Backend Response Time Degradation Hidden by Averages
warninglatencyUpdated Feb 6, 2026
Average response time metrics (haproxy_backend_response_time_average) mask tail latency issues, where p99 requests experience significantly worse performance than the mean.
Sources
How to detect:
Compare haproxy_backend_response_time_average against detailed timing breakdowns (Tq, Tw, Tc, Tr, Ta from logs). High Tr (server response wait time) with low Tc (connection time) indicates backend application slowness rather than network or queueing issues.
Recommended action:
Parse HAProxy access logs to extract timing distribution data beyond simple averages. Use log-based metrics to track p95/p99 latencies. Investigate backend application profiling when Tr dominates total response time. Consider implementing request sampling or distributed tracing for detailed latency analysis.