30K+ TPS Transaction Platform

Go services on Kubernetes with Kafka and Pub/Sub ingest, Redis state, and partner rails — built to never lose a transaction and survive partial failure

30K+ TPS Transaction Platform Go services on Kubernetes with Kafka and Pub/Sub ingest, Redis state, and partner rails — built to never lose a transaction and survive partial failure GCP us-central1 / GKE mTLS mesh Clients · merchants / apps · Architecture component Clients merchants / apps Load Balancer · GCLB :443 · GCP us-central1 / GKE Load Balancer GCLB :443 Ingest Bus · Kafka + Pub/Sub · GCP us-central1 / GKE · at-least-once Ingest Bus Kafka + Pub/Sub at-least-once Go Services · GKE autoscaled · GCP us-central1 / GKE › mTLS mesh · 30K+ TPS Go Services GKE autoscaled 30K+ TPS Redis · state + dedup · GCP us-central1 / GKE Redis state + dedup Observability · SLO + alerts · GCP us-central1 / GKE Observability SLO + alerts Cloud Spanner · durable ledger · GCP us-central1 / GKE › mTLS mesh Cloud Spanner durable ledger Partner Rails · banks / NPCI · Architecture component Partner Rails banks / NPCI HTTPS publish consume state commit settle metrics Legend External Cloud Message bus Backend Database Security

Durable Ingest

  • • Every request lands on Kafka or Pub/Sub first
  • • At-least-once delivery means no transaction is dropped
  • • Backpressure absorbs bursts above steady-state TPS

Stateful Core

  • • Go services autoscale on GKE past 30K TPS
  • • Redis holds hot working state and dedup keys
  • • Cloud Spanner is the durable, ordered ledger

Survive Partial Failure

  • • mTLS mesh isolates the services and ledger
  • • Partner rails settle out of the hot path
  • • SLO telemetry drives alerting and safe degradation