In 2016, we at Honeycomb first borrowed the term “observability” from the wikipedia entry for control systems observability, where it is a measure of your ability to understand internal system states just by observing its outputs. We then spent a couple of years trying to work out how that definition might apply to software systems. Many twitter threads, podcasts, blog posts, and lengthy laundry lists of technical criteria emerged from that work, including a whole ass book.

Engineering has come a long way since the days of delivering discrete, point-in-time products that were often packaged on a CD and shipped to customers. The days of physical media and long development cycles are long gone. The advent of cloud computing and the rise of Software-as-a-Service (SaaS) transformed the landscape, creating a new model of continuous development and service delivery. This shift has not only revolutionized how software is developed, but has also redefined the engineer’s role.

Sampling is a necessity for applications at scale. We at Honeycomb sample our data through the use of our Refinery tool, and we recommend that you do too. But how do you get started? Do you simply a set rate for all data and a handful of drop and keep rules, or is there more to it? What do these rules even mean, and how do you implement them?

As part of our recent failure testing project, we ran into an interesting failure mode involving the OpenTelemetry SDK for Go. In this post, we’ll show you why our apps stopped sending telemetry for over 15 minutes and how we enabled keepalives to prevent this kind of failure from happening in the future.

Simply put, full-stack observability is monitoring designed for modern, cloud-native architectures. It allows you to understand how your software system interacts at scale, across everything from traditional mainframes and legacy clients to modern serverless or Kubernetes-based services. 

Setting clear, measurable goals is essential for any successful team. However, aligning those goals with the technical work can be challenging in the fast-paced world of software engineering. Engineers might focus on reducing latency or improving uptime, while business leaders look at revenue and customer satisfaction. It gets tricky to track the impact between the two to justify when specific engineering initiatives are important, why, and how they impact the bottom line. Everyone may feel the work is important, but it's hard to see or remember why!

Alerts are a perennial topic, and a CoPE will need to engage with them. The bounds of this problem space are formed by two types of alerts:  Reactive alerts (in Honeycomb, we call these Triggers): They are alerts that fire after some event, like crossing a pre-determined boundary.  Proactive alerts (Burn Alerts based on Honeycomb’s SLO feature): These give notice before crossing a threshold; in the case of SLOs, that means before failing to meet the stated objective.

The previous post laid out the basic idea of instrumentation and how OpenTelemetry’s auto-instrumentation can get teams started. However, you can’t rely only on auto-instrumentation. This post will discuss the limitations in more detail and how a CoPE can help teams overcome them.

The Collector is the focal point for telemetry inside your cluster. Instead of your containerized applications sending directly to your OpenTelemetry-capable backend (the place that allows you to ask questions of your telemetry), we send that data to an internal location first, then forward the data on.

“How is my app performing?” is one of the most common, yet hardest questions to answer. There are myriad ways to measure this, like error rate, average response time, and so on. Enter the Application Performance Index (aka Apdex), a single metric that attempts to answer, “Are my application’s users happy?”

It’s one of my strongly held beliefs that errors are constructed, not discovered. However we frame an incident’s causes, contributing factors, and context ends up influencing the shape of the corrective items (if any) that get created. I’ll cover these ideas by using our June 3rd incident where a database migration caused a large outage by locking up a shared database and making it run out of connections.

The CoPE is made to affect, meaning change, how things work. The disruption it produces is a feature, not a bug. That disruption pushes things away from a locally optimal, comfortable state that generates diminishing returns. It sets things on a course of exploration to find new terrains which may benefit it more—and for longer.