Kubernetes in the Datacenter: Squarespace’s Journey Towards Self-Service Infrastructure
summary
Squarespace migrates hundreds of VMs into Kubernetes using Calico BGP on a leaf‑spine L3 fabric to announce pod and service IPs, replaces Graphite+Sensu with per‑team Prometheus operator instances and a central AlertManager, and uses Ceph RBD and CephFS via Kubernetes storageclasses for persistent storage. Java services hit garbage‑collector and thread‑count failures on 64‑core nodes, fixed by a base container that reads cgroup quotas and an LD_PRELOAD shim to override available processors, enabling rapid self‑service deployments and ~20 new services planned.
AI-written summary · about
chapters (42)
- 0:00Introduction
- 0:24Background on Squarespace
- 1:05Squarespace in 2013
- 1:37Squarespace in 2014
- 2:21Squarespace in 2016
- 2:48Manual provisioning workflow
- 3:58Microservices do not mix
- 4:42Selfservice infrastructure
- 5:09Kubernetes
- 5:32Service Definitions
- 7:46CPU Limit
- 9:07Stress Testing
- 9:56Java Libraries
- 10:36Base Java Container
- 11:13C Shared Library
- 11:52SelfService Compute
- 12:08Java Service Configuration
- 12:53calico
- 13:58topology
- 16:19Pod IPs
- 17:42SelfService Networking
- 18:26Federation
- 18:52Prometheus
- 21:37Prometheus and Kubernetes
- 22:38Prometheus Operator
- 23:27Prometheus Alert Manager
- 23:47Prometheus Alerts
- 24:20SelfService Storage
- 25:34SEF
- 26:17Automatic Provisioners
- 26:47RBD Provisionary
- 27:29Purpose Persistent Volume
- 28:00NFS Shared Storage
- 28:52Benefits
- 30:09Databases
- 30:39Prometheus Cluster
- 31:05New Services
- 32:27Redundancy
- 33:10Configuration
- 33:53Performance
- 34:37Memory
- 35:26Maces
related talks
35mKubernetes journey
38mBuilding the Perfect Infrastructure with Kubernetes
36mOptimizing Resource Usage in Kubernetes
41mKubernetes the Very Hard Way
32mRunning Kubernetes in a Manufacturing Line – What Could Possibly...
17mOur Journey to Kubernetes at Scale
40mOperating a Self-Healing Bare-Metal Kubernetes Platform at Global Scale
29mA history of GOV.UK's infrastructure: the long voyage to Kubernetes
