Hey everyone! I've been trying to follow the OpenTelemetry world while evaluating our observability stack. We're a small team, and the promise of vendor-agnostic instrumentation is super appealing to avoid lock-in.
I saw the announcement about Elastic's new native OTLP support and the updated APM agent integrations. On paper, it looks like it could simplify our pipeline—sending directly to Elastic without the OpenTelemetry Collector as a mandatory hop. But I'm a bit lost in the details.
Could anyone who's tried this new integration share some real-world impressions? I'm especially curious about:
* How does the cost/performance compare to using the Collector? If Elastic is now a "first-class" OTLP destination, does it handle high cardinality spans or metrics more efficiently?
* Did you run into any gaps compared to using the traditional Elastic APM agents? The demo I watched made setup look easy, but I'm worried about missing features or alerting quirks.
* For someone currently on a mix of tools (think: a SaaS APM, a DIY Prometheus stack), does this make Elastic a more compelling "all-in-one" option? Or is it mostly a convenience layer?
I've got demo fatigue at this point, so hands-on experiences would be really valuable. The idea of reducing moving parts is tempting, but I don't want to swap one complexity for another.
New here!
Just my two cents.
Nice to see someone else wading through the OTel hype. I've been testing this integration for a few weeks on a staging environment.
On your cost/performance point, sending OTLP directly felt about the same on ingestion as using the Collector for us, but the real win is one less component to manage and monitor. High cardinality still needs careful indexing management in Elastic itself, but it's no worse.
The main gap I found is around custom instrumentation. The traditional APM agents gave us a bit more fine-grained control for some legacy services. The OTel integration works great out of the box, but if you've heavily customized your spans, double-check the attribute mapping.
For your all-in-one question, it definitely makes Elastic more compelling. Having traces and metrics in the same place where you can correlate them without hopping tools is huge. Just make sure your team is ready for the Elasticsearch learning curve if they aren't already. It's powerful, but it's not a simple SaaS dashboard.
one stack at a time