On-prem and self-hosted
The same reports, run by you: one Helm chart in one namespace of your own Kubernetes cluster, on your own hardware or in your own cloud. Your cost data stays in your own database, sign-in goes through the install's own Keycloak, and nothing of what it collects reaches us.
Why run it yourself
- Your data stays with you. What your infrastructure costs, what runs where and what it is called is kept in a database you run, inside your own network.
- Hardware with no bill. A Kubernetes cluster on your own servers has no cloud bill to read. qloudpulse prices it from what you say a core-hour and a GiB-hour cost you, and splits that across every pod, with the idle capacity priced apart.
- No route in, and none out. Nothing outside your cluster connects to the install, and the install connects to nothing outside it unless you turn something on.
What it reads
Your Kubernetes clusters, today
One small pod in each cluster, from the public qloudpulse-public repository, sends what each pod requests and uses to the install. Every hour the collector splits each node's cost across the pods on it, by namespace and workload, and once a day it works out what is idle. A node's hour costs its allocatable cores times your price per core-hour plus its allocatable GiB times your price per GiB-hour, and a GPU node can add a price per GPU-hour. Prices are in US dollars.
VMware vSphere Pilot
Hosts, VMs, datastores and snapshots from each vCenter you name, every hour, with a vCenter login that holds the built-in Read-only role, and vCenter's certificate always checked. Each vSphere cluster is priced per core-hour, GiB-hour and GiB-month of storage, and each VM pays its share. The settings and the login are in the chart. A pilot: we set it up with you.
Your AWS accounts
The same read-only roles as on the hosted plan, read from inside your network. An install that only reads your clusters needs no AWS account.
What it installs
One chart, in one namespace that is the install's alone, with rights in that namespace only: no cluster-wide objects, no operator and no message broker. It runs:
- The report service, with the reports, the settings and a Troubleshooting page for the install's admins.
- The collector, with the schedule inside it, and workers that can take runs beside it.
- Keycloak, for sign-in, asking every user for an authenticator app unless you turn that off. Its admin console has no public route.
- Postgres, or a connection to one you already run.
- VictoriaMetrics, which keeps what the cluster pods send.
Nothing leaves your network
- Closed by default. A network policy for every pod allows only the paths it needs. Nothing outside the namespace reaches the database, and the database reaches nothing at all.
- Nothing calls us. The licence is checked inside your cluster. The install connects to the internet only if you turn on sending support bundles, or give it a vCenter to read.
- Encrypted between pods. Every connection between its pods is TLS, and each side checks the other's certificate.
- Locked-down pods. The namespace can enforce Kubernetes' restricted Pod Security, so nothing less locked down runs in it.
What you need
- Kubernetes 1.25 or later, with a network plugin that enforces NetworkPolicy.
- An ingress controller that can use HTTPS to the pods behind it, or your own route to them.
- Two addresses, one for the reports and one for sign-in, and a TLS certificate for them.
- A storage class for Postgres and VictoriaMetrics.
- About 3.5 GiB of memory for its pods.
During early access we set up each install with you: the images and the chart come from us, with a licence for your company.
The licence
An install collects while its licence is valid. The licence is a file of one line that we sign, for the period you have paid for, and it names your install, your company and the plan, never a person. Your company's admins download it, and you put it in a Secret.
| When | Collection | Reports |
|---|---|---|
| Until it ends | Runs | As always |
| 14 days more | Runs | A banner says when collection stops |
| After that | No new runs | Readable as before, with a banner |
Nothing is deleted or hidden: your data stays in your database whatever the licence says. To renew, apply the new file to the same Secret; the install reads it within two minutes, with no restart, and each job catches up. $300 a month per install, with the first month free. Pricing
Running it
- Troubleshooting page. The install's admins see what a sign-in, the database, the licence and each job are doing, and can run a failed job again. The same checks answer on a port that only
kubectl port-forwardreaches. - Support bundles you read first. When you need our help, an admin prepares a bundle of checks, settings without their passwords and the report service's recent log lines, with email addresses and tokens masked. They read it before it goes anywhere, and it is encrypted so that only we can open it, whether it is sent or downloaded and passed on another way.
- Upgrades. A
helm upgradekeeps your users, passwords and certificates. Back up your database before one: the chart does not do it for you yet.
Hosted and self-hosted compared
| Hosted | Self-hosted | |
|---|---|---|
| Who runs it | We do, on AWS in Ireland | You do, in your own cluster |
| Your data | A database of your company's own, in ours | Your own database |
| Sign-in | qloudpulse's login, or Google | The install's own Keycloak |
| Reads | AWS, Kubernetes, DigitalOcean and Datadog | AWS, Kubernetes and VMware vSphere (a pilot) |
| Price | By your AWS spend, free up to $2,500 a month | $300 a month per install |
A Kubernetes cluster on your own hardware can also send to the hosted plan, with one pod and nothing else to install. Choose self-hosted when the data itself has to stay inside your network.
Getting started
Write to us with the clusters you want to see, and we set you up with the chart, the images and a licence, with the first month free.