Pick what you want to know about a metric and the builder writes the PromQL, with a plain-English explanation. Below it: a cheat sheet of the queries people look up most.
Runs entirely in your browser — nothing you enter here is sent to a server.
The patterns below cover most dashboards and alerts. Replace the metric and label names with your own.
| Goal | Query |
|---|---|
| Requests per second | sum(rate(http_requests_total[5m])) |
| Requests per second, per service | sum by (service) (rate(http_requests_total[5m])) |
| Error percentage | 100 * sum(rate(http_requests_total{code=~"5.."}[5m])) / sum(rate(http_requests_total[5m])) |
| p95 latency | histogram_quantile(0.95, sum by (le) (rate(http_request_duration_seconds_bucket[5m]))) |
| CPU usage, % | 100 - avg by (instance) (rate(node_cpu_seconds_total{mode="idle"}[5m])) * 100 |
| Memory used, % | 100 * (1 - node_memory_MemAvailable_bytes / node_memory_MemTotal_bytes) |
| Disk full in 4 hours | predict_linear(node_filesystem_avail_bytes[1h], 4 * 3600) < 0 |
| Target is down | up == 0 |
| Top 5 by request rate | topk(5, sum by (handler) (rate(http_requests_total[5m]))) |
| Restarts in the last hour | increase(kube_pod_container_status_restarts_total[1h]) > 0 |
| Series count per metric | topk(10, count by (__name__) ({__name__=~".+"})) |
sum(rate(x[5m])) is right. rate(sum(x)[5m]) breaks on counter resets and is not valid without a subquery.
A raw counter only goes up and tells you nothing. A gauge is already a current value; use the *_over_time functions on it.
rate() needs at least two samples in the window. With a 30-second scrape, use [2m] or more; [5m] is a safe default.
histogram_quantile() needs the bucket boundary label. Aggregate with sum by (le, …), never plain sum.
rate() averages over the whole range and is stable, which makes it right for alerts and most graphs. irate() uses only the last two samples, so it reacts instantly but is noisy; use it for zoomed-in graphs.
They are the same calculation in different units. rate() gives the per-second average; increase() gives the total growth over the range, which is rate multiplied by the range in seconds.
Use histogram_quantile() on the rate of the metric's _bucket series, keeping the le label in the aggregation: histogram_quantile(0.95, sum by (le) (rate(metric_bucket[5m]))).
Check the label matchers first — a mistyped label value matches nothing without an error. Then check that the range is at least twice the scrape interval and that the target is still being scraped (up == 1).
Upzilla checks your sites, servers and APIs around the clock and alerts you the moment something breaks. 10 monitors free, no card required.