carlos / platform Public

Adds a "Database" stats tile to the per-app Overview page in the console (console.carloku.com/a/<account>/apps/<app>), alongside the existing Running / Resources / Web addresses / Release history tiles — requested after reviewing a Supabase "Primary Database" card (region, instance type, CPU/Disk/RAM/connections) as a reference shape.

Shows two numbers per app:

• Total on-disk SQLite size across the account's own routes (file + -wal), reusing the existing status.BoxMetrics.RouteDBBytes join (loadBoxJoin/dbSize) that already backs the carlos instances/carlos routes APIs — no new backend plumbing needed, that pipeline already existed end-to-end from an earlier "show database and bucket sizes" commit.
• The fullest disk-used fraction among the boxes backing the app's reporting instances (worst case, not an average) — reusing the existing internal/metrics sweep (BoxMetricsView.Disk) the same pipeline already used for the Resources tile's CPU/Mem.

Motivation: while scoping this I investigated the bdn/ips app's 3 instances failing repeatedly overnight (2026-10-01) — root cause was one instance's SQLite database hitting SQLITEFULL, crashing it, then every instance on the same box failing to even reopen its database (SQLITEIOERR_SHMSIZE) for the next 4h18m. Disk headroom per instance isn't visible anywhere on the tenant-facing console today; this tile is a step toward that, hence DiskFracMax deliberately surfaces the WORST box rather than an average.

Status: implemented, tested (new assembleOverview unit tests pin DiskFracMax-is-max-not-average and DBSizeSum-sums-regardless-of-sweep-reporting), build/vet/gofmt clean, internal/console's full test suite green. Two pre-existing test failures observed in this sandbox (internal/tunnelclient, internal/edge's fleet e2e tests) are unrelated tunnel/relay-subsystem tests with no import relation to this change — flagged in a separate discussion rather than investigated here.