Why it matters

Metadata perf affects query planning. Understanding shapes efficient Impala.

Advertisement

The architecture

Catalog service: reads HMS; publishes updates.

StateStore: distributes to Impalads.

Impalad: local cache.

Impala metadata flowCatalog servicereads HMSStateStoredistributes updatesImpalad cachelocal metadataINVALIDATE METADATA forces re-read; REFRESH updates single table
Metadata pipeline.
Advertisement

How it works end to end

Initial load: eager or lazy. Eager loads all metadata upfront.

Updates: HMS notifications for auto-refresh (newer versions).

Manual: INVALIDATE / REFRESH for staleness.

Sizing: large metadata increases memory + startup.