I'm helping a client with a large hybrid environment spanning AWS, Azure, and on-premises systems. They've used Datadog APM for several years, but legacy technical debt has left the setup inconsistent: APM libraries are outdated, source instrumentation is misconfigured, and distributed traces no longer connect reliably across important services. Datadog support advised correcting the application code and local agents, but coordinating changes across all the application teams will take time. A centralized observability pipeline is also being developed, although it won't be ready for several months. In the meantime, could an intermediate proxy, OpenTelemetry Collector, or other processing layer intercept telemetry and transform, enrich, or repair spans before they reach Datadog? Has anyone managed to reduce legacy APM gaps at the pipeline level without redeploying the applications?
1 Answer
There are useful things you can do in a collector or gateway, such as filtering spans, renaming attributes, standardizing tags, redacting data, sampling consistently, and adding metadata from the infrastructure layer. However, a pipeline cannot reliably recreate information that the application never emitted. If parent or child relationships, propagation headers, service names, or instrumentation spans are missing, the collector usually doesn’t have enough context to rebuild the original distributed trace. The durable fix is still updating the libraries, instrumentation, and local agents. A collector can be a good temporary mitigation, but I’d avoid treating it as a replacement for source-level fixes.
That makes sense. The client already uses tools such as Copilot and Claude, but getting the individual application teams to prioritize repository changes is the main obstacle. We may use the pipeline for normalization and visibility while planning the application updates in parallel.