Skip to main content
Verglas supports two common destinations for Pipeline records:
  • Sink + Catalog publishes Parquet files and Iceberg table metadata for interoperable lakehouse storage.
  • Query maintains bounded aggregate read models for application endpoints.

Iceberg publication

The Pipeline determines when a batch is ready. The Sink validates and deduplicates its batch identity. The Catalog publishes the table metadata only after the immutable data and metadata objects exist. Retrying the same batch returns the stored receipt rather than publishing a second snapshot. Catalog implements the Iceberg REST /v1 surface for namespace and table list, create, load, commit, register, rename, and delete operations. Standard Iceberg clients continue to treat the catalog and object store as the table interface; no Verglas-specific reader is required.

Query read models

Declare a fixed query binding:
Call one endpoint declared by that Query component:
Use describe() to inspect the materialization and its source watermarks:
Query definitions are immutable and operator-managed. They declare input sources, aggregate views, typed parameters, and bounded endpoints. A Query DO consumes deterministic Pipeline batches and durably stores both materialized rows and replay receipts.

Choose a destination