Does sampler influence the completeness of a trace?

[lmssy]

I set trace sampler rate as 0.1, I store all trace information in ElasticSeach.

I got a trace of my server which spanId is nveqeCbIYa0=, parentSpanId is qzJh8poTULY=, but I can't find any trace whose spanId is qzJh8poTULY=, it's not reasonable.

Is this because I set the sampler rate not 100%? I think a trace must be complete no matter what sampler rate is.

[anuraaga]

Hi @lmssy - as long as you are using a parentbased_traceidratio sampler, it will be complete. Is this how you configured it? If so, there is probably some other reason that the span for qzJh8poTULY couldn't be exported, perhaps the exporter isn't configured correctly for that server?

[lmssy]

hi @anuraaga - Thx for your reply, I check my config is traceidratio, I will change to parentbased_traceidratio and try again. Thanks!

[lmssy]

Hi @anuraaga , problem still here, but seems all those situation happen when my server call redis or mysql, is this something related to connection pool or thread pool? I mean the thread of the request to mysql or redis has been changed to another thread of mysql or redis' connection pool, then the trace chain is broken?

[anuraaga]

Oh can you confirm, is the trace breaking between two remote servers or is it spans within the same server? Also are you using auto instrumentation Java agent?

[lmssy]

Spans within the same server, I use auto instrumentation JavaAgent

[anuraaga]

In that case, we'd expect the agent to make sure context is propagated such that the trace is complete. Actually, we wouldn't expect a parent span ID to be set to a non-existent span either way. Can you try using the logging exporter to see if the spans are exported correctly when logging? If so, it could be an issue between your collector and Elasticsearch but not the app.