Observability

Prometheus / PromQL

PromQL selectors, aggregations, rate functions and common patterns for querying Prometheus metrics.

Instant Selectors

http_requests_totalAll time series for a metric

http_requests_total{job="api"}Filter by exact label value

http_requests_total{status=~"5.."}Regex match on label

http_requests_total{status!="200"}Exclude a label value

http_requests_total{job="api",status="200"}Multiple label filters

http_requests_total[5m]5-minute range vector

rate(http_requests_total[5m])Per-second rate smoothed over 5m

irate(http_requests_total[5m])Instantaneous rate (last 2 samples)

increase(http_requests_total[1h])Absolute increase over 1 hour

delta(temperature_celsius[10m])Change in gauge value over 10m

sum(http_requests_total)Sum all series

sum by (job) (http_requests_total)Sum grouped by job label

sum without (instance) (http_requests_total)Sum dropping instance label

avg(cpu_usage)Average across all series

max(memory_bytes)Maximum value

min(memory_bytes)Minimum value

count(up)Count of time series

topk(5, http_requests_total)Top 5 series by value

bottomk(5, http_requests_total)Bottom 5 series by value

metric_a + metric_bAdd two metrics (label matching)

http_errors_total / http_requests_totalError ratio

metric > 0.9Filter series where value exceeds threshold

metric * 100Multiply all values by 100

metric offset 1hQuery value from 1 hour ago

histogram_quantile(0.95, rate(http_duration_bucket[5m]))95th percentile latency from histogram

predict_linear(metric[1h], 3600)Linear prediction 1 hour ahead

absent(up{job="api"})Alert when a series has no data

label_replace(metric, "new", "$1", "old", "(.*)")Rename or transform a label

clamp(metric, 0, 1)Clamp values between 0 and 1