Druid

Apache Druid is a database for exploring large collections of events, such as website activity, application measurements or network records. It prepares that data for analytical questions like counts, trends and breakdowns across dimensions. It is useful for applications and dashboards whose users need interactive access to those aggregates. Operating Druid involves both bringing data in and serving queries, with separate storage and coordination needs behind the visible query endpoint.

Deployment and operating notes

Apache Druid’s current Kubernetes page links its Docker image, a Druid operator and Kubernetes discovery extensions that can support operation without ZooKeeper. These are deployment choices, not evidence that every old chart value has a direct equivalent. Separate ingestion, query-serving, metadata storage and deep storage when assessing the architecture.

For an existing cluster, inventory datasources, ingestion specifications, supervisors, segment storage, metadata database and authentication. Preserve metadata and deep-storage objects together; running new pods without those dependencies does not recover the data. Follow a version-supported rolling-upgrade sequence and verify extension compatibility, especially when changing discovery or task execution. Rehearse a representative ingestion job, historical query and late-arriving event, then compare segment availability and query results before moving clients. Keep source ingestion positions and a plan for duplicate or missed events during cutover. A Helm rollback alone cannot reverse metadata migrations or reconcile ingestion that continued after the switch.