Previously, in Why Histograms? and Histograms vs Summaries, I went over the basics of histograms and summaries, explaining the tradeoffs, benefits, and limitations of each. Because they’re easy to understand and demonstrate, those posts focused on so-called explicit bucket histograms. The exponential bucket histogram, also referred to as native histogram in Prometheus, is a low-cost, efficient alternative to explicit bucket histograms. In this post, I go through what they are, how they work, and the problems they solve that explicit bucket histograms struggle with.
In many ways, histograms and summaries appear quite similar. They both roll up many data points into a data structure for efficient processing, transmission, and storage. They can also both be used to track arbitrary quantiles such as the median or p99 of your data. So how do they differ? Let’s dive in.
A histogram is a multi-value counter that summarizes the distribution of data points. For example, a histogram may have 3 counters which count the occurrences of negative, positive, and zero values respectively. Given a series of numbers,
-1, the histogram would count
3positive values. A single histogram data point is most commonly represented as a bar chart.
A few weeks ago a coworker asked me to do an introductory presentation on the way we use Docker. This is an attempt at distilling my thoughts into a more consumable form. In this post I’ll cover everything from setup, to a basic dockerfile, to deployment and advanced usage.