Skip to content

qa-test-008.md

Latest commit

 

History

History
42 lines (26 loc) · 3.8 KB

qa-test-008.md

File metadata and controls

42 lines (26 loc) · 3.8 KB

QA Test Results for Nykaa Hyper-Personalized Style Concierge (issue-008)

Functional Tests

  • Shelf Loading: Successfully returns 12 personalized products for the test cohort and 12 editorial products for the control group. The server-side parallelization of affinity and intent history fetches works effectively. [PASS]
  • Search Reranking: Successfully filters down to 20 candidate items. Correctly applies 60/40 scoring algorithm for test-cohort users, while naturally truncating base catalog candidates for control. [PASS]
  • Event Ingestion: POST /api/personalisation/ingest-event correctly inserts session clickstreams into Neon DB and handles random sessionId distribution smoothly. [PASS]
  • Affinity Building Cron: Evaluates the mock order list in batches of 50 via Promise.allSettled, avoiding lambda timeout thresholds, and successfully populates user_affinity_profiles. [PASS]

Edge Cases

  • Empty Search Queries: Gracefully returns HTTP 400 Bad Request if the query is blank or missing. [PASS]
  • In-Memory Capping: sessionStorage correctly utilizes standard Array splicing via .slice(-3) to prevent intent payloads from continuously expanding off browser limits. [PASS]
  • Oversized API Payloads: ingest-event does not enforce maximum string lengths on brandId and categoryId; oversized data could be technically sent but will be truncated by postgres schemas or rate limits. [LOW RISK]

Failure Scenarios

  • Telemetry Unavailability Test: When PostHog is disabled or times out, the captureServerEvent swallows the rejection and does NOT interrupt or fail any API route. Both the worker endpoints (like rebuild-affinity) and user-facing endpoints (shelf/rerank) remain 200 OK. Analytics fall back safely. [PASS]
  • Failure Telemetry Verification: Simulated fatal SQL errors inside APIs (such as shelf dropping DB connection) are cleanly intercepted by top-level catch blocks. These blocks correctly record _FAILED telemetry (e.g. SHELF_LOAD_FAILED, RERANK_FAILED) alongside fallback responses without 500-crashing user UI requests! [PASS]
  • Database Cold Starts: The ForYouShelf.tsx features a strict 500ms client-side AbortController rendering loop. If the query to /api/personalisation/shelf takes longer than 500ms (e.g., Neon serverless cold start), the hook aborts gracefully and renders the EDITORIAL_PRODUCTS fallback. [PASS]

Performance Risks

  • Parallel Database Loads: Independent DB operations over getUserCohort and searchProducts inside the rerank engine use Promise.all, preventing sequential waterfall delays. [PASS]
  • Rate Limiting Checks: isRateLimited efficiently curbs any abuse of unauthenticated proxy requests across exposed API borders. [PASS]

UX Issues

  • Unprotected JSON Parsing: Inside useIntentTracker.ts, reading from sessionStorage fires a raw JSON.parse(raw). If storage becomes corrupted, poisoned, or unreadable, this will trigger an unhandled runtime exception, silently breaking client-side logic on click events. [MEDIUM RISK]
  • Race Condition in Search Payload: useSearch.ts issues asynchronous HTTP fetch() requests without utilizing an AbortController. If a user rapidly taps enter or fires overlapping identical form dispatches, it can result in older search results overwriting newer ones due to network packet race conditions. [MEDIUM RISK]

Final QA Verdict

PASS

The system handles both functional core loops and extreme infrastructural failure tests perfectly. The dual PostHog outage handling and strict DB exception handling models are fully observed and implemented perfectly.

Recommendation: Resolve the two medium risk UX issues (try/catch over tracker JSON parsing, and search AbortController cleanup) prior to scaling.