Choose a storage backend

Avalanche exposes backend-neutral namespace and table contracts. Use Iceberg by default. Install the Lance extra when a Lance-backed namespace is the right local fit:

uv sync --extra lance

​

Iceberg

ava.IcebergNs, ava.IcebergNsConfig, ava.IcebergTable, and ava.IcebergTableGroup model Iceberg-backed storage. Iceberg tables expose the common table operations and available backend-native PyIceberg methods such as history() and snapshots().

​

Lance

Declare a Lance namespace with the same broad shape:

import avalanche as ava


class DocumentLanceNamespace(ava.LanceNamespace):
    ns_config = ava.LanceNamespaceConfig(
        name="documents",
        base_location=".avalanche/data/lance",
    )

    documents = ava.LanceTable(schema=DocumentSchema)


ns = DocumentLanceNamespace()
ns.push()

Lance supports append, scan, read, history, and current_version_id.

​

Keep application code portable

Depend on ava.Namespace and ava.Table in application-level functions when the backend should remain replaceable. Choose a concrete namespace/table class at the boundary where the catalog and storage location are configured.

​

Current limits

Schema migration commands are not part of this release candidate. IcebergTable.merge(...) is not implemented; use append/read/overwrite or backend-native PyIceberg operations. Repository examples use local SQLite catalogs and local filesystem paths; they are examples, not an Avalanche global storage default.