Throughput (req/sec)
--Response Time (ms)
--Error Rate (%)
HTTP Status Distribution
Active VUs
API Endpoint Performance
6 endpoints| Endpoint | Avg (ms) | P95 (ms) | Error Rate | Status |
|---|---|---|---|---|
| Start a test to see endpoint metrics | ||||
AI-Generated Recommendations for HBB/Kaltura
Based on 10K VU Test ResultsAPI Gateway Returning HTML Instead of JSON Under Load
Issue: 29% of requests (206,342) received HTML error pages instead of JSON responses. Kaltura's API gateway/load balancer is returning default error pages when overwhelmed.
- Increase API gateway connection pool and timeout limits
- Configure proper JSON error responses (not HTML 502/503 pages)
- Add circuit breaker pattern to prevent cascade failures
- Review Nginx/HAProxy backend configuration for max connections
Request Timeouts Exceeding 10 Seconds
Issue: 16.8% of requests (119,411) timed out after 10s. Most affected: asset/action/list, categorytree/action/get, anonymouslogin.
- Scale Kaltura OTT backend horizontally (add more API nodes)
- Enable auto-scaling on the Kaltura cluster for Partner ID 5093
- Optimize database queries for asset listing and category tree endpoints
- Implement response caching for anonymous content (CDN-cacheable)
Anonymous Login Bottleneck
Issue: ottuser/action/anonymouslogin is called before every content request. Under 10K VUs, this endpoint becomes the bottleneck, causing downstream failures.
- Implement KS token caching (anonymous tokens should be reusable for 30-60 min)
- Add a token pool/pre-generation mechanism to avoid per-request login
- Consider a lightweight token validation instead of full login per session
No Rate Limiting / Graceful Degradation
Issue: Instead of rate-limiting and serving cached content, Kaltura returns hard failures. Users see blank pages or broken UI.
- Implement API rate limiting per partner (5093) with 429 responses
- Add stale-while-revalidate caching for content APIs
- Configure CDN edge caching for GET-able content (category tree, asset lists)
- Implement graceful degradation: serve cached content when backend is overloaded
Load Distribution — Now Distributed Across India
Update: All future tests will use distributed load from 5 unique Indian IPs across Mumbai (ap-south-1) and Hyderabad (ap-south-2) regions, routed via CDN to Kaltura backend. This eliminates single-IP rate limiting concerns.
- What is the provisioned API capacity for Partner 5093?
- Are there IP-based or partner-level rate limits?
- What CDN edge nodes serve Indian traffic for VZY?
Capacity Planning Gap
Issue: With 10K concurrent users and 45.9% error rate, the platform cannot support the projected 50K-100K launch-day users.
- HBB/Kaltura must provision capacity for minimum 50K concurrent API calls
- Conduct progressive load tests: 1K → 5K → 10K → 25K → 50K after each fix
- Establish SLA targets: P95 latency < 500ms, error rate < 1% at 50K VUs
- Setup real-time alerting for error rate > 5% in production