about the question i don't understand when i see the torials

Sounds like Indexer nodes are optional, and you can ingest using Realtime nodes directly?
Indexer node how to HA?
Without using Indexer, the Realtime nodes are not HA?
If using Indexer then you don’t need Realtime nodes?

You do need some method of ingestion, although there are two currently: the indexing service (Overlord + MiddleManagers) and standalone stuff (Realtime nodes, Hadoop CLI indexer). So if you’re using realtime nodes, you don’t need the indexing service (and vice versa).

The indexing service can be run fully HA by doing the following,

  1. Run multiple overlords

  2. Run in remote runner mode, with MiddleManagers

  3. Push data in with tranquility and set “replicas” to something > 1

Realtime nodes are not generally HA, although they can recover from failures or restarts fairly quickly.