2.2 KiB
SQLite latest schema definitions
These SQL files are embedded into the server binary and are the authoritative
source for the tables and columns managed by the application. When adding a
table or column, update its definition here; do not add a separate list of
ALTER TABLE ADD COLUMN migrations in startup or handlers.
Startup opens an isolated in-memory reference database with the latest SQL, compares it with the actual SQLite file, and adds missing tables and columns in one transaction before creating missing indexes and running business queries. Existing rows, extra tables and extra columns remain in place. Repeated startup is safe. The process does not change types or constraints of existing columns.
Column additions use their exact definition, including defaults. SQLite rejects some additions (for example PRIMARY KEY, UNIQUE, STORED generated columns, or a required column without a suitable default on a populated table). Such failures roll back structure repair and stop startup with the table/column and cause. Resolve the definition or explicitly plan a data migration, then restart; the repair engine does not invent required values or weaken constraints.
Historical data and foreign keys
Historical data backfills and foreign-key semantic changes are separate from missing-table/column repair. Existing data migrations remain responsible for time values, token usage, tool guards, and approvals. The historical vulnerability foreign-key migration creates the destination from the latest definition and copies existing columns dynamically, preserving extra columns, indexes, triggers, and dependent records. It is not the source of missing-column checks.
Both application and knowledge databases use the same reconciliation engine. HITL and notification-read tables are included in the embedded definitions, so request handlers do not need to create those tables at request time.
Regression coverage
Schema reconciliation tests cover missing tables and columns, complex defaults, repeated startup, rollback on failure, and preservation of existing data and extra schema objects. Keep these properties when changing definitions or historical migrations.