What WePay Learned From Processing Billions of Dollars on GKE Using Linkerd
summary
WePay migrates all synchronous REST and gRPC payment traffic on GKE to Linkerd and Namerd, moving discovery and telemetry out of services and into a data plane; tools: Linkerd1, Namerd, Prometheus, Grafana, NewRelic. A custom probing service for HA verifies internal health when external checks can't, exposing service liveness across clusters. Migrations hit a Namerd memory leak and an old HTTP client causing ~50ms latency spikes, both diagnosed via mesh metrics; the full migration of production completes with zero downtime after staged canary and rolling upgrades while having processed billions of dollars through the mesh.
AI-written summary · about
chapters (5)
- 0:00Start
- 5:55Micro Services
- 12:27Microservices
- 27:03Rolling Upgrade Pattern
- 30:59How Do You Handle Backward Incompatible Changes
related talks
39mSeamless multi-cluster Communication and Observability with Linkerd
35mUpgrading Your Service Mesh to Linkerd 2
06mBenchmarking Journey Through Service Meshes
28mHow a Service Mesh Helped Us Build Production Cloud-Native Services
20mOverview and State of Linkerd
25mScheduling 68k COVID Tests with Linkerd
30mService Mesh Is Still Hard. The Things we all Did Wrong & Tales of Woe
35mOverview And State Of Linkerd