OpenTelemetry at Delivery Hero: The Good, the Bad and the Vendor-Agnostic
Talkat Observability Day 2025 · on Tuesday, 2025-04-01· at 13:30· Level 1 | Hall Entrance S10 | Room C·
summary
Delivery Hero aims to cut $6M/year and escape vendor lock-in by consolidating 13 acquired entities onto OpenTelemetry, but ends up running a zoo of collectors: central DataDog and OTLP collectors, federated collectors, and per-entity collectors. Real work includes forked contrib components (DataDog receiver, span-matrix connector, load‑balancing exporter), a custom proxy plus header‑based routing for stateful processors, and a delta-to-cumulative processor; marshalling costs dominate CPU and memory and require ongoing custom ops.
AI-written summary · about
chapters (11)
- 0:00OpenTelemetry intro
- 1:46Migration strategy
- 3:34Complex infrastructure history
- 6:04Observability goals
- 7:52Migration trade-offs
- 10:25OTel component challenges
- 13:19Collector inefficiencies
- 16:07Stateful components overhead
- 18:54Pains and learnings
- 19:35Adopting open-source standards
- 20:18Q&A
related talks
18mOpenTelemetry and Observability: What, Why, and Why Now?
13mManaging Observability Data at the Edge with the OpenTelemetry Collector and OTTL
05mOpenTelemetry in production for a year plus: the good the bad the ugly
19mOpenTelemetry's Path to Observability's Future
35mOpenTelemetry for Busy People
40mInstalling OpenTelemetry, Today and Tomorrow
37mHow To Adopt OpenTelemetry in an Enterprise Where Incumbent Vendor Tools Reign Supre...
22mOpenTelemetry – An Observability Framework for Cloud-Native Software