b768441c13aed1c978522f59d2cc326f3bd5a708
4 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
|
|
82c565f7dc |
perf(website): cache prerendered pages in CF Cache API (withRegionalCache) (#20905)
## Summary
Recovers most of the TTFB the EKS→Cloudflare migration lost on
`twenty.com`. OpenStatus's P50 chart shows the regression clearly: TTFB
went from ~50–80ms (pre-migration, CF edge cache HIT) to ~250–350ms
(post-migration, every request hits Worker → R2 → respond).
## Why the existing Cache Rule stopped working
The zone-level `Twenty Website - Aggressive cache` Cache Rule was
correctly configured and was the reason pre-migration TTFB was low. It
still exists, still has `cache: true`, Edge TTL 1d. But it doesn't apply
to Worker responses on a Worker custom domain:
- **Pre-migration** request flow: `edge → Cache Rule lookup → HIT
(~20ms) / MISS → origin → cache the response`
- **Post-migration**: `edge → Worker runs first (custom domain) → Worker
generates synthetic response from R2 → return`
Cache Rules cache responses obtained via `fetch()` from the Worker, not
synthetic responses constructed inside the Worker. OpenNext for SSG
pages reads prerendered HTML from R2 and returns it — that's synthetic.
So the rule has no insertion point.
This is structural to how CF Workers handle custom domains; not a
misconfiguration on your side.
## The fix
`open-next.config.ts`:
```ts
const incrementalCache = withRegionalCache(r2IncrementalCache, {
mode: 'long-lived',
});
const baseConfig = defineCloudflareConfig({ incrementalCache });
```
OpenNext-native wrapper. The Worker still runs per request (~5–20ms
execution), but the ISR cache lookup goes through CF's per-region Cache
API (~5–20ms) instead of R2 (~50–150ms). For pages whose prerender
doesn't change between requests, that's the bulk of the TTFB recovered.
## Measured impact (live before/after on twenty.com today)
| URL | Before (avg of 3) | After cold (first 2 hits/region) | After
warm |
|---|---|---|---|
| `/` | 322ms | 600–640ms | **110–125ms** |
| `/pricing` | 267ms | 630–690ms | **104–110ms** |
| `/why-twenty` | 250ms | 175–270ms | **100–175ms** |
First 1–2 hits per CF region after this deploys will be slower than
baseline (regional Cache API populating from R2), then it sustains.
Steady state is significantly better than pre-fix.
## What this doesn't recover
Pre-migration `cf-cache-status: HIT` was ~20–30ms because the Worker
wasn't invoked at all. We can't get there without either:
- Moving SSG hosting off the Worker (back to a static origin Cache Rules
would cover)
- OpenNext gaining a "publish responses to caches.default" mode (doesn't
exist today)
Realistic-best on CF Workers + OpenNext is around the ~80–130ms range
we're now seeing.
## Live state
Already deployed to both prod (Version `40dfaa1a-...`) and dev (Version
`b45cc2de-...`) ahead of opening this PR, so the OpenStatus chart should
start improving immediately. This PR makes `main` reflect the change.
