MinIO

Use it when

• Lake storage must run outside a public cloud: on-premises, air-gapped, or at the edge, with the full S3 API so table formats and engines work unchanged.
• Development should mirror production; a single binary on a laptop behaves like the S3-backed real thing.
• The event fan-out matters: bucket notifications publish to Kafka, AMQP, NATS, Redis, Postgres, MySQL, or webhooks.
• Multi-site resilience is needed without a hyperscaler: bucket replication or active-active site replication.

Think twice when

• The AGPL-3.0 license does not fit how you distribute or offer the software; the commercial license is the alternative.
• Someone must own capacity, upgrades, and drive failures; managed cloud storage removes that work entirely.
• The cluster layout is fixed at deployment; erasure sets shape how it can grow, which rewards planning up front.
• Compliance demands third-party attestation of durability rather than your own operations.

How it runs

A single Go binary, from one drive on a laptop to distributed erasure sets of up to 16 drives with Reed-Solomon coding (EC:4 by default) and bitrot checksums. Identity and policy follow the IAM model; Object Lock covers governance and compliance WORM. Lifecycle rules tier objects out to S3, Azure, or GCS.