HistogramSummary

This guide shows how to instrument a Flask web application with Prometheus histogram and summary metrics to measure request latency per path and method. It walks through:

Creating a histogram metric
Recording per-request timings using Flask’s before_request and after_request hooks
Inspecting the generated Prometheus metrics (including default buckets)
Customizing histogram buckets
Using a summary metric and the Python-client limitation for quantiles

Tip: Use per-path and per-method labels to slice latency metrics in Prometheus and visualize percentiles in Grafana or another dashboard.

Basic histogram example

Create a histogram metric and observe latency for each request. Place the timing logic in Flask’s before_request and after_request hooks so every request is measured automatically.

from prometheus_client import Histogram, start_http_server
from flask import Flask, request
import time

app = Flask(__name__)

LATENCY = Histogram(
    'request_latency_seconds',
    'Request Latency',
    labelnames=['path', 'method']
)

def before_request():
    request.start_time = time.time()

def after_request(response):
    request_latency = time.time() - request.start_time
    LATENCY.labels(
        request.path,
        request.method
    ).observe(request_latency)
    return response

if __name__ == '__main__':
    start_http_server(8000)
    app.before_request(before_request)
    app.after_request(after_request)
    app.run()

How it works:

before_request() records the timestamp when the request arrives (time.time()).
after_request() computes the latency by subtracting the start time from the current time, then calls observe() on the histogram with the appropriate labels.
start_http_server(8000) exposes metrics on port 8000 for Prometheus to scrape.

Note: You can store the start_time on the request object because Flask exposes a request-local proxy for each incoming request.

Example metric output

When you scrape or query the histogram metric in Prometheus, the client creates buckets and count metrics. A scraped output may look like:

# HELP request_latency_seconds Request Latency
# TYPE request_latency_seconds histogram
request_latency_seconds_bucket{le="0.005",method="GET",path="/cars"} 0.0
request_latency_seconds_bucket{le="0.01",method="GET",path="/cars"} 0.0
request_latency_seconds_bucket{le="0.025",method="GET",path="/cars"} 0.0
request_latency_seconds_bucket{le="0.05",method="GET",path="/cars"} 1.0
request_latency_seconds_bucket{le="0.075",method="GET",path="/cars"} 3.0
request_latency_seconds_bucket{le="0.1",method="GET",path="/cars"} 3.0
request_latency_seconds_bucket{le="0.25",method="GET",path="/cars"} 4.0
request_latency_seconds_bucket{le="0.5",method="GET",path="/cars"} 6.0
request_latency_seconds_bucket{le="0.75",method="GET",path="/cars"} 6.0
request_latency_seconds_bucket{le="1.0",method="GET",path="/cars"} 8.0
request_latency_seconds_bucket{le="2.5",method="GET",path="/cars"} 8.0
request_latency_seconds_bucket{le="5.0",method="GET",path="/cars"} 8.0
request_latency_seconds_bucket{le="7.5",method="GET",path="/cars"} 8.0
request_latency_seconds_bucket{le="10.0",method="GET",path="/cars"} 8.0
request_latency_seconds_bucket{le="+Inf",method="GET",path="/cars"} 8.0
request_latency_seconds_count{method="GET",path="/cars"} 8.0
request_latency_seconds_sum{method="GET",path="/cars"} 1.234

The client provides helpful default buckets suitable for many web applications, but you will often want to customize them.

Customizing histogram buckets

To change bucket boundaries, pass the buckets parameter when you instantiate the Histogram. For low-latency, high-resolution needs, provide a list of bucket upper bounds:

LATENCY = Histogram(
    'request_latency_seconds',
    'Flask Request Latency',
    labelnames=['path', 'method'],
    buckets=[0.01, 0.02, 0.03, 0.04, 0.05, 0.06, 0.07, 0.08, 0.09, 0.1]
)

Adjust buckets to match the latency distribution of your application. Use finer buckets for low latency applications and wider buckets for higher-latency systems.

Using a Summary metric

A Summary records observed values and can calculate configurable quantiles (in some client libraries). The usage pattern is almost identical to a Histogram: create the metric and call .observe() with the measured latency.

from prometheus_client import Summary

LATENCY_SUMMARY = Summary(
    'request_latency_seconds',
    'Flask Request Latency',
    labelnames=['path', 'method']
)

# Later, in your after_request:
LATENCY_SUMMARY.labels(request.path, request.method).observe(request_latency)

Warning: The Python Prometheus client does not implement configurable quantiles for Summary metrics the same way some other language clients do. If you need server-side quantiles, prefer Histograms and compute quantiles in Prometheus (e.g., histogram_quantile()), or check the client library updates for support in your language.

Histogram vs Summary — quick comparison

Metric Type	Use Case	Notes
Histogram	Ideal for calculating quantiles in Prometheus (via `histogram_quantile`) and exposing bucketed counts	You must choose buckets; Prometheus can compute quantiles from buckets
Summary	Records quantiles and sum/count on the client side	Some client libraries (like Python) lack configurable quantiles; server-side quantile calculation may be preferable

Links and references

Use the examples above to add latency instrumentation to your Flask app, tune bucket boundaries to your observed latency distribution, and choose between Histogram and Summary depending on your quantile needs and client library support.

Introduction

Observability Fundamentals

Prometheus Fundamentals

PromQL

Dashboarding & Visualization

Application Instrumentation

Service Discovery

Push Gateway

Alerting

Monitoring Kubernetes

Conclusion

Scaling & Long Term Storage

Basic histogram example

Example metric output

Customizing histogram buckets

Using a Summary metric

Histogram vs Summary — quick comparison

Links and references

Watch Video

​Basic histogram example

​Example metric output

​Customizing histogram buckets

​Using a Summary metric

​Histogram vs Summary — quick comparison

​Links and references

Watch Video

Basic histogram example

Example metric output

Customizing histogram buckets

Using a Summary metric

Histogram vs Summary — quick comparison

Links and references