CloudStack

Apache CloudStack is open-source cloud computing software. It is used to deploy a infrastructure as a service (IaaS) platform on virtualization technologies such as KVM, VMware, and Xen. This is similar to OpenStack but is significantly simpler to setup and manage (albeit with less features).

This page contains my notes on setting up and using CloudStack 4.15. I am by no means a CloudStack expert so take my notes here with a huge grain of salt and feel free to make corrections.

Installation

This installation is based on CloudStack 4.15 using CentOS 8. The setup described below uses KVM and Open vSwitch. I'm basing the design decisions and approach from the installation guide at http://docs.cloudstack.apache.org/en/latest/quickinstallationguide/qig.html

Overview

I will have 1 management node and a few bare metal nodes. All nodes will have the same processor (Intel something) and memory (24GB).

Each node will have the same network configuration based on OpenVSwitch. There will be only 1 ethernet connection per node with various VLANs trunked to each node. The VLANs are:

Network	Vlan	Network subnet
Management	11, untagged	172.19.0.0/20
Storage	3205	172.22.0.0/24
Guest	100 - 200	n/a
Public	2	136.159.1.0/24

The network configs for the 4 nodes I'll be using are listed below. There is also a NFS server used for primary storage. The reason for the weird IPs is because this was set up on an existing network.

Node	Networks
management	Management: 172.19.12.141/20 Storage: 172.22.0.241/24
baremetal1	Management: 172.19.12.142/20 Storage: 172.22.0.242/24
baremetal2	Management: 172.19.12.143/20 Storage: 172.22.0.243/24
baremetal3	Management: 172.19.12.144/20 Storage: 172.22.0.244/24
netapp1	Storage: 172.22.0.19/24

Switch config

For completeness, here's the configuration of the HP Procurve switch that the nodes are connected to. The switch should have all the guest VLANs defined and tagged.

config

# Guest VLANs
vlan 100 name guest100
vlan 101 name guest101
...
vlan 200 name guest 200
interface 1-8 tagged vlan 100-200

# Public, management, storage VLANs
vlan 2 name public
vlan 11 name management
vlan 3205 name storage
interface 1-8 untagged vlan 11
interface 1-8 tagged vlan 2,3205

Node setup

Each node will be set up with the following sub-steps.

CloudStack Repos

Install CloudStack repos.

# cat > /etc/yum.repos.d/cloudstack.repo <<EOF
[cloudstack]
name=cloudstack
baseurl=http://download.cloudstack.org/centos/8/4.15/
enabled=1
gpgcheck=0
EOF

Install base packages

Install all other dependencies.

# yum -y install epel-release
# yum -y install bridge-utils net-tools

Install OpenVSwitch from CentOS Extras:

# yum -y install \
http://mirror.centos.org/centos/8/extras/x86_64/os/Packages/centos-release-nfv-openvswitch-1-3.el8.noarch.rpm \
http://mirror.centos.org/centos/8/extras/x86_64/os/Packages/centos-release-nfv-common-1-3.el8.noarch.rpm

Disable SELinux

The system should have SELinux disabled. Use setenforce and edit the selinux config:

# setenforce 0
# vi /etc/selinux/config 
## disable selinux

Disable firewalld

# systemctl stop firewalld
# systemctl disable firewalld

Configure Open vSwitch

# echo "blacklist bridge" >> /etc/modprobe.d/local-blacklist.conf
# echo "install bridge /bin/false" >> /etc/modprobe.d/local-dontload.conf

# systemctl start openvswitch
# systemctl enable openvswitch

We will be using network-scripts to configure the Open vSwitch bridges later. I removed NetworkManager but retained network-scripts to ensure NetworkManager doesn't interfere with my network setup. The install guide leaves NetworkManager around.

I create a 'shared' bridge that's tied to the network interface called nic0. This was done to make it easier to change the bridge setup during my testing but this could be simplified. Each of the physical networks I later set up in CloudStack are its own individual bridge to make it obvious how VMs get connected to the network.

# ovs-vsctl add-br   nic0
# ovs-vsctl add-port nic0 enp4s0f0 tag=11 vlan_mode=native-untagged
# ovs-vsctl set port nic0 trunks=2,11,40-49,3205

# ovs-vsctl add-br management0 nic0 11
# ovs-vsctl add-br cloudbr0 nic0 2
# ovs-vsctl add-br cloudbr1 nic0 100
# ovs-vsctl add-br storage0 nic0 3205

The node's management IP address needs to be removed from the primary network interface and then assigned on the management0 interface. If you're doing this to a node remotely, this might interrupt your connection.

