FluxyChatHow-to Guides
Room SQL point-in-time recovery
Snapshot and restore a room Durable Object SQLite database (30-day Cloudflare PITR) from the console.
Every FluxyChat room is a SQLite-backed Durable Object (new_sqlite_classes). Cloudflare retains 30 days of that database (SQL + KV). FluxyChat exposes it as a product, not a wrangler trivia fact.
Operator path
- Open a room in the dashboard.
- In Room insights, use Snapshot now before a risky cutover.
- After a bad write, Restore on the snapshot you want.
- The object applies
onNextSessionRestoreBookmarkand reloads that bookmark the next time it is constructed (hibernation / eviction). Live sockets reconnect.
API
# List current bookmark + named snapshots (any member)
curl "$WORKER_URL/rooms/$ROOM_ID/pitr" \
-H "Authorization: Bearer $MEMBER_JWT"
# Named snapshot (owner/admin)
curl -X POST "$WORKER_URL/rooms/$ROOM_ID/pitr" \
-H "Authorization: Bearer $ADMIN_JWT" \
-H "Content-Type: application/json" \
-d '{"action":"snapshot","label":"before-migration"}'
# Restore (owner/admin) — takes effect on next DO wake
curl -X POST "$WORKER_URL/rooms/$ROOM_ID/pitr" \
-H "Authorization: Bearer $ADMIN_JWT" \
-H "Content-Type: application/json" \
-d '{"action":"restore","snapshotId":"pitr_…"}'What this does and does not restore
| Restored | Not in the room SQLite |
|---|---|
| Room DO SQL tables + DO KV (schedules, cost ledger, PITR catalog) | D1 messages history (tenant DB) |
| Live stream offsets / alarm job map stored on the object | R2 attachments |
| Agent schedule rows on the room | Other rooms |
D1 remains the tenant system of record for durable chat history. PITR is the room kernel (hot state). Use both in an incident: restore the room object, then decide whether D1 rows need a separate export replay.
Tests
Hibernation + alarm queue + schedule claim run in workerd via pnpm --filter @fluxy-chat/worker test:pool (@cloudflare/vitest-pool-workers).