Latency SLO

The latency SLO tracks whether your HTTP endpoint consistently responds within your target response time. It is only available for http_check monitors on Business plan and above.

What is p95?

The 95th percentile (p95) response time is the threshold below which 95% of checks in the measurement window completed. It is more informative than an average because it represents the experience of users near the slower end of the distribution without being skewed by rare extreme outliers.

For example, a p95 of 320 ms means that 95% of your checks completed in 320 ms or less. The remaining 5% took longer — those are the "tail" requests.

How the latency SLO works

You set a threshold (in milliseconds) and a target percentage. The SLO is met when at least that percentage of checks in the window came in under the threshold. For instance, with a threshold of 500 ms and a target of 95%, the SLO requires that p95 ≤ 500 ms. 

latency_slo = fraction_of_checks_below_threshold ≥ target_pct / 100

Configuring the latency SLO

Open the monitor edit form and scroll to SLO Settings. The latency fields are:

  • Latency threshold (ms) (latency_threshold_ms) — the response time ceiling, in milliseconds. Checks below this value are counted as "fast."
  • Latency target (%) (latency_target_pct) — the minimum fraction of checks that must come in below the threshold. A value of 95 means 95% of checks must be fast.
  • Exclude from latency SLO (latency_excluded_regions) — a per-region toggle. Excluded regions still run checks and contribute to uptime; their response times are removed from the p95 aggregate and burn-rate math. Useful when a geographically distant region has an inherently high round-trip time that would otherwise drain your budget for users who are not in that region.

7-day Current vs. 1-hour Recent

The latency panel shows two p95 figures side by side:

  • Current p95 — the 95th-percentile response time computed over the full SLO window (default: last 7 days). This is the number that feeds the error-budget calculation. It changes slowly and reflects your steady-state latency.
  • Recent p95 (1 h) — the 95th-percentile over the last hour only. This reacts quickly to recent changes — a deployment, a traffic spike, or a regional network event — and lets you distinguish a brief degradation from a sustained trend without waiting for it to affect the 7-day figure.

A rising Recent p95 that has not yet moved the Current p95 is an early warning sign. A rising Current p95 means the degradation has been sustained long enough to affect your budget.

Per-region latency breakdown

Below the main latency card, a grid shows the p95 for each probe region. This lets you identify whether a latency problem is global or limited to one or two regions — which often points to regional infrastructure, CDN routing, or network path issues.

Regions excluded from the SLO are shown with an excluded badge. Their p95 values are still displayed so you can monitor them for informational purposes. Regions whose p95 exceeds your threshold are highlighted so they stand out at a glance.

Latency SLO and plan tiers

The latency SLO and the per-region breakdown are available on Business and Founder plans for http_check monitors only. On lower plans the latency panel is hidden. Upgrade from Settings → Plan.