Spotify's journey to GCP

Transcript

1. Spotify’s journey to GCP
2. ~# whoami Alexey Lapitsky / @vzctl alexey@spotify.com Site Reliability Engineer @Spotify PM for the service track of the GCP migration
3. Agenda • History • Why GCP? • The biggest mistake • Learnings • Questions!
4. Spotify’s cloud history 2010 Manual VM provisioning 2012 Cloudstack, reached ~700 instances 2013 Amazon EC2 2015 Google Cloud Platform
5. Current status • 4 data centers, ~10k physical servers • Micro-service architecture, ~1k different services • Mostly commodity hardware (Dell) • Fully automated, self-service server provisioning Moving everything to GCP
6. Focus
7. Why GCP?
8. Why GCP? • BigData • Culture • Timing • Partnership
9. Our worst mistake
10. Spoticloud CLI REST DB Cloudstack AWS Puppet DNS Server • CLI Tool + REST Server • User authentication with ssh agent: github.com/spotify/crtauth SSH Jumphost
11. Looks awesome! • No need to distribute AWS accounts • No need to manage ACLs • Extremely easy to use ~# spcloud create --name web --project alexey
12. … or not really :( • Testing environment, shamelessly misused for production • No accountability or a proper ownership tracking ◦ 60% of instances were wasting money • Access only to the basic cloud primitives ◦ Which are already available on-premise Currently being deprecated.
13. GCP Migration
14. Cross-project networking (XPN)
16. Things we learned • VPN is tricky • GCP network is great • Keep an eye on quotas • Live VM migration! • Think about ACL mgmt in advance • BigTable, GCS, BigQuery are awesome
17. Questions?