# ip addr del 172.19.12.141/20 dev enp4s0f0
# ip addr add 172.19.12.141/20 dev management0
# ip route add default via 172.19.0.3
# ip addr add 172.22.0.241/24 dev storage0

# ip link set management0 up
# ip link set storage0 up

Network configuration

Once the Open vSwitch bridges are set up, configure the interfaces as follows:

Network Interface	Role	Configuration
enp4s0f0	primary NIC in the host	up on boot; no IP
nic0	network OVS switch that connects to the other bridges to the NIC	up on boot; no IP
cloudbr0	public traffic.	up on boot; no IP
cloudbr1	guest traffic	up on boot; no IP
management0	management traffic	up on boot; assigned with management IP
storage0	storage traffic	up on boot; assigned with storage network IP

Network configs are applied using network-scripts. The idea here is to have the network interfaces be configured when the system boots automatically. For interfaces that require a static IP address, I used the following network-scripts file. Adjust the device name and IP address as required.

# cat <<EOF > /etc/sysconfig/network-scripts/ifcfg-cloudbr0
DEVICE=cloudbr0
TYPE=Bridge
ONBOOT=yes
BOOTPROTO=static
IPV6INIT=no
IPV6_AUTOCONF=no
DELAY=5
IPADDR=172.16.10.2
GATEWAY=172.16.10.1
NETMASK=255.255.255.0
DNS1=8.8.8.8
DNS2=8.8.4.4
USERCTL=no
NM_CONTROLLED=no
EOF

For devices that don't require a static IP:

cat <<EOF > ifcfg-cloudbr0
DEVICE=cloudbr0
TYPE=OVSBridge
DEVICETYPE=ovs
ONBOOT=yes
BOOTPROTO=none
HOTPLUG=no
NM_CONTROLLED=no
EOF

Once configured, verify that your node comes up with the proper network settings on a reboot.

Management node setup

On the management node, set up the network configs and the CloudStack management packages.

Setup Storage

If you intend to use the management server as the primary and secondary storage, you will need to set up a NFS server. If you intend to use an external NFS server as the primary storage, you can skip this step.

# mkdir -p /export/primary /export/secondary
# yum -y install nfs-utils
# cat > /etc/exports <<EOF
/export/secondary *(rw,async,no_root_squash,no_subtree_check)
/export/primary *(rw,async,no_root_squash,no_subtree_check)
EOF
# systemctl start nfs-server
# systemctl enable nfs-server

CloudStack management services

Install MySQL. MariaDB isn't supported and the installation fails with it.

# rpm -ivh http://repo.mysql.com/mysql80-community-release-el8.rpm
# yum -y install mysql-server
# yum -y install mysql-connector-python

## edit /etc/my.cnf to have the following lines.
cat >> /etc/my.cnf <<EOF
[mysqld]
innodb_rollback_on_timeout=1
innodb_lock_wait_timeout=600
max_connections=350
log-bin=mysql-bin
binlog-format = 'ROW'
EOF

# systemctl enable mysqld
# systemctl start mysqld

Setup CloudStack.

# yum -y install cloudstack-management

# cloudstack-setup-databases cloud:password@localhost --deploy-as=root
# cloudstack-setup-management
# systemctl start cloudstack-management
# systemctl enable cloudstack-management

After starting cloudstack-management for the firs time, it might take from 2-10 minutes for the database to set up completely. During this time, the web interface won't be responsive. In the mean time, you will need to seed the system VM images to the secondary storage. If you are using an external NFS server for your secondary storage, adjust the mount point in the following command accordingly.

## Seed the systemvm into secondary storage
# /usr/share/cloudstack-common/scripts/storage/secondary/cloud-install-sys-tmplt -m /export/secondary -u https://download.cloudstack.org/systemvm/4.15/systemvmtemplate-4.15.1-kvm.qcow2.bz2 -h kvm -F

We will continue the setup process via the web interface after setting up a bare metal node.

Bare metal node setup

You should set up at least one bare metal node which will be used to set up your first zone and pod.

On a bare metal node, set up everything outlined in the Node setup section above. The node should have the CloudStack repos, Open vSwitch, SElinux/firewalld, and the networking configured. The agent node must have virtualization enabled on the CPU and KVM should be installed. You should be able to find /dev/kvm on the system.

CloudStack Agent

To set up the node, install the cloudstack-agent package.

# yum -y install cloudstack-agent

Configure qemu and libvirtd.

## edit /etc/libvirt/qemu.conf 
vnc_listen=0.0.0.0

## edit /etc/libvirt/libvirtd.conf
listen_tls = 0
listen_tcp = 1
tcp_port = "16509"
auth_tcp = "none"
mdns_adv = 0

The CloudStack install guide instructs you to edit the libvirtd arguments to --listen, but this will prevent libvirtd from starting using systemd. Instead, you should skip this step entirely because the CloudStack agent will configure this for you when you add the node to a zone.