|
||
|
|
a2acf88a57 |
feat(website): per-PR preview deploys via Worker versions (#20762)
## Summary Adds review apps for the marketing site. Every PR that touches `packages/twenty-website/**` or `packages/twenty-shared/**` gets a per-version Worker preview URL, sticky-commented on the PR, auto-cleaned up when the PR closes. Same Cloudflare machinery skew protection rides on, just used for previews — no extra plan, no extra services. Cleaner than the GitHub-Actions-runner + Cloudflare-tunnel pattern: previews persist for the life of the version, accessible from anywhere, no warm-up. ## Files - **`.github/workflows/website-pr-preview.yaml`** — on PR open/sync/reopen: builds the Worker with a per-PR `DEPLOYMENT_ID`, runs `wrangler versions upload --tag pr-<N>` (no production traffic), sticky-comments the preview URL. Skipped on fork PRs because GitHub doesn't pass secrets to forks anyway. - **`.github/workflows/website-pr-preview-cleanup.yaml`** — on PR close: walks the Worker version list via the CF API, deletes anything tagged `pr-<N>` (with message-based fallback if the annotation key changes), updates the sticky comment. - **`open-next.config.ts`** — `maxNumberOfVersions: 10 → 50` to leave room for PR previews on top of skew protection's prod-version retention. ## How it looks on a PR The bot leaves a sticky comment like: > 🔍 **Website preview** is up at **https://abc12345-twenty-website-dev.twentyhq.workers.dev** > > | | | > |---|---| > | Version | `abc12345-...` | > | Commit | `<sha>` | > | Bindings | shared with the `dev` Worker (R2 cache + secrets) | > > Updates on every push. Auto-deleted when the PR closes. On close it becomes: > 🧹 Website preview for this PR was cleaned up after close. ## Twenty repo credentials already provisioned - `secret CLOUDFLARE_API_TOKEN` — same scoped token the `twenty-infra` workflow uses - `var CLOUDFLARE_ACCOUNT_ID` = `67b2bbe4381006564d2b0aa6ce6177be` - `var CF_PREVIEW_DOMAIN` = `twentyhq` (no `.workers.dev` suffix — OpenNext appends it; [opennextjs-cloudflare#811](https://github.com/opennextjs/opennextjs-cloudflare/issues/811)) ## Known limitations - **Shared dev bindings**: PR previews use the dev Worker's R2 bucket + secrets (Stripe test key, JWT private key). Fine for a read-mostly marketing site; if two simultaneous PRs ever fight over ISR cache state we can prefix R2 keys per-PR later. - **Fork PRs don't get previews**. GitHub Actions doesn't pass `secrets.*` to fork-PR runs (security), and the wrangler upload requires the CF token. To enable forks, would need to switch to `pull_request_target` and gate on a maintainer label — not done here because the security tradeoff isn't worth it for a marketing-site preview. - **Version cap**: 50 versions is the new ceiling, and `maxVersionAgeDays: 14` auto-prunes anything older. Cleanup-on-close should keep us well under in steady state. ## Test plan - [ ] CI on this PR triggers the preview workflow itself; check that the sticky comment appears with a working URL - [ ] Hit the URL, click around — should look like a fresh marketing-site build with this PR's changes - [ ] Close (don't merge) → cleanup workflow should run; sticky comment switches to the "cleaned up" message; the version is gone from `wrangler versions list --name twenty-website-dev` |
||
|
|
00fad657f4 |
feat(website): enable OpenNext skew protection + tune CF cache (#20760)
## Summary
The original goal of this whole migration: **cross-deployment skew is
now handled by OpenNext's per-version routing instead of by users having
to refresh.**
A client holding a stale tab from deployment X requests assets with
`?dpl=X` — the Worker compares X to the current `DEPLOYMENT_ID`, looks
it up in `CF_DEPLOYMENT_MAPPING`, and routes to the matching old Worker
version via its per-version preview URL
(`<old-version>-twenty-website-<env>.twentyhq.workers.dev`). The old
version serves the old assets / RSC payloads / Server Actions
consistently.
**Verified end-to-end on dev**:
| | Marker in HTML |
|---|---|
| Current Worker (`twenty-main.com/`) | `9npeiytir8EPOtW71cqDZ` |
| Stale request (`twenty-main.com/?dpl=<previous-deploy-id>`) |
`B9OC_TNl1vaGcJ5oUUty6` |
| Direct hit on old preview URL | `B9OC_TNl1vaGcJ5oUUty6` ← matches the
skew-routed response |
## Changes
**`open-next.config.ts`** — enable skew protection
```ts
const baseConfig = defineCloudflareConfig({ incrementalCache: r2IncrementalCache });
export default {
...baseConfig,
cloudflare: {
...baseConfig.cloudflare,
skewProtection: {
enabled: true,
maxNumberOfVersions: 10,
maxVersionAgeDays: 14,
},
},
};
```
(`defineCloudflareConfig` doesn't accept `skewProtection` directly — has
to be merged in)
**`next.config.ts`** — `deploymentId: process.env.DEPLOYMENT_ID`. CI
sets `DEPLOYMENT_ID` per-build; Next bakes it into prerendered HTML,
`?dpl=…` on asset URLs, Server Actions, and RSC fetch headers.
