![Terraform
Build infrastructure
across multiple
regions
One set of templates
Variables for
differences between
environments
module "sso-instance-template" {
source = "git::ssh://git@foo/tfmodules/instance-tmpl.git"
machine_type = "${var.sso_machine_type}"
preemptible = "${var.preemptible}"
envname = "${var.envname}"
service = "sso"
domain = "${var.envname}.${var.domain}"
fw_tags = ["global", "sso", "gluster-client"]
needs_nat = "natted"
startup_script =
"gs://${var.bootstrap_bucket}/lush-sso-bootstrap.sh"
disk_image = "${var.disk_image}"
gcp_region = "${var.gcp_region}"
}
module "sso-group-manager" {
source =
"git::ssh://git@foo/tfmodules/tf-gce-instance-group-manager.git"
service = "sso"
instance_template = "${module.sso-instance-template.link-subnets}"
}](https://image.slidesharecdn.com/lushnextlondontechsession-170507191549/85/Lush-E-Commerce-Migration-Google-NEXT-London-20-320.jpg)
More Related Content
What's hot (8)
Similar to Lush E-Commerce Migration - Google NEXT London (20)
Recently uploaded (20)
Lush E-Commerce Migration - Google NEXT London
- 1. Lush E-Commerce Migration Google Cloud in a crazy timescale
- 2. Why did we do this? What did we do? How we did it What were the benefits? Whats next? Agenda
- 5. Why Move? Not aligned to Lushs Ethics Blackbox / Closed Solution No Flexibility Inside the Platform Poor Support Limited Capabilities Excessive Costs Poor Billing Model Vendor Lock In
- 6. Why GCP? Future Proofing Lushs Technology Stack Aligns to Lushs Ethical Approach No Vendor Lock-in (Open Cloud) Full Control over Infrastructure Fast Network and Robust Backbone Per minute billing Sustained Use Discounts Preemptible VM Instances Brilliant Support
- 7. Timescales November December 11th November Project kicks off 18th November Contract signed 17th - 20th December Go live period 22nd December End of service from previous supplier
- 9. Warehouse SOLSOL SML QMS OS OSA SOLR SOLR WEB SOLR UK SITE Warehouse SML QMS SOL OS OSA SOLR SOLR WEB SOLR UK SITE NEW COUNTRY CUSTOMER SERVICES SINGLE SIGN ON CONTENT REPO
- 10. Lift and shift + select managed services Compute Engine Cloud SQL 2nd generation Cloud Load Balancing Google Stackdriver
- 11. Memcache Compute Engine Memcache Compute Engine Memcache Compute Engine Varnish Compute Engine Memcache Compute Engine Cloud SQL Cloud SQL Cloud SQL Cloud SQL Solr (Search) Compute Engine Frontend Compute Engine Content Repo Compute Engine Order Store Compute Engine SSO Compute Engine Varnish Compute Engine Cloud Load Balancing
- 12. europe-west1 UK and EMEA asia-east1 India, HK, JP, and APAC us-central1 US and LATAM
- 13. Production Infrastructure 3 Locations, 3 Regions, 9 Zones 175+ Instances (Autoscaler) 36 Cloud SQL 2nd generation Instances More of the same for non-production, i.e. development and staging
- 14. How we did it Sam Bashton
- 15. Philosophy
- 16. MVC: Minimum Viable Cloud Any machine can be terminated at any time and will self heal Infrastructure active/active in at least two zones Infrastructure defined as code Use drop-in replacement services where available Automated deployments
- 17. Infrastructure as Code Modern sysadmin (devops): 60% dev 40% ops Dont build infrastructure - build templates that build infrastructure
- 18. Infrastructure as Code Pre-built building block templates - Terraform modules 80/20 rule - 80% of the work is standard Allows us to concentrate of the interesting 20% Massively reduces time to deliver
- 19. Reducing Repetition Templates heavily re-used: Multiple environments Dev Stage Production Multiple Regions APAC EMEA US
- 20. Terraform Build infrastructure across multiple regions One set of templates Variables for differences between environments module "sso-instance-template" { source = "git::ssh://git@foo/tfmodules/instance-tmpl.git" machine_type = "${var.sso_machine_type}" preemptible = "${var.preemptible}" envname = "${var.envname}" service = "sso" domain = "${var.envname}.${var.domain}" fw_tags = ["global", "sso", "gluster-client"] needs_nat = "natted" startup_script = "gs://${var.bootstrap_bucket}/lush-sso-bootstrap.sh" disk_image = "${var.disk_image}" gcp_region = "${var.gcp_region}" } module "sso-group-manager" { source = "git::ssh://git@foo/tfmodules/tf-gce-instance-group-manager.git" service = "sso" instance_template = "${module.sso-instance-template.link-subnets}" }
- 21. Production Infrastructure Code 3 Locations, 3 Regions, 9 Zones 175+ Instances (Autoscaler) 36 Cloud SQL 2nd generation Instances < 2200 lines of Terraform code
- 22. Ready Meal Images Part-baked images created with Packer Boot time configuration with Puppet
- 23. Autoscaling
- 24. Production Stage Development Github Branch: develop Jenkins .deb build Multiple Regions Jenkins .deb install GCS Bucket Deb Repository Multiple Regions Jenkins .deb install Multiple Regions Jenkins .deb install GCS Bucket Deb Repository GCS Bucket Deb Repository Jenkins .deb promote Github Branch: stable Or master Jenkins .deb build Deployments
- 25. Logging and Monitoring Stackdriver Agent installed on every instances, exports logs and metrics
- 26. Difficulties Faced Poor Terraform support for Cloud SQL 2nd Generation Yay for Open Source - we could fix this ourselves (PRs awaiting merge) Old PaaS used MySQL - but no mechanism for database replication available Required a full dump and restore, meaning multi-hour downtime
- 27. Money Saving Autoscaling ensures we only run the infrastructure necessary for the current levels of traffic Switching dev/staging to Preemptible VMs has saved 贈贈贈 Right-sizing recommendations saved more 贈贈 40% saving
- 28. Key Improvements Global network Simplified Infrastructure Cheaper! Autoscaling
- 29. Whats Next? Ryan Kerry, Lush
- 30. Whats Next? Now: Service Orientated Architecture Transformation Containerisation of Application (GKE) Data Warehouse (Big Query & Data Proc and Dataflow) Cloud Native Components (Pub/Sub & Dataflow). Future: No-ops approach - Cloud Functions. Machine Learning and AI Capabilities. Speech, Vision and Video APIs.
- 31. Thank you
- 32. Lush E-Commerce Migration Google Cloud in a crazy timescale