Spike/base site approach a - #21853
Conversation
Resolve SSR baseSiteId in plain Node, no Angular. Cacheless by design: every resolve() fetches OCC base sites fresh and matches the request URL against each site's urlPatterns, falling back to the app-configured default (context.baseSite[0]) on no match. No cache because SSR runs as multiple instances that die/restart independently, so a per-process cache would drift between nodes and yield inconsistent results across the fleet; the OCC call is cheap, so consistency beats the saved call. Reliability (spec factor #1 - never hang/DoS the SSR process): - concurrency cap (default 10) sheds load fast, throwing ConcurrencyLimitError before touching OCC; - AbortController timeout (default 3000ms) throws OccUnavailableError. Framework provides createBaseSiteRequestHandler(): it resolves the baseSiteId from a trust-proxy-aware request URL and maps both typed errors to 503 + Retry-After. The app wires only the route and a render callback; the framework carries no HTTP or site knowledge.
- use protected instead of private for resolver fields and methods - rename misleading _config constructor param to config - apply contentType to the 503 error response in the request handler - drop stale "switch approaches" note from the resolver contract
🚨 PR Title Validation Failed 🚨Your pull request title does not follow the required format. Please update it to match the expected pattern: Expected format: Allowed Types
Example of a valid PR title✅ ❌ Merge is blocked until the PR title is corrected. |
CX_BASE_URL is not a Node runtime env var on CCv2; read OCC base URL from build-time environment.occBaseUrl first, then occ-backend-base-url meta tag in browser/index.csr.html (substituted by the deploy script). Missing URL degrades gracefully — warning logged, AI-SEO handlers disabled, core SSR unaffected (previously crashed on startup). Resolver gains 60 s TTL cache with initPromise dedup to avoid thundering-herd on cache miss; concurrency cap guards the miss path.
🚨 PR Title Validation Failed 🚨Your pull request title does not follow the required format. Please update it to match the expected pattern: Expected format: Allowed Types
Example of a valid PR title✅ ❌ Merge is blocked until the PR title is corrected. |
There was a problem hiding this comment.
for all these files, since the concern is base-site and not site-context at large, I recommend the next directory to be renamed to base-site instead of site-context
| err instanceof ConcurrencyLimitError || | ||
| err instanceof OccUnavailableError |
There was a problem hiding this comment.
minor comment: since this is a spike
We are defaulting to 5 second retry for both concurrency limit error and occ unavailable.
I don't they they should share the retry value.
for ConcurrencyLimitError, I would suggest 2-3 seconds
for OccUnavailableError, I would suggest 20-30 seconds
2 participants
No description provided.