Continuous Delivery Lessons Learned Through Hypergrowth
summary
Built deployment tooling that runs a one-minute sync dbops cron across 10k+ machines using SSH, HDPool and monit, supporting daily deploys with teams of 10–40; growth drives time-to-deploy from about 20 minutes to 3 days as operational overhead and human coordination explode. Moved ~700 PB into Magic Pocket in 24 months, showing infra can be provisioned separately from apps. Key outcomes: keep provisioning distinct from service deploys, design isolation-aware convergence (not linear pipelines) to handle tenant, locality and version-pin complexity.
AI-written summary · about
chapters (5)
- 0:00Start
- 21:07Custom Approvals Configuration Differences Variance in Versions
- 22:16Communicating is Hard (With Distributed Systems)
- 22:43Distributed Systems Are The Norm
- 23:10System Interdependencies.
related talks
47mDeploying Like a Boss: From 2 Servers to 20,000
43mContinuously deploying a distributed monolith
52mDevops @Scale
26mHuman Trust in Continuos Deployment
47mContinuous Delivery Warstories
42mHow Houghton Mifflin Harcourt Went from Months to Minutes for Infrastructure Delivery
44mHype Driven Development
35mLiving in a Hybrid World