In early May, we released the first version of our new natural language querying interface, Query Assistant. We also talked a lot about the hard stuff we encountered when building and releasing this feature to all Honeycomb customers. But what we didn’t talk about was how we know how our use of an LLM is doing in production! That’s what this post is all about.

When considering a migration to Kubernetes, as with any major tech upgrade or change, it’s imperative to understand the motivation for doing so. The engineering time and labor to execute a complex migration will take away from other priorities, making it crucial to have org-wide alignment on why the change makes sense. 

We saw a shift this year in how the technology sector honed in on sustainability from a cost perspective. In particular, looking at where they’re spending that revenue in the infrastructure and tooling space. Observability tooling comes under a lot of scrutiny as it’s perceived as a large cost center—and one that could be cut without affecting revenue. After all, if the business hasn’t had a problem in the last few months, we mustn’t need monitoring—right?

While Kubernetes comes with a number of benefits, it’s yet another piece of infrastructure that needs to be managed. Here, I’ll talk about three interesting ways that Honeycomb uses Honeycomb to get insight into our Kubernetes clusters. It’s worth calling out that we at Honeycomb use Amazon EKS to manage the control plane of our cluster, so this document will focus on monitoring Kubernetes as a consumer of a managed service.

Kubernetes helps teams of all sizes optimize their microservices architecture by enabling seamless automated containerized app deployment, easy scalability, and efficient operations. But Kubernetes also has a reputation for being difficult to learn and complex to manage, and when you’re new to something, it’s hard to know what you don’t know. That’s where Honeycomb observability comes in—distributed traces provide real-time visibility into container operations so they can be fine-tuned for optimal performance and bugs can be found and fixed faster, paving the way to a successful delivery.

Running a Kubernetes cluster isn’t easy. With all the benefits come complexities and unknowns. In order to truly understand your Kubernetes cluster and all the resources running inside, you need access to the treasure trove of telemetry that Kubernetes provides. With the right tools, you can get access to all the events, logs, and metrics of all the nodes, pods, containers, etc. running in your cluster. So which tool should you choose? Since we are all in on OpenTelemetry, we think the best solution is the OpenTelemetry Collector.

Frontend observability is a tricky problem. No website is free of errors or slowdowns; sites break down in weird ways for all kinds of reasons. Accounting for every possible combination of platform, browser, extensions, and (sometimes baffling) user behavior would be an impossible task. How do we decide which errors are important? One useful framework for making these frontend development decisions is customer-centric observability.

In a simple deployment, an application will emit spans, metrics, and logs which will be sent to api.honeycomb.io and show up in charts. This works for small projects and organizations that do not control outbound access from their servers. If your organization has more components, network rules, or requires tail-based sampling, you’ll need to create a telemetry pipeline.

When people hear the word “migration,” they typically think about migrating from on-prem to the cloud. In reality, companies do migrations of varying types and sizes all the time. However, many teams delay making critical migrations or technical upgrades because they don’t have the proper tools and frameworks to de-risk the process.

Modern software development—where code is shipped fast and fixed quickly—simply can’t happen without building observability in before deployments happen. Teams need to see inside the code and CI/CD pipelines before anything ships, because finding problems early makes them easier to fix. And don’t take “easier to fix” lightly: incidents found late not only impact the customer experience (and perhaps their loyalty), but they also hit developer productivity and job satisfaction hard.

At Traceloop, we’re solving the single thing engineers hate most: writing tests for their code. More specifically, writing tests for complex systems with lots of side effects, such as this imaginary one, which is still a lot simpler than most architectures I’ve seen.

There’s a lot of hype around AI, and in particular, Large Language Models (LLMs). To be blunt, a lot of that hype is just some demo bullshit that would fall over the instant anyone tried to use it for a real task that their job depends on. The reality is far less glamorous: it’s hard to build a real product backed by an LLM.