Customers should be able to debug your product in their own observability stack, not just your dashboards. Nityananda Gohain and Dhruv Ahuja make the case for native OTLP export instead of proprietary integrations that leave users maintaining polling scripts.
The first architectural decision is who runs the software:
- Self-hosted application: ship instrumentation and let users configure the endpoint; export happens in their environment.
- Managed platform: provide configurable destinations and forward workload and platform telemetry from your infrastructure.
A useful export contract includes:
- Logs, traces, and metrics with consistent attributes and preserved correlation.
- Destination URLs, transport choice, and authentication headers.
- Per-signal toggles to control ingestion volume and cost.
- A simple shared destination by default, with per-signal routing where needed.
- Documentation and working configuration examples, not just an “OpenTelemetry supported” checkbox.
The Collector topology has a real trade-off: one per tenant buys isolation at greater resource cost; shared Collectors with per-tenant pipelines simplify operations but leave you managing an expanding routing configuration.
The takeaway is portability as product design. Standardizing the transport removes bespoke integration work, but you still own the delivery contract, tenant boundaries, and semantics that make exported data useful.