Principle:Apache Druid Supervisor Submission
| Knowledge Sources | |
|---|---|
| Domains | Streaming_Ingestion, Task_Management |
| Last Updated | 2026-02-10 00:00 GMT |
Overview
A supervisor deployment principle that submits a streaming ingestion specification to create a long-running Druid supervisor that continuously reads from a message broker.
Description
Supervisor Submission creates a persistent streaming ingestion supervisor in Druid. Unlike batch tasks that run once and complete, supervisors:
- Run continuously, creating new indexing tasks as needed
- Manage consumer offsets / sequence numbers
- Handle task failures with automatic retry
- Support hot-swapping of specs without data loss
- Can be submitted in a suspended state for staged deployments
The supervisor spec is POSTed to /druid/indexer/v1/supervisor. If a supervisor with the same ID already exists, the new spec replaces the old one (spec update). The wizard optionally shows a diff between the current and new spec when editing.
Usage
Use this principle as the terminal step of the streaming ingestion wizard, after all schema and configuration steps are complete. Submitting the supervisor starts continuous data ingestion from the configured stream.
Theoretical Basis
Supervisor submission follows a declarative deployment pattern:
Supervisor Spec (JSON) → POST /druid/indexer/v1/supervisor
Response → { id: string } // Supervisor ID
Supervisor lifecycle:
PENDING → RUNNING → (continuously creates indexing tasks)
Can be: suspended, resumed, reset, terminated
Spec update:
POST same endpoint with new spec → hot swap (no data loss)
Compare old vs new spec to verify changes