Operations | Monitoring | ITSM | DevOps | Cloud

On October 29th, 2025, Azure Front Door suffered an outage that impacted Microsoft services on a global level, including Microsoft 365, Outlook, Xbox Live, Copilot, and more. It also affected Microsoft Azure, meaning companies like Costco, Starbucks, and Alaska Airlines ran into issues for both customer-facing and internal systems. The root of the issue was a misconfiguration in the data plane for Azure Front Door and the Azure Content Delivery Network.

KubeCon Atlanta 2025 felt different this year — and CloudZero had a full team on the ground to capture it. Engineers, product leaders, sales reps, and CTO Erik Peterson spent three days embedded across the show floor. Their vantage points were complementary: the outbound conversations, the inbound questions, the demos, the technical deep-dives, and the quieter moments between sessions. Five perspectives stood out.

Running multiple LLMs without aggregation can feel like managing five different clouds with no dashboard. Sure, you can make it work, but you won’t like the bill. And most SaaS teams didn’t start with a multi-LLM strategy. It just happened. You added one model for reasoning, another for summarization, or maybe a fine-tuned version for customer support. Fast-forward six months, and your AI stack looks like a tangle of APIs. And each charges tokens on its own terms.

You know that uneasy pause before opening your monitoring dashboard? The one where you're hoping nothing's broken—but a part of you knows something probably is. Performance issues often start quietly: a few slow endpoints, a checkout that takes longer than usual, a graph that looks a little off. Before long, those small signals turn into alerts and support tickets.

Automating chaos engineering with Terraform eliminates manual setup across environments by enabling you to version control your entire chaos infrastructure, from service discovery to security governance policies. The Harness Terraform provider supports end-to-end automation including Kubernetes infrastructure setup, custom image registries, Git-based ChaosHub management, and granular security controls that ensure safe experiment execution in production.

Behind every reliable dashboard and morning report stands a system of accountability: SQL Server Agent jobs. They’re what keeps backups on time, analytics flowing, and data pipelines steady when everything else moves fast. However, to sustain that reliability, SQL Server Agent must operate with precision at scale.

This post is a deep dive into how we improved the P95 latency of an API endpoint from 5s to 0.3s using a niche little computer science trick called a bloom filter. We’ll cover why the endpoint was slow, the options we considered to make it fast and how we decided between them, and how it all works under the hood.