| name | status-page-data-layer |
| description | Conventions for the SQLite data layer in this status-page app (lib/db.ts). Use this skill whenever adding tables, columns, queries, or dbOps methods, writing migrations, or persisting configuration (monitors, settings, checks). Trigger on phrases like "add a table", "add a column", "store a setting", "query the database", "add a dbOps method", "migrate the schema", or any change touching lib/db.ts, lib/seed.ts, or SQLite persistence. Apply automatically when working with data persistence here.
|
Status Page Data Layer
This project persists everything in a single file-based SQLite database via the sqlite3 package.
There is no ORM. All access goes through lib/db.ts.
Core rules
-
Use the promise helpers, never raw callbacks. lib/db.ts wraps sqlite3 in three helpers:
run(sql, params) — INSERT/UPDATE/DELETE/DDL; resolves to the statement (this, has lastID/changes).all<T>(sql, params) — multi-row SELECT.get<T>(sql, params) — single-row SELECT (or undefined).
Always pass values as params (parameterized ?), never string-interpolate user input.
-
Migrations are additive and idempotent, inside initDb(). There is no migration runner.
Schema evolution = add CREATE TABLE IF NOT EXISTS ... / CREATE INDEX IF NOT EXISTS ... to initDb(),
or guarded ALTER TABLE for new columns (check PRAGMA table_info before adding, since SQLite has no
ADD COLUMN IF NOT EXISTS). initDb() runs on every server start via instrumentation.ts, so it MUST be
safe to re-run. Never drop or rename columns that hold live data.
-
Never break checks.site_id compatibility. The checks table is keyed by site_id TEXT.
Monitor IDs are TEXT slugs and must stay stable so historical check rows remain linked. Do not switch
monitor IDs to integers, and do not rename existing seeded IDs.
-
Settings are a key/value JSON table. The settings table is (key TEXT PRIMARY KEY, value TEXT, updated_at INTEGER)
where value is a JSON blob. One row per logical group: branding, notifications, retention, admin.
Read with getSetting<T>(key) (JSON.parse, returns undefined if absent); write with
setSetting(key, obj) using INSERT ... ON CONFLICT(key) DO UPDATE SET value=?, updated_at=?.
Do NOT add a column per setting — add a new key instead.
dbOps method naming
All query logic lives in the exported dbOps object. Conventions:
listX(opts?) → array; getX(id) → single or undefined; insertX(obj); updateX(id, patch) (always set updated_at = Date.now()); deleteX(id); countX().- Timestamps are epoch milliseconds (
Date.now()), stored as INTEGER. Booleans are 0/1 INTEGER.
- Keep batch/dashboard queries (e.g.
getDashboardData(siteIds[], days)) returning raw rows; shape them in the route, not the dbOps method.
Delete semantics
Deleting a monitor removes the monitors row but retains its checks history (so an accidental delete
doesn't destroy uptime data). Provide history purge only as a separate explicit operation (pruneChecks(beforeTs)).
Where things live
- Schema + helpers +
dbOps: lib/db.ts
- First-run seeding (only when a table/row is empty):
lib/seed.ts
- Typed setting resolvers with env fallback:
lib/settings.ts
