🗓️ 28072025 1359
📎

prometheus_histograms

A histogram in Prometheus samples observations into buckets — useful for tracking latency, durations, sizes.

Records:

• count: total # of observations
• sum: total of all values
• bucket: # of observations ≤ a certain value

📐 Data Format

Each metric is split into:

<metric_name>_bucket{le="0.1"}  → count ≤ 0.1 <metric_name>_bucket{le="0.5"}  → count ≤ 0.5 <metric_name>_bucket{le="+Inf"} → total count <metric_name>_sum              → total sum <metric_name>_count            → total number of events

You get a distribution across buckets.

🧮 Common PromQL Queries

1. Rate of requests per second: rate(http_request_duration_seconds_count[1m])

2. Average duration: rate(http_request_duration_seconds_sum[1m]) / rate(http_request_duration_seconds_count[1m])

3. Percentile approximation (via histogram_quantile): histogram_quantile(0.95, rate(http_request_duration_seconds_bucket[5m]))

Note: This is an approximation!

Java

static final Histogram requestLatency = Histogram.build()
      .name("myapp_request_duration_seconds")
      .help("Request duration in seconds.")
      .buckets(0.1, 0.3, 0.5, 1, 2, 5) // custom buckets
      .register();

🧠 Tips

• Always use rate() or irate() with counters like _bucket, _count, _sum
• Bucket count is cumulative up to the le boundary
• Use +Inf bucket for total event count
• Higher granularity buckets = more precision = more cardinality

---

References