Storage

Land files.
Name tables. Serve SQL.

Buckets hold bytes. Catalogs register Datasets. Warehouse tables answer SELECT. Pipelines write here. Ontology may read here. Storage is its own product.

Assembly

Files, a named table, serving SQL

sark · ga4_sessions

Bucket

landing

s3_…

Dataset

ga4_sessions

ice_…

Warehouse table

ga4_sessions

wt_…

You do not have to use every box. A bucket alone is files. A Dataset needs a catalog. A warehouse table needs an engine. They compose; they do not stack.

Model

Bucket, table, engine

Bucket

Amazon S3 or Cloudflare R2. Keys and bytes — CSV, Parquet, dumps, anything with a path. Not a table. Not SQL.

Dataset

An Iceberg table on Glue. Schema, snapshots, a name you can read. The catalog points at files in a bucket.

Primitives

Vendor products, not Sark nouns

You create or bind the thing they already have a name for. Sark wraps it. It does not become a different SKU.

Amazon S3

s3

Object store. Files only. Hosted, managed, or BYO.

Cloudflare R2

r2

Object store. Files only. Same job as S3, different cloud.

AWS Glue

glue

Catalog for Iceberg Datasets. Firehose, Lake Formation, and Athena are backing — not separate creates.

R2 Data Catalog

r2-catalog

Catalog for Iceberg Datasets on Cloudflare. Registers tables; it is not the query engine.

Dataset

iceberg

The table you create (ice_…). Iceberg is the format. Athena reads it.

Assemblies

Use one, or wire two

Bucket

Files only

Create S3 or R2 and you get keys. Upload, archive, land dumps. Nothing can SELECT until a catalog or an engine knows the columns.

Catalog

Named tables

Glue registers Datasets on those files. iceberg.read / iceberg.write. Athena is the query. The bucket is backing.

Are

What Storage is

  • Buckets, catalogs, and Datasets — the store, not the graph
  • Vendor primitives you create or bind: S3 and Glue
  • A peer of Connect and Pipelines. A pipeline writes here. It does not own the lake
  • Usable without ontology. Funnel may read a table; it does not live in the bucket

Are not

What it is not

  • A bucket that is secretly a warehouse. Drop a CSV and you still have a file
  • Ontology Object Store. Instances are not ice_… or r2sql_… rows
  • Connect. A warehouse you already run is a connector, not Sark storage
  • Pipelines. The graph moves data. Runs belong to the pipeline

FAQ

Frequently asked questions

Get started

Create a bucket, a catalog, or an engine. Land a table when you have one.

Get started