March 15, 2000. OPRA, which consolidates and publishes US options quotes and trades, takes in 3,544 messages a second over one minute. Its capacity is about 3,540, and the SEC soon writes that the exchanges are approaching it almost daily.

A pod's IP may last seconds, so address rules lag. Cilium enforces policy in the kernel with eBPF, and Kubernetes delivers the policies. Those policies live in etcd. We keep yours in a dedicated, highly available etcd cluster for each tenant. Which of your namespaces has none?

Linux 7.3 is out. Cgroup scheduling gets new weight modes (cgroup_mode, default 'concur') and a single runqueue. Pod CPU limits are enforced through cgroups, so roll it one node pool at a time. Your control plane runs with us, untouched by any drain. Which pool gets 7.3 first?

Kubernetes 1.34 lost patch support on Tuesday. A cluster still on it owes three upgrades to reach 1.37, one minor at a time, and 1.38 is due December 16. We run your control plane, dedicated, and plan that climb with you. Map the route: https://www.kestrelinfrastructure.com/?x=1

Kestrel | Managed Kubernetes for the servers you already own

You own the compute. We manage the state. Kestrel runs your Kubernetes control plane; your bare metal, GPUs and VMs join with one kubeadm command. Priced per cluster, never per core.

kestrelinfrastructure.com

Tandem 16 processors said "I'm alive" every second and checked their peers every two. Silence meant presumed dead. etcd keeps a like watch: if the leader goes quiet, members elect another. We run yours dedicated and highly available. When did it last elect a leader on purpose?

Every Deployment, Secret and role binding in your cluster is an entry in one ledger, etcd, and only the API server holds the pen. So we keep those books on a dedicated, highly available etcd cluster for each tenant, across zones. Place yours with us: https://www.kestrelinfrastructure.com/?x=1

Managed Kubernetes Control Planes for Your Own Servers | Kestrel Infrastructure

We run a dedicated, highly available Kubernetes control plane for the servers you already have: bare metal, GPU servers or VMs in any cloud. One flat monthly price, nothing per CPU.

kestrelinfrastructure.com

Flux 2.10 is out. It supports Kubernetes 1.35 through 1.37, with 1.35.2 as the floor. So a GitOps upgrade starts with the control plane beneath it. We plan those upgrades with you, one minor version at a time. Run kubectl version on each cluster Flux manages. Oldest goes first.

9:02 on a Saturday, a warehouse loses its uplink. Nothing visible happens. The edge node keeps running its pods, and somewhere in the control plane a clock starts.

etcd 3.6.16 is out, closing CVE-2026-63209 and a Hash/Restore data race. The maintainers patched 3.5 and 3.7 the same day. Someone must know which line each cluster runs. We do, for a dedicated, highly available etcd cluster per tenant. How many etcd lines does your fleet run?

Patching nodes should be dull: drain, reboot, next. Unless they also host your etcd, and each reboot spends the one member of three you can lose. We keep your API server and etcd off your nodes, across zones, so a drain only moves pods. Which of yours needs a quorum check?

etcd 3.5.35 is out. A member saving a leader's snapshot now syncs the directory too; on Linux, a file sync alone does not persist its directory entry. We sweat that on a dedicated, highly available etcd cluster for each tenant. Leave the fsyncs to us: https://www.kestrelinfrastructure.com/?x=1

Managed Kubernetes Control Planes for Your Own Servers | Kestrel Infrastructure

We run a dedicated, highly available Kubernetes control plane for the servers you already have: bare metal, GPU servers or VMs in any cloud. One flat monthly price, nothing per CPU.

kestrelinfrastructure.com

EKS, GKE and AKS agree on one price: about $438 a month per cluster to stay on a Kubernetes version past standard support. The fee buys time; the upgrade still waits. We plan yours with you, one minor at a time, on a dedicated control plane we run. How many of yours pay it?

etcd 3.7.3 is out, fixing a leader replication slowdown, a Hash/Restore data race and five Go vulnerability advisories. Your cluster state lives in etcd. We keep each tenant's dedicated, highly available etcd cluster current. Run etcdctl endpoint status. Which version answers?

Friday, 3:00 PM. Your control plane's board: etcd: quorum held. API servers: up across zones. Certificates: rotated on schedule. Access: audited. We keep that board on a dedicated control plane per tenant. Your nodes stay yours. For a quiet weekend: https://www.kestrelinfrastructure.com/?x=1

Managed Kubernetes Control Planes for Your Own Servers | Kestrel Infrastructure

We run a dedicated, highly available Kubernetes control plane for the servers you already have: bare metal, GPU servers or VMs in any cloud. One flat monthly price, nothing per CPU.

kestrelinfrastructure.com

Apollo flew to the Moon on programs woven by hand into core rope. Whatever changed in flight had to fit in 2,048 words of RAM. Your cluster keeps its changing state in etcd. We run a dedicated, highly available etcd cluster for each tenant. Can your team list what is in yours?

