- Initial Average Latency: ~3.5 seconds per request (100 connections).
- Problem Identified:
A CPU-intensive calculation (synchronous, CPU-bound task) was running on the main thread, blocking the Event Loop and delaying all incoming requests.
- The flame graph showed that the heavy calculation function consumed most of the CPU time.
- This confirmed that the Event Loop was blocked, preventing I/O and network tasks from being processed efficiently.
-
Cause of Event Loop Blockage:
The blocking occurred because the calculation was executed directly on the main thread, which is single-threaded by default in Node.js. -
Chosen Solution – Worker Threads:
- Offloaded the heavy computation into Worker Threads, which run in isolated threads.
- Worker Threads were chosen instead of Clustering because:
- Workers share memory more efficiently.
- Ideal for CPU-heavy calculations.
- Keeps the main process light while handling concurrent requests.
-
Communication Strategy:
- The main thread sends tasks to the worker via
postMessage. - The worker executes the heavy function and responds with the result via
parentPort. - This ensures the main thread remains non-blocking and responsive to requests.
- The main thread sends tasks to the worker via
| Metric | Value (Optimized) |
|---|---|
| Latency (Avg) | 54.71 ms |
| P50 (Median) | 44 ms |
| P97.5 | 128 ms |
| Max Latency | 385 ms |
| Requests/sec | ~1,813 |
| Bytes/sec | ~462 KB |
- Before: ~3.5 seconds latency.
- After: ~54.71 ms latency.
- Latency Improvement: ~98.4% faster 🚀
- Main thread now shows minimal blocking.
- Heavy calculation time moved into worker threads, freeing the Event Loop.
By offloading the CPU-bound workload to Worker Threads, the Event Loop is no longer blocked.
- Latency dropped by ~98%.
- Throughput increased significantly (handling thousands of requests concurrently).
- The system is now responsive, scalable, and production-ready.
