Types of Metrics (Counters, Gauges, Histograms)

Introduction Counters Gauges Histograms Best Practices FAQ

Introduction

In observability, metrics are crucial for monitoring the performance and health of systems. Metrics can be broadly categorized into three types: Counters, Gauges, and Histograms.

Counters

A Counter is a cumulative metric that only increases over time. It is ideal for tracking events such as the number of requests received, errors, or completed tasks.

Key Characteristics of Counters

Monotonically increasing: Counters never decrease.
Resettable: They can be reset to zero, but it's uncommon.
Used for aggregating counts over time.

Example usage in Prometheus:

http_requests_total{method="POST", handler="comments"} 1027

Gauges

A Gauge is a metric that represents a single numerical value that can arbitrarily go up and down. Gauges are suitable for tracking values such as temperature, memory usage, or current active sessions.

Key Characteristics of Gauges

Can increase or decrease: Gauges can reflect both upward and downward trends.
Useful for measuring instantaneous values.

Example usage in Prometheus:

memory_usage_bytes{instance="localhost:9090"} 524288000

Histograms

A Histogram samples observations and counts them in configurable buckets. It is useful to visualize the distribution of data points, such as request durations or response sizes.

Key Characteristics of Histograms

Tracks the count and sum of observations: Histograms can calculate averages and percentiles.
Configurable buckets: You can define the ranges for the histogram.

Example usage in Prometheus:

http_request_duration_seconds{method="GET", handler="homepage"} 0.0045

Best Practices

Choose the right metric type based on what you are measuring.
Use consistent naming conventions for metrics.
Monitor cardinality to avoid performance issues.
Aggregate metrics over time for better insights.

Note: Always consider the context of the metric to ensure it provides meaningful insights into system performance.

FAQ

What is the difference between a Counter and a Gauge?

A Counter only increases over time, while a Gauge can increase or decrease, reflecting the current state.

When should I use a Histogram over a Gauge?

Use a Histogram when you need to observe the distribution of a data set, such as response times, instead of a single value.

Can Counters be reset?

While Counters can technically be reset to zero, this is typically not done in practice, as it would lose the historical data.