[thomasluizon]

Closes thomasluizon/orbit-tickets#1532

Both production API connection strings now always use the managed Render Postgres database. Removing the migration selector prevents a default plan from choosing the deleted Supabase database.

• infra/variables.tf: remove api_database.
• infra/configuration.tf: retain the existing Render Npgsql conversion, map both connection keys to it unconditionally, and remove the obsolete SSM data source. Keeping the existing conversion preserves the deployed Render connection string format.
• infra/README.md: remove the database cutover instructions and obsolete SSM prerequisites; document the Render source and ordered cleanup.

No new external response fields are read. The existing Render connection conversion is unchanged.

Test evidence

• terraform -chdir=infra init -backend=false -input=false: passed without production backend access.
• terraform -chdir=infra fmt -check: passed.
• terraform -chdir=infra validate -no-color: passed.
• dotnet build Orbit.slnx: passed with zero errors; generated OpenAPI remained unchanged.
• node --test infra/google-play-isolation.test.mjs infra/bff-forwarding.test.mjs infra/ses-isolation.test.mjs: 17 passed.
• dotnet test tests/Orbit.Domain.Tests: 716 passed, exit 0.
• dotnet test tests/Orbit.Application.Tests: 5520 passed, exit 0.
• dotnet test tests/Orbit.Infrastructure.Tests: 3541 passed, exit 0.

No existing test covers the production database selector. An unchanged-test baseline and a strengthened-test failure with the defect present were not obtained. No test was added for this configuration removal; formatting and validation check the local configuration, while the production plan below must establish resource equivalence.

Plan proof

Production plan proof is pending the orchestrator, as required by the work order. No production state, secret values, or AWS parameters were accessed by this worker. No SSM parameters were deleted.

In the existing production Terraform execution environment, with the approved ignored infra/local.tfvars and its current image digests, domains and production IP allow list preserved, remove the obsolete api_database assignment and run from the repository root:

terraform -chdir=infra init -input=false
terraform -chdir=infra plan -input=false -var-file=local.tfvars -out=local.tfplan

Expected result: no resource changes. Every environment group, service and database must remain unchanged. Paste the actual plan summary here before deleting the obsolete parameters. This expected result is not an observed production plan.

Manual steps

1. In the existing production Terraform execution environment, remove the obsolete api_database key from ignored infra/local.tfvars. Run the commands under Plan proof against the S3 production backend configured in infra/versions.tf. The actual plan must prove no resource changes and its summary must be recorded in this PR before cleanup. Stop on any proposed resource change.
2. Only after that proof, in AWS Systems Manager > Parameter Store in us-east-2, delete /orbit/production/api/ConnectionStrings__DefaultConnection and /orbit/production/api/ConnectionStrings__SessionConnection. Do not open or retrieve their values. Verify both exact parameter names are absent from the Parameter Store list and record completion in this PR. This cleanup belongs to the orchestrator and remains pending.

[pullfrog]

✅ No new issues found.

Reviewed changes across all three files, including production environment-group wiring and remaining references to the removed selector.

• Render-only database configuration: Both production connection keys now use the unchanged Render Npgsql conversion; the Supabase selector and its SSM data source are removed.
• Cleanup sequencing: Documentation requires a production plan showing no resource changes before deleting the obsolete SSM parameters. That operational proof remains pending, as disclosed in the PR.
• Verification: All 17 targeted infrastructure regression tests passed. Independent Terraform validation was unavailable because terraform version returned bash: line 1: terraform: command not found; live production equivalence was not verified.

  ^{｜ View workflow run ｜ Using gpt-6.1-sol ｜ 𝕏}

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[thomasluizon]

Plan proof (orchestrator)

Production terraform plan against the S3 production backend, run read-only (-lock=false) with the Cloudflare and Render credentials and the ignored infra/local.tfvars (its api_database key removed for this branch), in two trees with identical inputs:

• base 1bd0d265 (this pull request's base): Plan: 7 to add, 14 to change, 0 to destroy.
• head e71f6fc8 (this pull request): Plan: 7 to add, 14 to change, 0 to destroy.

The two plan texts are byte-identical once the refresh lines are removed (diff prints 0 lines), so removing the api_database selector and the SSM data source changes no production resource. The 21 pending changes in both plans are existing drift from other unapplied work (the redesign web service and routing Worker, the BFF forwarding secret, the uploads IAM policy and the env group links that carry them); they are not part of this pull request and stay unapplied here. Both plans set production_redesign_web_digest to the same placeholder digest because the ignored tfvars predates that variable; the redesign service it creates is identical in both.