ADR-013: Seat Hold TTL Alignment with Checkout Session TTL

Status: ✅ Approved — 2026-04-12 Impacts: schema-commerce.md (§seat_reservations, §checkout_sessions)

---

Context

SeatReservation originally used a 10-minute hold TTL (hold_expires_at), while CheckoutSession used a 30-minute session TTL (expires_at). This mismatch creates a broken UX: a passenger can hold a valid checkout session but lose their seat at the 10-minute mark. This causes a confusing failure at payment time ("Your seat is no longer available").

Conversely, holding seats too long blocks inventory and reduces conversion for other shoppers.

Options Evaluated

#	Option	Pros	Cons
1	Extend seat hold to match session TTL (30 min)	Simple, consistent, no UX failure risk	Blocks inventory longer
2	Heartbeat refresh during checkout	Inventory-friendly (short default + extend on activity)	Complex frontend logic, requires WebSocket or polling
3	Re-acquire at payment time	Minimal hold time	Risk: another shopper may already hold the seat, poor UX

Decision

Option 1 for V1: Align seat hold TTL to checkout session TTL (30 minutes).

• SeatReservation.hold_expires_at is set to the same value as CheckoutSession.expires_at at session creation time.
• Both expire simultaneously, maintaining consistency.
• The seat_hold_cleanup Hasura Scheduled Trigger (runs every 60s) releases expired holds.
• The checkout_abandoned_sweep Hasura Scheduled Trigger (runs every 5 min) expires abandoned sessions.

Phase 2 consideration: The team may implement Option 2 (heartbeat refresh) as a yield optimization when inventory pressure increases. This would shorten the default hold to 10 minutes but extend on frontend activity (every 5 minutes), capped at the session TTL.

Consequences

Positive:

• Zero UX failures from mismatched TTLs
• Simple implementation — single TTL source, no frontend polling
• Consistent cleanup: both triggers work in harmony

Negative:

• 30-minute seat holds are longer than typical e-commerce (5–15 min). Acceptable for bus tours where operators measure inventory in dozens of seats, not thousands.

Neutral:

• Phase 2 heartbeat optimization is backwards-compatible (reduces TTL, doesn't change cleanup mechanics)