seamless-database

seamless-database is the checksum-based metadata and caching service for the Seamless framework. It acts as the distributed computation cache that allows Seamless workflows to avoid recomputing identical transformations, both within a single session and across the entire cluster.

How it works

Seamless uses content-addressed storage: every piece of data (buffers, code, parameters) is identified by its checksum. When a transformation (computation) is submitted, its inputs are hashed into a transformation checksum. Before executing the computation, Seamless components (such as seamless-dask) query the database: "has this transformation been computed before?" If a cached result is found, the result checksum is returned immediately, skipping the computation entirely.

The database stores the following kinds of records:

All data is persisted in a single SQLite file (typically seamless.db).

Role in the Seamless ecosystem

Other Seamless components interact with the database over HTTP:

• seamless-dask checks the database cache before scheduling a transformation on the Dask cluster, and writes results back after computation.
• seamless-remote provides the DatabaseClient / DatabaseLaunchedClient classes that other components use to communicate with the database server.
• seamless-config defines the launch template for the database server (port range, host, timeout, read/write mode).

The server exposes a JSON-over-HTTP protocol: clients send {"type": "<record_type>", "checksum": "<hex>", ...} via GET (read) or PUT (write) requests.

Installation

pip install seamless-database

Usage

# Start a writable database server on a random port
seamless-database seamless.db --port-range 5520 5530 --writable

# Start a read-only server on a fixed port
seamless-database seamless.db --port 5522

If --port and --port-range are both omitted, seamless-database picks a random free port in the dynamic/private range (49152-65535).

Status-file protocol

seamless-database does not require a status file. If --status-file is omitted, it runs independently.

If --status-file is provided, the file is used for two things:

1. Report the chosen port, especially when --port-range is used.
2. Report whether startup succeeded ("running") or failed ("failed").

The status-file protocol is simple:

1. Wait for the status file to exist and parse it as JSON.
2. Reuse the existing JSON object as the base payload. An empty JSON object {} is sufficient.
3. Choose or validate its listening port.
4. Once the HTTP server is up, rewrite the same file with "status": "running" and the selected "port".
5. If startup fails before the server is running, rewrite the file with "status": "failed" instead.

If remote-http-launcher is used, it may pre-populate the JSON with fields such as the PID, workdir, or "status": "starting". seamless-database preserves such fields when it writes back the final status.

CLI options

CLI scripts

Installing seamless-database also provides:

• seamless-database