Skip to main content

This is for internal use by the PaaS team. Public-facing documentation is located at docs.cloud.service.gov.uk.

Production Ireland Decommissioning

Introduction

The purpose of this document is to provide a guide for decommissioning the production environment in Ireland.

This is only a guide, the process has not been tested outside of a development environment. There are likely steps missing and the process may need to be adapted as you go.

Warning

  • The production ireland environment is deployed into the same aws account as the production london environment. We only want to remove resources from the ireland region. Cross region resources should be left. As such do not try to decommission the paas-aws-account-wide-terraform project. This is account wide.
  • Logit and Aiven prod are shared between London and Ireland, so there is no need to decommission for just Ireland.

When “Notify” is decommission

(Optional) When notify have decommissioned their environments we can remove their isolation segments. This can provide instant cost savings. Change the number_of_cells in the following two files to 0:

Pre-checks

Before starting the decommissioning process, ensure that the following checks have been completed:

  • [ ] Ensure all tenants have been migrated off the environment.
  • [ ] Ensure all final bills have been sent. Decommissioning the environment will stop the billing process.
  • [ ] Ensure logit graphs do not show any traffic to the environment other than the normal platform traffic.
  • [ ] Ensure all user applications have been removed or stopped.
  • [ ] Ensure all user services have been removed from the environment. PaaS services will be removed as part of the decommissioning process. The pipeline script will try to remove all services, however it may fail if a service is not ready for removal (For example, if an s3 bucket is not empty).

Before decommissioning

Before decommissioning the environment, ensure that the following steps have been completed:

  • [ ] Take a pg_dump of the billing database and store somewhere safe. We will have a final rds snapshot as well, but it is good to have a backup in case we have post-decommissioning billing queries.
  • [ ] Take a pg_dump of the audit database and store somewhere safe. This contains all the cf events since the auditor was deployed.
  • [ ] Decommission london monitoring of ireland production.
    • Remove the following symlink. Merge and deploy.
    • Manually delete the pipeline from concourse:
      • fly login -c https://deployer.london.cloud.service.gov.uk/ -t prod-lon
      • fly -t prod-lon destroy-pipeline --pipeline monitor-prod
  • [ ] Remove all peers from the terrraform config. Set to [] to ensure removal. Merge and deploy.

Actual Decommissioning

  • [ ] Extract pingdom credentials from paas-credentials. Log into pingdom and remove the ireland checks.
  • [ ] Remove protection for the production environment. Merge to main.
  • [ ] Add ‘$(eval export ENABLE_DESTROY=true)’ to prod section in paas-cf Makefile. Merge to main.
  • [ ] Announce on #cyber-security-notifications (slack) using ‘Action Notification’ your intention to decommission the environment. Getting team member approval.
  • [ ] Run gds aws paas-prod-admin -- make prod pipelines to push the destroy pipeline to concourse.
  • [ ] Start the ‘destroy-cloudfoundry’ pipeline from concourse
    • Note: It is likely the terraform destroy might fail on s3 buckets. They may need to be manually emptied and concourse job re-run. It is also an option to add force_destroy to the terraform if it is missing.

= DO NOT CONTINUE UNTIL THE DESTROY PIPELINE HAS COMPLETED SUCCESSFULLY =

  • [ ] Add $(eval export ENABLE_DESTROY=true) to prod section in paas-bootstrap Makefile. Merge to main.
  • [ ] Spin up a production ireland vagrant vm with gds aws paas-prod-admin -- make prod deployer-concourse bootstrap
  • [ ] Start destroy-bosh-concourse pipeline from the vagrant machine concourse. Ensure this runs to completion successfully.
  • [ ] Remove vagrant vm with gds aws paas-prod-admin -- make prod deployer-concourse bootstrap-destroy

Post Decommissioning Checks

  • [ ] Click around AWS console and enable the resource explorer in Ireland to look for orphaned items. Check:
    • [ ] ec2
    • [ ] ebs
    • [ ] ebs snapshots
    • [ ] elbs
    • [ ] cloudfront (remember this is global, it won’t be empty so check with care)
    • [ ] s3 (remember is this global, check with care). Prod state bucket will still probably be there and can now be removed.
    • [ ] rds
    • [ ] rds snapshots (we expect to still have snapshots)
    • [ ] sqs
    • [ ] eips
    • [ ] amis (Bosh might have left a few amis). Clean up.
    • [ ] cloudwatch
    • [ ] elasticache redis caches
  • [ ] Check aiven project “paas-cf-prod”. Ensure we don’t have any services that start “prod-”. They should start “prod-lon-”. Any “prod-” are likely orphaned instances. Check carefully and manually remove.

  • [ ] Check AWS billing in the following days to see we aren’t being charged for anything unexpected in the Ireland region.

  • [ ] Database snapshots will no longer automatically time out. We will need to remove them all once we are happy we don’t need them. Leave for a few weeks to be sure before cleanup.

  • [ ] Create ticket to evaluate current logit scaling. We have probably reduced logging volume by half.

Post Decommissioning Clean Up

  • [ ] Update documentation to remove references to the Ireland production environment. This includes:
  • [ ] Remove environment config in paas-credentials.
  • [ ] Remove references to prod environment (not the prod account) in paas-cf and paas-bootstrap. We have a ‘prod’ and ‘prod-lon’ environment that both use the ‘prod’ aws account.
  • [ ] Remove alerting for ireland in Pagerduty