DoxLon | Life with kube, containers and microservices
- 1. Life With Kube, Containers and Microservices Richard Marshall, Head of Platform
- 2. Regulated Robo-Advisors • FCA Regulated • Financial Advisors • Automating Financial Advice • Lots of regulations • and compliance • and audits • and risk committees. • And not much traffic (yet) • But really big customers. • We’re Hiring.
- 3. In the beginning • Linux, Apache, Mongo(!), PHP • Development turn-around times taking to long. • Complex Code base. • 15 releases in a year • (most of those were fixing the planned release)
- 4. #MicroKubeTainers • Microservices + Containers + Kubernetes • Engineering team kick-start • Complexity is moved from the app to the platform. • But it’s repeated complexity! • Kube makes it SO easy to deploy, run and scale apps. • Relearning most of the past 10 years
- 5. Production ready?? • Deploying a kube cluster is easy (Kube- up, KOPS) • Kube PoC deployed, declared production-ready (done…) • d’oh! • Kube evolves fast! Be comfortable rebuilding. • Have an infrastructure pipeline ready. (before you tell people you’re ready to go) • Try “Kube the hard way.” • AWS “interesting” selection. Use GCE.
- 6. Engineered Pipeline • Deploying to kube is easy. too easy • Have code defining what is deployed to Kube • Maintain cluster state outside Kube. • Helm is available now. YMMV • Built our own using git, consul+consul-template (and Bash) • Happy to share this code, if you’re interested.
- 7. • Kube re-implements what your I(aaS) is doing. • Networking is re-implemented within Kube (so your firewalls/SGs don’t help much) • Make use of Kube Tools. We’re growing to love Weave (and NetPol) • Learn and Understand what kube is doing. It makes it easy until it’s not working.
- 8. • Security Hygiene, Washing Hands. • Easier with a scheduler. • Host patching a breeze. • Now containers need patching, and tracking, too. • *Solving*: quay.io/Clair/docker Hub. • _Always_ patching.
- 9. •AWS Persistent Storage •IO Burst Credits •High Disk IO Demand •Cascade failures. capacity? •du -s •Node maturity, NVM instrumentation •Slow Pod deployment times •IP Exhaustion •AWS ELB TLS Policies
- 10. Easy to deploy, hard to know what’s happening.
- 11. Declarative means kube only confirms that we asked for a deployment
- 13. • Are we better off? • Platform is way more complex, but interesting • Techies love the stack • Auditors seem pretty pleased. • Code is quicker to develop, Time To Deliver smaller • So yes. #MicroKubeTainers