**`wrangler.jsonc`**:
- `compatibility_date: 2026-04-15` (was `2025-01-15`; build was warning)
- `assets.run_worker_first: true` — Worker must intercept asset requests
so the skew handler can route stale `/_next/static/*` to the old
version. CF edge cache absorbs hot paths so this isn't a 5×
billable-invocation tax
- `preview_urls: true` — required; skew routes via the per-version
preview URL which only exists when previews are enabled
- Per-env `services: [{ binding: WORKER_SELF_REFERENCE, service:
twenty-website-<env> }]` — OpenNext's recommended setup for
fire-and-forget ISR revalidation
- Per-env `vars`: `CF_WORKER_NAME` + `CF_PREVIEW_DOMAIN` (bare
`twentyhq`, *not* `twentyhq.workers.dev` — OpenNext appends
`.workers.dev` itself, see
[opennextjs-cloudflare#811](https://github.com/opennextjs/opennextjs-cloudflare/issues/811))
- Kept `global_fetch_strictly_public` in compat flags — without it, CF's
optimised intra-account routing self-loops the cross-version fetch and
522s out. With it, the fetch takes the public-Internet path which routes
correctly.
**`public/_headers`** — deleted (with `run_worker_first: true` the
assets pipeline doesn't process it; Next sets the same `Cache-Control:
immutable` on `/_next/static/*` anyway).
## Companion infra PR
https://github.com/twentyhq/twenty-infra/pull/__ — wires the four CF env
vars (`DEPLOYMENT_ID`, `CF_WORKER_NAME`, `CF_PREVIEW_DOMAIN`,
`CF_ACCOUNT_ID`, `CF_WORKERS_SCRIPTS_API_TOKEN`) into the deploy
workflow.
## Known limitation
Skew routing only works for Worker versions deployed AFTER this PR
(older versions don't have `preview_urls: true` and don't have
`DEPLOYMENT_ID` bindings OpenNext can read). Users on tabs older than
the first post-merge deploy still fall through to the current Worker
(same behaviour as today).
OpenNext marks `skewProtection` as **experimental** in their type docs
("might break on minor releases") — worth keeping an eye on.
|
||
|
|
658bdf3e57 |
chore(website): rename twenty-website-new → twenty-website (#20745)
## Summary Follow-up to the Cloudflare/OpenNext migration (#20741). Now that the legacy `twenty-website` package was already removed in #20270, the `-new` suffix on the marketing site package is no longer meaningful. ## What changes - **Directory rename**: `git mv packages/twenty-website-new packages/twenty-website` (1213 files moved, no content change) - **Package + nx config**: `package.json` and `project.json` name fields updated, `sourceRoot` repointed - **Source refs**: `load-local-articles.ts` and `load-local-release-notes.ts` had a hardcoded `'twenty-website-new'` segment in their monorepo-root fallback path; `app/[locale]/releases/page.tsx` had display strings showing where to add content - **External refs**: root `package.json` workspaces, root `CLAUDE.md` / `README.md`, `twenty-sdk` + `create-twenty-app` READMEs, `.vscode/twenty.code-workspace`, `.cursor/rules/changelog-process.mdc`, Crowdin config + the three `website-i18n-*` CI workflows + `ci-website.yaml` - **Docker cleanup**: `packages/twenty-docker/twenty-website-new/Dockerfile` deleted; the two Makefile targets (`prod-website-new-build` / `prod-website-new-run`) that referenced it removed — EKS deploy was retired in the Cloudflare migration - **`yarn.lock`** regenerated against the new workspace path ## What's deliberately not in this PR The dev hostname `website-new.twenty-main.com` in `wrangler.jsonc` stays for now. Migrating it to `website.twenty-main.com` needs coordinated DNS deletion (current CNAME points at the legacy Docusaurus NLB and serves 503s) and removal of the matching legacy `website` Helm chart in `twenty-infra`. Flagged as a separate cleanup. Companion infra PR: https://github.com/twentyhq/twenty-infra/pull/682 (workflow paths + Terraform ECR + docs) ## Test plan - [x] `yarn install --immutable` resolves clean against the new path - [x] `npx nx typecheck twenty-website` passes - [x] `npx nx lint twenty-website` passes - [ ] CI on this PR confirms the same on a fresh checkout - [ ] After merge: trigger `Deploy Website` workflow against `environment=dev` to confirm the renamed working-directory deploys correctly |