Apache CloudStack — A Production Deployment Guide

When building the Lab if you hit the following error:

ERROR [kvm.storage.LibvirtStorageAdaptor] (AgentRequest-Handler-1:[]) (logid:5a91eee6) Failed to create RBD storage pool: org.libvirt.LibvirtException: failed to connect to the RADOS monitor on: 10.100.99.21,: No such file or directory
2026-09-02 01:36:54,642 ERROR [kvm.storage.LibvirtStorageAdaptor] (AgentRequest-Handler-1:[]) (logid:5a91eee6) Failed to create the RBD storage pool, cleaning up the libvirt secret

The trouble shooting steps are as follows:

This is a known, documented, one-line fault, and it's not the server hardware. It's the Ceph authentication client, not your networking or your monitor address.

What the error really means. When you add the Zone, CloudStack asks a KVM host to register the RBD (Ceph) primary storage pool, and libvirt opens a connection to your monitor at 10.100.99.21. That monitor address is correct. The No such file or directory message is misleading — the true cause is that the cephx user CloudStack logs in as (client.cloudstack) is missing its mgr capability.

Here's the mechanism: creating the pool makes libvirt stat the cluster, and modern Ceph serves that statistic through the manager (mgr) daemon. A client with only mon and osd capabilities can happily list pools and create/delete RBD images from the command line — so every smoke test passes — but the moment libvirt asks the mgr for cluster stats, it gets a file-not-found from a cluster that is otherwise working perfectly. That's exactly why retrying the wizard four times gave the same result: the wizard was never the problem.

Confirm it first (30 seconds). On ceph-01 (10.100.99.21):

sudo cephadm shell -- ceph auth get client.cloudstack

Look at the caps lines. You want to see all three of these:

caps mon = "profile rbd"
caps osd = "profile rbd pool=rbd-prod, profile rbd pool=rbd-dev"
caps mgr = "profile rbd pool=rbd-prod, profile rbd pool=rbd-dev"

If the caps mgr line is missing (or the client doesn't exist at all), that's your fault confirmed.

The fix. Set the full capability set — ceph auth caps replaces all caps, so list all three. Run this on ceph-01:

sudo cephadm shell -- ceph auth caps client.cloudstack \
    mon 'profile rbd' \
    osd 'profile rbd pool=rbd-prod, profile rbd pool=rbd-dev' \
    mgr 'profile rbd pool=rbd-prod, profile rbd pool=rbd-dev'

Important and reassuring: this does not change the key. The RADOS Secret you already entered in CloudStack stays valid — you don't need to touch the storage form. Just go back and add the Zone again; the pool will register this time.

(If ceph auth get client.cloudstack instead reports that the client doesn't exist, you skipped the client-creation step — create it with the command in the deployment guide §5.7 and use the printed key as your RADOS Secret, with RADOS User set to cloudstack.)

Where this is in the book, so you can see the full explanation: it's the deployment guide, Phase 5, 5.7 "The Dedicated cephx Client" — the paragraph right after the ceph auth get-or-create command describes this precise error.

The troubleshooting guide's Section 6 (CloudStack Layer) and the diagnostic sweep in Section 8 are your companions for anything else.

Two quick sanity checks only if the above somehow isn't it (it almost certainly is):

1. On the KVM/compute host (not the Ceph nodes), confirm ceph-common and qemu-block-extra are installed — deployment guide Phase 7.2, troubleshooting §6.6. These give the hypervisor its RBD driver.
2. In the CloudStack primary-storage form, the RADOS User field should be exactly cloudstack (no client. prefix), matching the cephx user you just granted caps to.

If you made it this far you are one capability away from a working zone.

This article is powered by MatrixClaw using Testlab hardware and ebook content as the system build & diagnostic aide tools.

Thank you to everyone who takes the time to read this: I created "Apache Cloudstack Production Deployment Guide" In the hope it will help others fast track what took me a little longer. Most people reading this are most likely here as looking for a new Hypervisor. Apache Cloudstack on it's own is not that out of the box, its an orchestrator for hypervisors that connects the dots with Storage, Compute, networking at Cloud level with API's and a script engine for automation.

Do not let that put you off, the whole aim of this book is to give you the tools to deploy all. Show you how you can run open source software on bare metal infrastructure even in Production environments for enterprise, hyperscaler, IaaS, PaaS, SaaS provider, education, financial services, healthcare as well as small business and home labs. The book covers the current LTS version of Apache Cloudstack 4.22.1.0.

KVM is the chosen hypervisor and if like me you thought there was only one true tier 1 hypervisor to rule them all, I can advise that assessment is misguided.

Apache Cloudstack and KVM hypervisor along with Ceph are like a swiss army knife.

I hope if you purchase this book and take nothing else way it is that there is more than one and that is good. The book uses KVM as that is open source. It uses Ceph for storage, that is open source.

To help the book has a back story as to why 'Leaf Spine Books' a fictional organisation, needed Apache Cloudstack. The book also provides the how. Starting at a test lab level where you can use a single laptop, or a Server, it expands to multiple servers, networking and scalable storage if you need to build production grade. However if all you have is a single laptop then that is all you need to build everything. Management, Storage, Compute, network, monitoring.

Ubuntu 24.04 LTS linux is used as the base OS for bare metal regardless of laptop or server(s). You can Even run Multiple laptops if you want to stretch to Production grade and do not have servers.

To help with all of this if you make your way to GitHUB: https://github.com/leafspinebooks/matrixclaw-nano

You will find a set of Python scripts we have engineered and shared for all under AGPL-3.0. See LICENSE - the full, unmodified text from gnu.org.

Copyright (c) 2026 Spiral Matrix Limited and Michael John Leslie Hinsley.

See DISCLAIMER.md for the full statement.

The reason to publish on Leanpub is that we are an AI first engineering company and consultancy. The book is fully tested using our framework called MatrixClaw that has allowed us to fast track testing. MatrixClaw nano is a non AI driven set of Python scripts that we have run ourselves. Based on using 2 different laptops with their base OS replaced with Ubuntu 24.04 LTS server then installing ubuntu-desktop over the top so the laptop is still a true laptop but also an engineering station.

With that in mind every chapter of the book is built and tested, ratified and updated if we find a fix that needs to be applied. Rather than a done once and published book our ethos is run, test and test again. Any fixes we find will be applied should we find any and the book re published. Any scripts that need updating the same and will be both updated in the book as well as on our GitHUB repository.

I hope everyone enjoys both the story line in the book which is fictional, and the full engineering which is real.

I have previously spoken at the European Apache Cloudstack meet up in London. and spent over 32+ year in IT 20 of which around enterprise infrastructure and hypervisor technology. The last 10 years building out Private Cloud and SaaS.

In March 2026 I decided to go back to being a consultant and founder This book in part is a piece of that journey. I hope you will join me on that journey as I publish other titles, show how Apache Cloudstack is the key in all, including AI factories, Genesis systems, even IOT / Edge systems Allowing you to focus on your business, research, or personal goals.

For anyone interested in Apache Cloudstack Aquaponics and Apiculture you can find my talk online here: https://youtu.be/t7nvgegjjv4?si=6mIW_n3NWc2-iY1L