TechPrufer
Get started
Grab
Grab logoGrab · Feature trail

Service Mesh Evolution (Consul to Istio)

2 posts · Oct 2016 – Jul 2025

What this trail covers

Grab has evolved its service mesh infrastructure from a Consul-based setup with a fallback mechanism (Catcher) to Istio. This migration addresses challenges with single points of failure, limited advanced features, and multi-cluster support. The new architecture utilizes multiple, resilient Istio control planes deployed on dedicated Kubernetes clusters, supporting both HTTP and gRPC traffic across multi-cloud and multi-organization environments. The migration journey involved cross-cloud (GCP to AWS) and cross-mesh connectivity, with a focus on stability, advanced traffic management, and enhanced security policies.

All Grab trails
  1. Grab's service mesh evolution: From Consul to IstioThis post details Grab's migration from Consul to Istio for its service mesh. Key technical contributions include designing a resilient architecture with multiple external control planes on dedicated Kubernetes clusters for high availability and isolation, supporting both HTTP and gRPC traffic across multi-cloud (GCP to AWS) and multi-organization environments. The migration involved careful planning for gradual traffic shifting, quick rollback mechanisms, and maintaining service continuity. Lessons learned emphasize questioning assumptions, balancing innovation with stability, and adapting to changing needs.
  2. This Rocket Ain't Stopping - Achieving Zero Downtime for Rails to Golang API MigrationThis post details the migration of Grab's passenger app APIs from a Rails + NodeJS stack to a Golang Service Oriented Architecture. It outlines a zero-downtime migration strategy involving rigorous testing (unit, load, staging QA, shadow), phased endpoint migration, progressive deployment, and instant rollback capabilities. Key technical challenges and solutions discussed include: the use of gRPC between Go and Rails for latency optimization, issues with ELB TCP balancing and ASG scaling events, a memory leak in the Ruby gRPC gem, latency mismatches, and the development of a JSON tokenizer for response comparison during shadow testing. The post also covers handling idempotency for PUT/POST/DELETE requests during shadowing, progressive shadowing with RPS increments, prewarming ELBs, and a gradual DNS switch for the final rollout.