Design Juspay's Payment Orchestration Layer Across Multiple Gateways

Problem

Design Juspay's payment orchestration layer — the routing and retry engine that sends payment requests to the optimal gateway and handles failures.

Requirements

Functional:

• Route payment to best gateway based on: success rate, latency, merchant preference, instrument type
• Retry on failure: different retry strategies per failure type (timeout vs decline)
• Support A/B testing of gateway routing rules
• Provide unified status to merchant regardless of which gateway processed

Non-functional:

• 500K TPS at peak
• End-to-end latency < 1.5s (NPCI mandate)
• Zero fund loss: idempotent retries, no double charge
• 99.99% availability

Critical discussion points

1. Idempotency key propagation through retry chains
2. Gateway health scoring: real-time rolling success rate
3. Fallback routing: when primary gateway is degraded
4. Saga pattern for atomic debit + routing + confirmation
5. Audit log for every routing decision (regulatory requirement)