PVBLIC Foundation
|
cf3635ba25
|
Update mistral.py
1. Intelligent Error Handling
Added _is_retryable_error() method to distinguish retryable vs non-retryable errors
Prevents unnecessary retries on client errors (4xx) that won't succeed
Caps retry delay at 30 seconds to prevent excessive waiting
2. Optimized Timeout Configuration
Upload: Capped at 2 minutes (was using full 5-minute timeout)
URL requests: 30 seconds (should be fast)
OCR processing: Full timeout (can take time)
Cleanup: 30 seconds (should be quick)
3. Enhanced Connection Pool
Increased connection limits: 20 total, 10 per host
Longer DNS cache TTL (10 minutes vs 5 minutes)
Increased keepalive timeout (60s vs 30s)
Added async DNS resolver for better performance
Granular timeout controls (connect, read, total)
4. Concurrency Control for Batch Processing
Added semaphore-based concurrency control (default: 5 concurrent)
Prevents API overwhelming while maintaining throughput
Configurable concurrency limit per workload
5. Memory Efficient Result Processing
Early exit for empty content validation
Better error metadata for debugging
Added content length tracking
Streamlined page processing logic
6. General Performance Improvements
Better error logging with truncated responses
Optimized metadata creation
Improved debug logging efficiency
|
2025-05-30 20:06:29 -07:00 |
|
