Work in progress · Version 1.1, last updated 6 November 2026
Governance

Architecture decisions

The significant decisions taken so far, newest first. Each will be recorded as an architecture decision record in its repository.

Decision log

Date Decision Alternatives considered Reason
5 Nov 2026 Telemetry runs in production only; Datadog is the single observability tool and Sentry is removed Sentry alongside Datadog One tool and one bill; no telemetry cost for non-production
5 Nov 2026 Finder production logs are indexed lean (errors, 5xx, slow requests and a 5% sample, kept 30 days) and archived in full to S3 for a year Index everything; Cloudflare log export Cost
4 Nov 2026 Finder production alerts are raised in Datadog CloudWatch alarms through chat One alerting surface
4 Nov 2026 Start finder production with one of each component and scale to the specified size before load tests and launch Launch at full size Lower running cost until traffic arrives
4 Nov 2026 Finder production reads its own PADI Travel change stream Share the staging stream A production reload never blocks staging
3 Nov 2026 Finder compute on EC2 autoscaling groups Containers on Fargate (in the original specification) Same image, tooling and monitoring agent as development and staging
3 Nov 2026 Search queries authenticate with IAM SigV4 Shared passwords No shared secrets; deploy smoke test proves a refused signature fails
3 Nov 2026 Monitoring defined as code in Terraform Datadog CloudFormation types The CloudFormation types for dashboards and SLOs are no longer maintained
3 Nov 2026 One NAT gateway and an S3 gateway endpoint only A NAT gateway per zone; interface endpoints Cost; zone risk accepted and documented
30 Oct 2026 Performance changes must leave pages identical; riskier ones ship behind switches that are off by default Optimise freely No regressions before launch
29 Oct 2026 DAM data stays cached in memcache, with memcache sized up Shorter DAM cache lifetime Evictions, not the lifetime, caused the slowness
29 Oct 2026 The Cloudflare zone's rules, Workers and DNS records are managed in Terraform with a nightly drift check Dashboard changes Audit trail and no silent drift
Oct 2026 URLs carry country and language; country, region and language are independent; three content regions Language-only URLs Cacheable pages with market-specific content
Oct 2026 Finder text is machine-translated only Human review PADI decision
19 Oct 2026 Staging deploys need no approval (edge Workers excepted); production does Approval on staging Delivery speed with a production safeguard
18 Oct 2026 Per-hostname client certificates for origin pulls Cloudflare's shared certificate The shared certificate can be presented by any Cloudflare customer
16 Oct 2026 Staging carries production PADI Travel data through its own stream Staging data Realistic testing
Sep 2026 All systems deploy through AWS CodePipeline with one shared library GitHub Actions to EC2 (retired); Acquia Pipelines One delivery platform for web, edge and finder
Jun–Sep 2026 Frontend on Acquia Front End Hosting Advanced Vercel; AWS Amplify (supported Next.js 15 only) Single vendor with the Drupal hosting; contract in place