Switching Jaeger Distributed Tracing to ClickHouse® to Enable APM
summary
Migrates Jaeger storage from Cassandra to ClickHouse, replacing a Bitnami Cassandra Helm chart with the Kubernetes ClickHouse operator and wiring the Jaeger gRPC storage plugin to a ClickHouse installation. Claims: near-zero Kafka consumer lag remains, ClickHouse gives much smaller storage, batch ingestion, SQL aggregates and materialized columns that enable P90 and top-50 slow-transaction dashboards in Grafana. Gotchas: ArgoCD annotations, default load‑balancer service, no read‑only user, slow deletes (handled with TTL retention), and backups via a CronJob image — overall the stack works and enables richer APM analytics.
AI-written summary · about
chapters (7)
- 0:00Introduction and motivation
- 1:34Understanding Jaeger traces
- 4:28Tracing architecture and storage
- 7:28Why move to ClickHouse
- 9:33Implementation and gotchas
- 13:15Operations and visual insights
- 15:36Q&A and concluding remarks
related talks
38mDistributed tracing with Jaeger and ClickHouse
37mDeep Dive: Jaeger – Steve Flanders
55mBuilding Fast ClickHouse Analytics for MySQL Transaction Data Using Kubernetes
41mJaeger: Intro and Deep Dive
28mDeep Dive: Jaeger
33mDeploy, Scale and Extend Jaeger
26mO11y in One: Exploring ClickHouse as a Unified Telemetry Database
37mDeep Dive: Jaeger