Measuring tail latency in PHP without losing your weekend

Average response time is a comforting lie. Your slowest 1% of requests is what kills user trust — and on the right SaaS dashboard it's often a 10x multiplier on your reported median.

This article walks through three production-grade patterns: middleware-level histograms, Horizon job-level metrics, and outbound HTTP client wrappers that track per-host tail latency.

None of these require an APM vendor. All three emit Prometheus-compatible histograms via a tiny in-process buffer, flushed by a queue worker. Total overhead: under 0.2ms per measurement.

Pair this with a Grafana panel that buckets by route and status, and you have most of what a $400/month APM gives you, at zero recurring cost.