Understand that delta temporality reports changes since the last export while cumulative reports the running total since process start
Configure the exporter's preferred temporality setting; Prometheus prefers cumulative while many OTLP backends prefer delta
Use the temporality selector API in the SDK to assign temporality per instrument kind rather than globally if mixed backends are needed
Test temporality behavior by restarting the instrumented process and observing whether counters reset to zero (delta) or continue accumulating (cumulative)
Document the chosen temporality for each pipeline so consumers know whether to use rate functions or raw values in queries
Known gotchas
Switching a metric from cumulative to delta on a live pipeline causes a gap or spike in dashboards; coordinate the change with a dashboard update
Some backends silently convert temporality on ingest; verify end-to-end that the values stored match your expectations rather than assuming pass-through
Delta counters that miss an export interval produce a gap in the series; cumulative counters that restart produce a counter reset which rate functions handle but naive queries do not
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp