Reduced GCS Egress for Random Reads
Optimized initial read handling to reduce egress by over 50%. Previously, all initial requests were treated as sequential, triggering 200MB requests. Now, requests with non-zero offsets are treated as random, utilizing a 1MB reader instead. (Related: #4401)
Metrics: Updated Histogram Buckets for Latency Distribution
This change refines the monitoring of GCSFuse performance by adjusting the histogram bucket boundaries for several distribution metrics. The previous configuration was overly granular for very low latencies and lacked sufficient detail for higher-magnitude events. (Related: #4348)
Key Highlights:
- Improved Visibility: The new buckets provide more detailed information at higher magnitudes while reducing granularity for low-value latencies where precision is less critical.
- Metric-Specific Tuning: Boundaries have been segregated to better fit metrics measured in microseconds versus those measured in milliseconds.