I'm not necessarily a fan of OpenTelemetry, as every time I looked at it it was still "in progress" (though the last time was almost 2 years ago), but in general I agree.
We started with tracing+logging, but as time passed we moved more and more stuff into tracing. But not just span-per-service kind of tracing - detailed orchestration on the level of "interesting" functions, with important info added as tags.
Right now we use almost exclusively tracing (and metrics ofc) and it's working great, even if each trace often has hundreds of spans.
In our case we use Datadog, which works well for most bits and purposes, but we also send all traces over Firehose to S3 so that we can query them with Athena at 100% retention. This way you get your cake and can eat it too.
If you've got the time for instrumenting something relatively small, it's definitely worth picking up OTel again to see how it works for you. Tracing is stable in most languages (Go included) and there's a lot more support in each language's ecosystem.
FWIW I think OTel will always be in a state of "in progress" though. There's more scenarios and signal types that will get picked up (profiling, RUM) and more things to spread across the ecosystem (k8s operator improvements, collector processors for all kinds of things, etc.) and more integration into relevant tools and frameworks (.NET does this today, maybe flask or spring could do it tomorrow). But I wouldn't let that stop you from trying it out again to see if you can get some value out of it!
I'm not necessarily a fan of OpenTelemetry, as every time I looked at it it was still "in progress" (though the last time was almost 2 years ago), but in general I agree.
We started with tracing+logging, but as time passed we moved more and more stuff into tracing. But not just span-per-service kind of tracing - detailed orchestration on the level of "interesting" functions, with important info added as tags.
Right now we use almost exclusively tracing (and metrics ofc) and it's working great, even if each trace often has hundreds of spans.
In our case we use Datadog, which works well for most bits and purposes, but we also send all traces over Firehose to S3 so that we can query them with Athena at 100% retention. This way you get your cake and can eat it too.