Capabilities

Field Manual Section 4 - Operational Limits

Not all tanks are built for every terrain. While Tank provides a unified API, the underlying engines (Postgres, Redis, Mongo, etc.) have vastly different mechanical limits. This matrix details which operations are natively supported, emulated, or impossible on each driver.

Feature Matrix

Feature	Postgres	SQLite	MySQL/MariaDB	DuckDB	MongoDB	ScyllaDB/Cassandra	Valkey/Redis
Transactions	✅	✅	✅	✅	✅	❌	❌
Joins	✅	✅	✅	✅	❌	❌	❌
Bulk Insert	✅	✅	✅	✅	✅	✅	✅
Filters (WHERE)	✅	✅	✅	✅	✅	⚠️	❌
Aggregations	✅	✅	✅	✅	✅	❌	❌

Semantic Differences

Transactions

• Interactive: Standard SQL drivers and MongoDB allow you to read, perform logic, and write within the same transaction context.
• Batch / Pipeline: ScyllaDB and Valkey operate on a queue-and-commit model. You queue up multiple write operations, and they are executed atomically when you commit. You cannot read data from the transaction before it is committed.

Filters

• Standard: SQL drivers and MongoDB support complex WHERE clauses with nested conditions, ranges, and logical operators.
• Key-Value / Wide-Column: ScyllaDB and Valkey are optimized for primary key lookups. ScyllaDB supports some filtering on clustering keys and secondary indexes but with performance caveats. Valkey only supports direct key lookups.

Returning Data

• RETURNING: Postgres, SQLite, DuckDB, and MongoDB can return the fully generated row (including auto-increment IDs and default values) immediately after insertion.
• LastInsertId: MySQL provides the ID of the last inserted row, but not the full row content. Tank handles this transparently where possible.
• None: ScyllaDB and Valkey do not return the inserted data. You must know the ID beforehand or query it separately (which may not be consistent in eventually consistent systems).

Study the specs. Pick the right armor. Dominate the field.