceState = parentSpanContext.traceState;
	parentSpanId = parentSpanContext.spanId;
	}

Would it be feasible instead to fake the parentSpan? Getting really out of depth here, but perhaps a custom propagator?

[Flarna]

JS is single threaded so you could store your wanted traceId into some global variable and use it if present, otherwise fall back to a random.

parentSpan comes from context. If you set the wanted span/spanContext into a context and pass it to startSpan() you get the parent you want.

But having predictable traceId and tinkering with span hierarchy is quite a different game.

[danielo515]

I find quite annoying that I have to read the library source-code in order to be able to do something that should be relatively simple, as it is setting the traceId that I want. If any of the provided IDs (traceId or spanId) is not valid according to the library internal rules, it is silently rejected and a new one is generated, without any error or any warning. This silent behavior is very annoying and made me waste some precious time. I understand the intention is to not generate noise, but there should be some kind of communication, or provide valid strategies to generate IDs derived from other strings (we use ULID for example)