What Are The 4 Types Of Metrics In Prometheus

Prometheus supports four main metric types, each suited for specific use cases. These types help capture various aspects of system performance and behavior.

1. Counter

A counter is a cumulative metric that only increases over time or resets to zero. It is ideal for tracking counts, such as the number of requests or errors.

Characteristics:

• Monotonically increasing.
• Resets only when the application restarts.

Use cases:

• Total number of HTTP requests.
• Count of failed operations.
• Number of tasks completed.

Example:

from prometheus_client import Counter

http_requests = Counter('http_requests_total', 'Total HTTP requests')
http_requests.inc()  # Increment by 1
http_requests.inc(5)  # Increment by 5

2. Gauge

A gauge represents a value that can go up or down. It measures instantaneous values, such as memory usage or active connections.

Characteristics:

• Can increase or decrease.
• Tracks current state at a specific moment.

Use cases:

• CPU or memory usage.
• Number of in-progress requests.
• Size of a queue.

Example:

from prometheus_client import Gauge

memory_usage = Gauge('memory_usage_bytes', 'Current memory usage in bytes')
memory_usage.set(512)  # Set value
memory_usage.inc(128)  # Increment
memory_usage.dec(64)   # Decrement

3. Histogram

A histogram samples observations and categorizes them into predefined buckets. It is used to measure the distribution of values, such as request durations or response sizes.

Characteristics:

• Buckets define ranges for observations.
• Provides cumulative counts for each bucket and a +Inf bucket for all values.
• Also tracks the total sum and count of observations.

Use cases:

• Measuring request latency.
• Analyzing response size distribution.
• Tracking database query durations.

Example:

from prometheus_client import Histogram

request_latency = Histogram(
    'http_request_latency_seconds', 'Request latency in seconds',
    buckets=[0.1, 0.5, 1, 2.5, 5, 10]
)
request_latency.observe(1.2)  # Record an observation

Metrics exposed:

• http_request_latency_seconds_bucket{le="0.5"} 15
• http_request_latency_seconds_sum 25
• http_request_latency_seconds_count 30

4. Summary

A summary is similar to a histogram but focuses on providing quantiles (e.g., 95th percentile). It tracks the total number of observations, their sum, and configurable quantile values.

Characteristics:

• Quantiles are configurable per summary.
• Does not use buckets.
• Best for small-scale scenarios where exact quantiles are needed.

Use cases:

• Measuring request latency with specific percentiles.
• Monitoring exact error rates.

Example:

from prometheus_client import Summary

request_latency = Summary('http_request_latency_seconds', 'Request latency in seconds')
request_latency.observe(1.2)  # Record an observation

Comparison of metric types

Got an article suggestion? Let us know

Explore more

What Is A Bucket In Prometheus?

Questions

Limitations of Prometheus Labels

Questions

How To Manage Prometheus Counters

Questions

What is the Difference Between a Gauge and a Counter?

Questions

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.