test: make Nx own test environments - #675
Conversation
Route pgflow test startup, cleanup, focused slow checks, and interruption recovery through Nx so deliveries reuse one owned environment without bypassing setup dependencies. Add endpoint-specific E2E readiness, integration database config guards, pidfd-safe cleanup, and complete cache inputs. Keep E2E at full-suite scope because its workers and database state are shared.
|
|
View your CI Pipeline Execution ↗ for commit 3d5d5f0
💡 Verify your cache is correct by running tasks in a sandbox. Read docs ↗ ☁️ Nx Cloud last updated this comment at |
Serialize shared Supabase and integration resources through Nx-owned locks, validate focused test selectors before execution, and make E2E startup and cleanup track live process and container state. The first lifecycle pass could reuse stale database state, trust shell-expanded selectors, and certify or leak failed E2E servers. This correction records applied fingerprints only after readiness, declares actual cache inputs, and adds deterministic lifecycle regression checks.
🔍 Preview Deployment: Website✅ Deployment successful! 🔗 Preview URL: https://pr-675.pgflow.pages.dev 📝 Details:
_Last updated: _ |
jumski
left a comment
jumski
left a comment
There was a problem hiding this comment.
i feel like this is very complicated setup - maybe this complexity is required, maybe its overneigneered? we need to understand exacly. need to invoke ponytail for help
scripts/ensure-migrations.sh: wow why is it needed? why its so complicated?
scripts/functions-server.sh: why is it needed and why its so complex? its concerning tbh
scripts/require-test-file.sh: why is it needed?
scripts/run-e2e.sh: why is it needed?
scripts/test-lifecycle.sh: why is it needed?
scripts/with-integration-lock.sh: why is it needed?
tools/nx-executors/focused-file.cjs: how this works in simple terms?
| declare -a stack_lock_fds=() | ||
|
|
||
| # Nx-owned lifecycle targets this cleanup may stop. The exact workspace, | ||
| # project, and target come from Nx task environment variables, not argv. |
There was a problem hiding this comment.
this looks like something that will be hard to maintain - would it be possible to use some kind of tag system built into nx and mark the targets via this tag system in each project.json then retrieve all tagged targets here? would it be a good idea? ponytail?
| pnpm nx test-env:fresh edge-worker | ||
| ``` | ||
|
|
||
| The target takes the environment lock first. It then previews and locks every pgflow stack resource, the integration database, and the core upgrade fixture before it stops or removes anything. It stops only Nx-owned lifecycle processes from this workspace. If another worktree still holds a resource lock, recovery stops before it changes that resource. Locks live in the repository Git common directory, so sibling worktrees and sandboxes share them. |
There was a problem hiding this comment.
damn... is it done so only one worktree can run at the given time? this could be easily solved via supabase/config.toml customizable env-based port specifiers, so each new worktree gets a new .env file with specified ports for all services, guaranteed to not conflict with others... right?
🚀 Production Deployment: Website✅ Successfully deployed to production! 🔗 Production URL: https://pgflow.dev 📝 Details:
Deployed at: 2026-09-09T14:11:38+02:00 |
1 participant
Summary
Why
Direct test-runner commands bypassed Nx setup and caching, while interrupted runs could leave stale processes or shared infrastructure. The prior lifecycle also allowed stale migrations, shell-expanded selectors, false readiness, and cross-worktree resource races.
The updated lifecycle keeps setup, recovery, cache validity, and live-resource ownership inside Nx. Full E2E suites remain intact because their workers and database state are shared.
Local checks
Passed on commit
3d5d5f01:Local edge-worker integration did not complete successfully. The full target timed out, and its focused reproduction failed one unchanged
performanceMapFlow.test.ts15-second completion bound while two tests passed. The change does not touch worker, schema, migration, integration-helper, or performance-test behavior. CI's isolated integration job will add evidence about whether this is local load or a reproducible regression.Closes #673