Zebrium Blog

The State of Monitoring

Monitoring is about catching unexpected changes in application behavior. Traditional monitoring tools achieve this through alert rules and spotting outliers in dashboards. While this traditional approach can typically catch failure modes with obvious service impacting symptoms, it has two limitations:

A new release of your web service has just rolled out with some awesome new features and countless bug fixes. A few days later and you get a call: Why am I not seeing my what-ch-ma-call-it on my thing-a-ma-gig? After setting up that zoom call it is clear that the browser has cached old code, so you ask the person to hard reload the page with Ctrl-F5. Unless its a Mac in which case you need Command-Shift-R. And with IE you have to click on Refresh with Shift. You need to do this on the other page as well. Meet the browser cache, the bane of web service developers!

Automatically Spot Critical Incidents and Show Me Root Cause

We believe the future of monitoring, especially for platforms like Kubernetes, is truly autonomous. Cloud native applications are increasingly distributed, evolving faster and failing in new ways, making it harder to monitor, troubleshoot and resolve incidents. Traditional approaches such as dashboards, carefully tuned alert rules and searches through logs are reactive and time intensive, hurting productivity, the user experience and MTTR. We believe machine learning can do much better – detecting anomalous patterns automatically, creating highly diagnostic incident alerts and shortening time to resolution.

Introduction

At Zebrium, we provide an Autonomous Monitoring service that automatically detects anomalies within logs and metrics. We started by correlating anomalies across log streams to automatically raise incidents that require our user's attention. Now, we have taken the step to augment our incident detection by detecting anomalies within a group of related metrics and correlate those with anomalies found in logs.

Based on our experience with hundreds of incidents across nearly a hundred unique application stacks, we have developed deep insights into the specific ways modern software breaks. This led us to thoughtfully design a multi-layer machine learning stack that can reliably detect these patterns, and identify the collection of events that describes each incident. In simple terms, here is what we have learned about real-world incidents when software breaks. You can also try it yourself by signing up for a free account.

A few months ago, our friends at Maya Data joined our private Beta to give our Autonomous Log Monitoring platform a test run. During the test, they used their newly created Litmus Chaos Engine to generate issues in their Kubernetes cluster, and we managed to detect all of them successfully using our Machine Learning completely unsupervised. Needless to say, they were impressed!

TL;DR

Monitoring today puts far too much burden on DevOps and developers. These teams spend countless hours staring at dashboards, hunting through logs, and maintaining fragile alert rules. Fortunately, unsupervised machine learning can be applied to logs and metrics to autonomously detect and find the root cause of critical incidents. Read more below, or start using our Autonomous Monitoring Platform for free - it takes less than 2 minutes to get started.  

Introduction

Monitoring today is extremely human driven. The only thing we’ve automated with monitoring to date is the ability to alert on rules that watch for specific metrics and events that occur when something known goes wrong. Everything else - building parsing rules, configuring and maintaining dashboards and alerts, and troubleshooting incidents - requires a lot of manual effort from expert operators that intuitively know and understand the system being monitored.

As the principal responsible for the design of middleware software at Zebrium, I’m writing to share some of the choices we made and how they have held up. Middleware in this context means the business-logic that sits between persistent storage and a web-based user interface.

As part of my job, I recently had to modify Fluentd to be able to stream logs to our (Zebrium) Autonomous Log Monitoring platform. In order to do this, I needed to first understand how Fluentd collected Kubernetes metadata. I thought that what I learned might be useful/interesting to others and so decided to write this blog.

In our last blog we discussed the need for Autonomous Monitoring solutions to help developers and operations users keep increasingly large and complex distributed applications up and running.

Observability means being able to infer the internal state of your system through knowledge of external outputs. For all but the simplest applications, it’s widely accepted that software observability requires a combination of metrics, traces and events (e.g. logs). As to the last one, a growing chorus of voices strongly advocates for structuring log events upfront. Why? Well, to pick a few reasons - without structure you find yourself dealing with the pain of unwieldy text indexes, fragile and hard to maintain regexes, and reactive searches. You’re also impaired in your ability to understand patterns like multi-line events (e.g. stack traces), or to correlate events with metrics (e.g. by transaction ID).

Logs are the source of truth when trying to uncover latent problems in a software system. They are usually too messy and voluminous to analyze proactively, so they are used mostly for reactive troubleshooting once a problem is known to have occurred.

At Zebrium, we have a saying: “Structure First”. We talk a lot about structuring because it allows us to do amazing things with log data. But most people don’t know what we mean when we say the word “structure”, or why it allows for amazing things like anomaly detection. This is a gentle and intuitive introduction to “structure” as we mean it.

Kubernetes makes it easy to deploy, manage and scale large distributed applications. But what happens when something goes wrong with an app? And how do you even know? We hear variations on this all the time: “It was only when customers started complaining that we realized our service had degraded”, “A simple authentication problem stopped customers logging. It took six hours to resolve.”, and so on.

A widely prevalent application monitoring strategy today is sometimes described as “black box” monitoring. Black box monitoring focuses just on externally visible symptoms, including those that approximate the user experience. Black box monitoring is a good way to know when things are broken.

Kubernetes and other orchestration tools use abstraction to hide complexity. Deploying, managing and scaling a distributed application are made easy. But what happens when something goes wrong? And, when it does, do you even know?

It is especially tricky to identify software problems in the kinds of distributed applications typically deployed in k8s environments. There’s usually a mix of home grown, 3^rd party and OSS components – taking more effort to normalize, parse and filter log and metric data into a manageable state. In a more traditional world tailing or grepping logs might have worked to track down problems, but that doesn’t work in a Kubernetes app with a multitude of ephemeral containers. You need to centralize logs, but that comes with its own problems. The sheer volume can bog down the text indexes of traditional logging tools. Centralization also adds confusion by breaking up connected events (such as multi-line stack traces) in interleaved outputs from multiple sources.

A quick, free and easy way to find anomalies in your logs

Why understand log structure at all?

Have you ever spent time tracking down a bug or failure, only to find you’ve seen it before? Or a variation of this problem: at the completion of automated test you have to spend time triaging each failure, even though many are caused by the same bug. All this can impact productivity, especially in continuous integration and continuous deployment (CI/CD) environments, where things change rapidly.

Developers and testers constantly use log files and metrics to find and troubleshoot failures. But their lack of structure makes extracting useful information without data wrangling, regexes and parsing scripts a challenge.

It takes great skill, tenacity and sometimes blind luck to find the root cause of a technical issue. And for complex problems, more often than not, it involves leveraging log files, metrics and traces.  Whether you’re a tester triaging problems found during automated test, or a developer assisting with a critical escalation, dealing with data is painful.

According to Gartner, product analytics, “help manufacturers evaluate product defects, identify opportunities for product improvements, detect patterns in usage or capacity of products, and link all these factors to customers". The benefits are clear. But there are barriers – product analytics are expensive in terms of time, people and expertise.

We structure machine data at scale