Time Matters: Choosing Between Delta and Cumulative Metric Temporalities for Effective Monitoring
summary
OpenTelemetry Collector now includes a delta-to-cumulative processor (and a cumulative->delta pair) that works but is stateful and requires sticky routing, so converters need all measurements for a series to hit the same collector. Delta uses far less memory and suits short‑lived jobs but silently overcounts when samples drop and omits idle zeros; cumulative tolerates network loss but needs more state. Grafana Labs ships initial Datadog receiver support that emits delta today.
AI-written summary · about
chapters (8)
- 0:00Temporality in metrics
- 1:08Delta vs cumulative
- 2:13Memory and state
- 3:39Sample loss handling
- 5:07Periodicity differences
- 6:11Converting temporalities
- 8:40Choosing temporalities
- 10:06Graal Labs Delta support
related talks
23mDoes It Add Up? Exploring the Delta Temporality in OpenTelemetry & Beyond
28mOTEL delta temporality support in Prometheus
19mTelemetry Showdown: Fluent Bit vs. OpenTelemetry Collector - A Comprehensive Benchma...
27mTricked Out Traces
25mSponsored Session: Unifying Observability at DoorDash using Open Stand...
26mScaling OpenTelemetry for Modern Workloads: From Mobile To Ma...
35mBuilding a Real Time Metrics Database
23mControlling Data Overhead with the OpenTelemetry Collector