The 38-point Cluster Health Checklist
The exact list we work through in a paid audit. Quorum and fencing, failure domains, backup restorability, upgrade paths, GPU allocation idleness.
Get the checklist → Resources
What we know, written down
Most of what we publish is the answer to a question a customer asked. No gated whitepapers written by marketing, no "10 trends" listicles. Just the engineering, including the parts where the conventional answer is wrong.
Available now Planned This is a content plan, published as a plan. Nothing here is presented as existing until it does.
Why your HA cluster has never actually failed over
Fencing and STONITH, explained by the failures they prevent. If your cluster cannot fence its peer, it is not highly available.
Ceph CRUSH map mistakes we keep finding
The placement rules that decide durability and rebalance behaviour — and the three configurations that show up in almost every broken cluster.
PostgreSQL HA: Patroni vs repmgr vs pgpool, honestly compared
What each one actually guarantees, where each one fails, and which topology fits which constraint.
Kubernetes 1.34 DRA: what changes for GPU clusters
Dynamic Resource Allocation replaces the device plugin model. What that changes for scheduling, sharing and multi-tenancy.
Slurm vs Kubernetes for LLM training: pick by workload, not by fashion
Gang scheduling, topology awareness, and why the dividing line is about job shape rather than ideology.
The real cost of self-managed Kubernetes (with a TCO worksheet)
Control plane pricing is public. Everything that makes the number real is not. Here is the model to fill in with your own figures.
VMware → Proxmox VE + Ceph: a migration runbook
Wave planning, rollback rehearsal, validation. Written from migrations we have actually run.
We broke our own cluster on purpose: quarterly chaos drill report
A real drill report, published with the numbers, including the parts that went worse than expected.
Backups you have never restored are not backups
A method for timing restores, and why "the backup job succeeded" is the most dangerous sentence in infrastructure.
Bare-metal Kubernetes on Debian: the reference build
The package list, kernel parameters, networking and storage choices we start from, with reasoning.
Managed vs self-managed: what you are actually paying for
A line-by-line comparison of what a managed contract includes versus the hours it displaces.
Free · no call required
The 38-point Cluster Health Checklist
The exact list we work through during a paid audit: quorum and fencing, storage failure domains, backup restorability, upgrade paths, certificate expiry, GPU allocation idleness, and the eleven things that most often turn out to be wrong.
- 38 items, each with what "good" looks like and how to verify it
- Written for engineers, not for procurement
- Includes the questions we ask that usually make people go quiet
Want the checklist and nothing else?
That's fine. Download it, use it, and if it finds something you'd rather not fix yourself, we're easy to reach.