Migration 054 — input files

Migration 054_input_files.sql ships with file inputs (type: file). It adds tables only. It does not change an existing table and adds no index to a large table.

The tables

• input_file holds one row per uploaded file: the ID, the workspace and task it was uploaded for, the uploader, the origin (api, mcp_inline or mcp_url), the original and the safe name, the content type, the size, the SHA-256 and the storage key.
• job_input_file links a job and a field to a file (primary key (job_id, field)). Many jobs can use one file.
• input_file_upload_token holds the single-use MCP upload URLs. Only the SHA-256 of a token is stored.

The migration also adds two indexes, both on the new tables (input_file.created_at for the orphan sweep, job_input_file.file_id). No manual pre-run is needed. The server applies the migration at startup.

job.job_id and "user".user_id gain foreign keys from the new tables. Job retention now deletes a job’s job_input_file rows in the same transaction as the job. Deleting a user keeps that user’s uploads (uploaded_by_user_id becomes NULL) and deletes the user’s unused upload URLs.

Behaviour change: file is a reserved type name

file is now a primitive input type. A workspace that defines a connection type named file behaves differently after the upgrade:

• A task or action input written as type: file is now a file field, not a reference to that connection type. A caller that sends a connection name for it gets Input field '…': the file cannot be used.
• stroem validate rejects the workspace with Connection type name 'file' is reserved (collides with a primitive input-field type or alias).

Rename such a connection type (and every type: file input that refers to it) before you upgrade the server.

Rollout order

An older server replica treats type: file as a connection type and answers 400 for a task that uses it. Do the steps in this order:

1. Upgrade every server replica.
2. Upgrade every worker. A step that has files and runs on an older worker fails with this step has file inputs, which the worker does not support; upgrade the worker.
3. Add workflow YAML that uses type: file.

Check the blob storage for uploads before step 3: see Storage.