Operations | Monitoring | ITSM | DevOps | Cloud

An honest 2026 comparison of Railway, Render, Fly.io and deploying into your own AWS, GCP, Azure or Scaleway account - with the six measurable signals you have outgrown managed PaaS, a priced cost model, and a four-step framework with if-then verdicts. Romaric founded Qovery to make Kubernetes accessible to every engineering team. He writes about platform strategy, developer experience, and the future of cloud infrastructure.

Table of Contents A few months ago, we introduced Kubex support for the KAI Scheduler to improve GPU sharing for production inference workloads. The basic model is simple: The KAI Scheduler handles placement and GPU sharing. Kubex continuously observes usage and adjusts those allocations as demand changes. KAI provides the scheduling foundation. It lets multiple workloads share a GPU while accounting for the amount of GPU each workload requests. Kubex then closes the loop.

Table of Contents For most of Kubernetes’ history, memory management has been a blunt instrument. Cross your limit, and the kernel kills your container. There has been no equivalent to CPU throttling, no graceful backpressure, just a hard stop. With Kubernetes 1.37, that changes: Memory QoS, built on cgroups v2, graduates to Beta and is enabled by default.

Changing a Kubernetes resource request takes two numbers: what the workload requests, and what it uses. The CloudZero MCP server now returns both, by cluster, namespace, or workload. This gives you a number you can defend. Usage comes back as P95 over the date range you query, 30 days by default. When an engineering lead asks whether a service runs on a smaller request, that is the figure that settles it. Over-provisioning and under-provisioning show up on the same query.

Every organization running AI agents has already made a hosting decision. Most made it by accident. The sales team switched on the agent built into their CRM. Engineering is piloting a coding agent in a vendor’s cloud. Someone on the data team deployed a LangGraph service to a VM with a database key in an environment variable, and someone else is running an agent framework on a laptop with production credentials in a dotfile. Each of these is a hosting decision.

Somewhere in most production setups there is a container that stays awake all week to do one job on Friday. Maybe it re-indexes a database. Maybe it runs a batch report, or a data migration that only matters at month end. Either way, the resources are billed for seven days. Task containers close that gap. They are ephemeral, API-triggered containers that spin up inside an Upsun Cloud environment, run exactly one command, and remove themselves when that command exits. Nothing sits idle between runs.

Kubernetes normally excels at preserving uptime during maintenance tasks, but it’s not always perfect. Even something as benign as consolidating nodes after a traffic spike could take your application offline if not done carefully. This is where PodDisruptionBudgets (PDBs) come in. In this blog, we’ll explain why PDBs are important, what the risks are of not implementing them, and how you can find out which of your own deployments are missing PDB definitions. ‍

If you've ever inherited a cluster that nobody fully understands, or watched two teams solve the same infrastructure problem in completely different ways, you've already felt the cost of not having a template strategy. It's not always dramatic; it starts with a different CNI here, a missing cert-manager version there, but it compounds. Six months later, cluster configuration is tribal knowledge, and the platform team has become the bottleneck.