Go

For years Google has consistently been able to cut down multiple percentage points in their fleet-wide resource usage every quarter, using techniques described in their “Google-Wide Profiling” paper. Ad-hoc profiling has long been part of the developer’s toolbox to analyze the CPU and memory usage of a running process. However, through continuous profiling, and the systematic collection of profiles, entirely new workflows suddenly become possible. The presenter will start this talk with an introduction to profiling applications, and demonstrate how one can practice it using open-source continuous profiling tools, and how continuous profiling allows for an unprecedented fleet-wide understanding of code at production runtime. ...

In modern days, we run our applications as loosely coupled micro-services on distributed, elastic infrastructure as (mostly) stateless workloads. Under these circumstances, observability has become a key attribute to understand how our applications run and behave in action, in order to provide highly available and resilient service. There exist several observability signals, such as “log”, “metric”, “tracing” and “profiling” that can be collected from a running service, which we can also call pillars of observability. Using these signals, we can create real-time, actionable alerts, create panels where we can monitor applications closely, and perform in-depth analysis to find the root of the systems’ failures. Within the Go and CNCF community, there are a variety of tools that can collect and make these observable signals useful. ...

Observability is the key to understand how your application runs and behaves in action. This is especially vital for distributed environments like Kubernetes, where users run Cloud-Native microservices often written in Go. Among many other observability signals like logs and traces, the metrics signal has a substantial role. Sampled measurements observed throughout the system are crucial for monitoring the health of the applications and, they enable real-time, actionable alerting. While there are many open-source robust libraries, in various languages, that allow us to easily instrument services for backends like Prometheus, there are still numerous possibilities to make a mistake or misuse those libraries. ...