TechPrufer
Get started
Slack
Slack logoSlack · Feature trail

Internal Compute Platform (Bedrock)

3 posts · Oct 2021 – Mar 2022

What this trail covers

This thread tracks the evolution of Slack's internal compute orchestration platform, code-named "Bedrock." Initial posts detail the foundational architecture, which enables developers to build, package, and deploy Docker containers, abstracting away complexities of Kubernetes, Jenkins, Consul, Vault, Envoy, and Nebula. Bedrock aims to simplify production-grade service launches, reduce configuration, and accelerate feedback loops. This post details the implementation of a simple, stateless admission controller for Bedrock, enhancing its ability to manage Kubernetes StatefulSet rollouts with enhanced control, addressing limitations of native Kubernetes strategies for stateful applications.

All Slack trails
  1. Applying Product Thinking to Slack’s Internal Compute PlatformThis post introduces the concept of applying product thinking to internal platforms, specifically Slack's internal compute orchestration platform, Bedrock. It details how Bedrock leverages Kubernetes and other infrastructure components to simplify service deployment. The post emphasizes the importance of staying close to internal customers (developers) through user experience interviews, user surveys, NPS, dogfooding, a user advisory group (Bedrock Advisory Group), and self-paced training materials to gather feedback and inform the platform's roadmap and development.
  2. A Simple Kubernetes Admission WebhookThis post details the implementation of a simple, stateless Kubernetes admission webhook in Go. It explains the core concepts of admission controllers and webhooks, and provides a practical example of creating both validating and mutating webhooks. The post covers the Kubernetes configuration required for deploying and registering the webhook (Deployment, Service, Secret, ValidatingWebhookConfiguration, MutatingWebhookConfiguration) and the Go code for handling admission requests, including parsing `AdmissionReview` objects and returning `AdmissionResponse`.
  3. Infrastructure Observability for Changing the Spend CurveThis post details the implementation of a simple, stateless admission controller for Bedrock, enhancing its ability to manage Kubernetes StatefulSet rollouts with enhanced control, addressing limitations of native Kubernetes strategies for stateful applications. It also discusses the use of adaptive capacity and oversubscription to improve CI throughput and cost-effectiveness, and the implementation of circuit breakers to prevent downstream service overload during CI degradations.