Why Observability is Essential for Kubernetes


Kubernetes observability will become table stakes for teams looking to minimize disruption, maintain velocity, and improve business performance.

As today’s cloud architectures become more complex, engineers are waking up to the fact that many of the most common problems in operating Kubernetes aren’t simple bugs to be fixed but are instead endemic challenges that will require strategic changes. When developer teams struggle to track the state of Kubernetes, serverless functions, or other aspects of their cloud architectures, they need to address the root of their problems: lack of observability.

When we talk about the idea of observability, we’re not just talking about a single product. Observability is not just another word for logging or metrics, and it’s not just a feature that your team can buy off the shelf and consider accomplished. Observability is the idea of how long your team spends trying to understand a problem — how long does it take to recognize that an issue is taking place, identify the root cause and come up with a solution? If you can look at a dashboard and immediately understand the source of your problems, then your organization has excellent observability. On the other hand, if it takes hours for you to get to the bottom of an issue and often ends with a manual restart to fix the problem, then the biggest issue you need to address is a lack of observability.

See also: Researchers Find 380,000 Open Kubernetes Servers

The unrealized potential of Kubernetes

Kubernetes makes it possible for organizations to improve uptime, efficiency, and overall performance, allowing them to gain a competitive advantage in a hectic technology marketplace. A well-orchestrated cluster is one of the strongest tools to deliver stronger performance than competitors, but the full potential of Kubernetes clusters can only be realized by those who practice and maintain true observability.

When organizations increase the use of microservices in their architectures, they simultaneously increase their surface area and frequency of software changes. While these adjustments are made in the name of efficiency, they also bring about an increased need for visibility into cloud-native applications and infrastructure. Close monitoring is not only needed to identify anomalies as they occur but also to recognize where excess capacity in the system can be put to better use.

Kubernetes observability in practice

What does it mean to have good observability in a Kubernetes cluster? These four benchmarks indicate that an organization is using true observability to track, visualize and troubleshoot its entire Kubernetes environment:

Tracking overall health and dynamic behavior

Many engineers in 2022 would consider “infrastructure monitoring” to be an old-fashioned concept, but in reality, it has never been more important than in cloud architectures. When unexpected behaviors and performance issues pop up, the first step should be to evaluate a cluster’s overall health.

Understanding the behavior of the cluster as a whole requires particular attention to dynamic behaviors like new deployments, health checks, and autoscaling. Your cluster’s real-world performance will be determined by the Kubernetes control plane, and tracking dynamic events is critical. An organization with good observability practices will track API server stats, scheduler, and more to understand what’s happening at any given moment.

Correlating log data and performance information

Fundamentally, observability is about speed — how much time are you investing in identifying and solving a problem? In Kubernetes, context switching can be the ultimate factor in determining speed. When trying to solve a critical issue, developers often find themselves switching contexts from logs to overall monitoring to tracing tools. This fragmented user experience makes it difficult to correlate data and, therefore, even harder to solve problems at speed. Open-source observability tools like OpenTelemetry work to address this problem by developing “logs in context,” connecting logging data to other monitoring tools. These connections then make it easier for engineers to correlate causes and see what incident triggered a certain issue.

Another key aspect of Kubernetes observability is correlating performance information with business intelligence. Tracking the usage level of specific customers or their parent organization can provide crucial intelligence when trying to solve a problem, revealing patterns that otherwise may not have been obvious. Correlating information like user geography or parent organization allows engineers to think outside the box and consider new solutions to nagging problems.

Understanding in-cluster communication

One of the most common challenges in Kubernetes is understanding communication between the nodes and pods within a cluster. Kubernetes metadata can make it possible to get performance and distributed traces from applications — this instrumentation can be achieved using standards like OpenTelemetry deployed in Kubernetes clusters, open-source tools like Prometheus, StatsD, or Zipkin, or through agents from observability players like New Relic. Regardless of the tool, tracking in-cluster communication provides valuable insight into metrics like throughput, error rates, and transaction times.

Tracing requests throughout the tech stack

In an ideal world, engineers would be able to follow every request as it begins in a mobile app or front end and then moves throughout the entire system. In reality, even the best systems aren’t able to cover every step of a request’s path. Distributed tracing measures timing information from all parts of your tech stack, which provides an invaluable tool when trying to overcome these monitoring gaps and chase an intermittent bug throughout the system. For effective Kubernetes observability, engineering teams need a solution that will display Prometheus monitoring data alongside other telemetry data for unified visibility. This removes the overhead of managing storage and availability of Prometheus, making it possible to focus instead of deploying and scaling software.

A final word

Kubernetes and other complex cloud architectures will continue to play a key role in development and engineering. As competition heats up, Kubernetes observability will become table stakes for teams looking to minimize disruption, maintain velocity, and improve business performance.

Drew Decker

About Drew Decker

Drew Decker is Principal Product Manager, Kubernetes Ecosystem, at New Relic, where he is responsible for shaping New Relic's Kubernetes and Pixie offerings. Prior to this, Drew served several solutions architect roles at New Relic, focusing on Kubernetes, cloud native, and infrastructure. .

Leave a Reply

Your email address will not be published. Required fields are marked *