CoreDNS 1.15.0 is out, built on Go 1.27.3 to cover seven CVEs, with fixes for several plugin panics. It serves your cluster DNS from your own nodes, so schedule the upgrade. We plan those with you, beside your dedicated control plane. Which CoreDNS version answers your pods?

Each business day before the open, FINRA Rule 4590 has firms sync their clocks to NIST's atomic clock, within a second, and log it. We run your Kubernetes control plane on that routine: certificates rotated on schedule, access audited. When did your certificates last rotate?

A cloud's managed Kubernetes wants your nodes in that cloud; hybrid costs extra. So the control plane decides where your compute lives. We run yours dedicated. Your nodes join with kubeadm join from your own accounts or racks. Join the waitlist: https://www.kestrelinfrastructure.com/?x=1

Hosted Kubernetes Control Planes for Bare Metal & AI | Kestrel

Dedicated Kubernetes control planes for bare metal, GPU and AI clusters, neoclouds, and data centers. We run the API server and etcd; you keep the nodes.

kestrelinfrastructure.com

37signals spent $3.2M on cloud in 2022. So it bought about $600K of Dell servers and projected some $7M saved over five years. Owned servers still need a control plane. We run yours, dedicated; your nodes join with kubeadm join. What would your biggest node pool cost owned?

A node joins on a bootstrap token, valid 24 hours by default. kubeadm join trades it for the kubelet's own TLS credentials, and it answers to the API server. That API server and its etcd, we run for you alone. The node stays in your rack. We are admitting a few firms at a time.

2013. At Stanford, Diego Ongaro and John Ousterhout publish Raft, built to be understood. A change commits once a majority agrees. Your etcd keeps its books that way, so we run a dedicated, highly available etcd cluster for each tenant. How many members can yours lose?

A store's uplink drops. Its edge node keeps serving, since running pods carry on without the API server. Only changes wait for the link. So your etcd can live elsewhere. We run it dedicated and highly available, across zones. How many of your sites keep etcd in a closet?

A kubelet may trail its API server by three minor versions, never lead it. Generous, until the control plane moves four times and the nodes never do. We run your dedicated control plane and plan that drift with you, one minor at a time. Request access: https://www.kestrelinfrastructure.com/?x=1

Hosted Kubernetes Control Planes for Bare Metal & AI | Kestrel

Dedicated Kubernetes control planes for bare metal, GPU and AI clusters, neoclouds, and data centers. We run the API server and etcd; you keep the nodes.

kestrelinfrastructure.com

Your own on-prem server can join EKS. AWS then bills per vCPU-hour for it, on hardware you already paid for. We run a dedicated control plane instead. Your nodes join with a standard kubeadm join and stay in your racks. What did your own servers cost in fees last month?

1964. SABRE runs American Airlines' bookings on two IBM 7090s: one live, one as backup. The airline would not leave its book to one machine. Nor should your cluster. We run a dedicated, highly available etcd cluster for each tenant. Join the waitlist: https://www.kestrelinfrastructure.com/?x=1

Hosted Kubernetes Control Planes as a Service — Kestrel Infrastructure

Hosted, dedicated Kubernetes control planes as a service. We run the API server and etcd; your worker nodes stay on your own bare metal or cloud.

kestrelinfrastructure.com

Every user, controller and kubelet in your cluster comes through one front door: the API server. So we never share that door. Each tenant gets a dedicated control plane, etcd included, and every access is audited. If that is the discretion you expect, we should talk.

etcd keeps every revision until compacted. Let it outgrow its space quota and an alarm limits it to reads and deletes. Your cluster can no longer change. We keep that ledger: a dedicated, highly available etcd cluster for each tenant. How close is your DB size to its quota?

containerd began inside Docker, joined the CNCF in 2017 and graduated in 2019. Today it pulls images and runs containers on your nodes. That half of Kubernetes was always yours. We run the other half: a dedicated control plane, across zones. The waitlist is on our profile.

Racks in your data center, GPU servers at a bare-metal host, boxes at the edge. Give each a network path to the API server, and all three can join one control plane. We run that one, dedicated and across zones. Your nodes stay yours. How many control planes do your sites run?

Kubernetes 1.34 reaches end of life on October 27. Eleven days. Reaching 1.37 takes three upgrades: the control plane climbs one minor at a time, then the nodes follow. We plan each step with you, on a dedicated control plane across zones. On 1.34? kestrelinfrastructure.com