## The install guide suggests editing /etc/sysconfig/libvirtd to use the listen flag.
## However, this only works if you're not using systemd or using the libvirtd-tcp socket.
## I skipped this step since the agent will configure this later on.
LIBVIRTD_ARGS="--listen"

Start the CloudStack agent. The CloudStack agent should also automatically bring up libvirtd (it's a service dependency).

Allow sudo access

Ensure that /etc/sudoers does not require TTY. In the older documentation, CloudStack requires that the 'cloud' user be able to sudo with the addition of Defaults:cloud !requiretty. However, looking at the installation on the CentOS 8 box, the agent actually runs as root, so perhaps root needs to be able to sudo?

Setting up your first zone

At this point in the process, you should have at least one bare metal host and your management node should be up and running and it should be serving the CloudStack web UI at http://cloudstack:8080/client. Login using the default admin / password credentials.

You will be greeted with a setup wizard. I have had no luck with this and it's better to ignore it. Instead, navigate to Infrastructure -> zones and manually set up your first zone.

Description	Screenshot
There are 3 types of zones that you can create: Basic zone - All guest VMs are placed on a single shared flat network. There is no isolation or security policies in place to prevent guest VMs from seeing each other. Advanced zone - Guest VMs can be placed in one or more VLAN based networks. Guest networks can either be isolated or L2. Isolated networks (depending on the chosen network offering) comes with a virtual router (VR) which offers NAT/SNAT and firewall services and uses one or more public IP addresses. L2 networks are similar but doesn't have a virtual router but instead requires these services to be offered externally. Tenants can also create something called a virtual private cloud (VPC). A VPC is like a regular isolated guest network but with additional features. A VPC allows the user to: Create multiple subnets (called tiers) which can route with each other Network traffic between tiers can be controlled through Network ACLs One or more public IPs can be associated to a VPC. Like an isolated guest network, all subnets can be NATed out through a single public IP You can create a private gateway (and therefore static routes) within a VPC You can create a VPN connection to a VPC Advanced zone with security groups - Guest VMs are placed on a shared network that is publicly routable. There is no concept of a 'public' network because the guest network should also be public. As a result, there is no ability to create any other kind of guest networks or VPCs. The only benefit here is the ability to define security groups per-VM (which is implemented via IPTables on the bare metal host). Because enabling security groups in a zone will restrict that zone from being able to create isolated guest networks or VPCs, the security group feature only appears useful in an environment where guests only need to connect to the internet. Be aware of each type's limitations before continuing. We will be creating an advanced network zone.
We will add the DNS resolvers for the zone and specify the hypervisor type (KVM). Empty the guest CIDR since we're going to allow users to specify their own.
When using the advanced zone, you need to specify the physical networks for the management, storage, and public networks. These should correspond to the physical network devices on the hypervisor. Recall that in the previous step where we set up the Open vSwitch bridges, we created the following bridges for each role: management - management0 storage - storage0 public - cloudbr0 guest - cloudbr1
Specify the public network. The addresses defined here populates the 'Public IP' pool. All isolated guest networks and all VPCs will use one of the addresses defined in this pool for the SNAT/NAT. The addresses specified here should therefore be accessible from the internet.
Create a new pod. The pod network here should cover your management network subnet. The reserved IP addresses here will be used by system VMs that require access to the management network.
Specify the guest network VLAN range. Because we're using VLAN as an isolation method, this range specifies what VLANs the guest networks will use over the guest physical network.
Specify the storage network. The reserved start/end IPs will be used by system VMs that require access to the primary storage. If you are assigning static IPs on your bare metal hosts, ensure that the reserved addresses don't overlap with the IP range specified here (because I had CloudStack assign a VM with the same IP as a bare metal host)
Specify a cluster name.
Add your first bare metal host. You must add one host now and can add additional ones later.
Specify your primary storage. The server should be accessible from the storage network.
Specify your secondary storage. You need to have at least one NFS secondary storage that has been seeded with the system VM template. Secondary storage pools should be accessible from the management network (confirm?)
Launch the zone. This step might take a few minutes. If all goes well, you can then enable the zone shortly after. If you run into any problems, check the logs on the management node at /var/log/cloudstack/management.

Once your zone has been enabled, it should automatically start a Console Proxy VM and secondary storage VM. You can find this under Infrastructure -> System VMs. If for some reason the System VMs are not starting, check that your systemvm template is available in your secondary storage and that the cloud0 bridge on each host is up. You should be able to ping the link local IP address (the 169.254.x.x address) from the hypervisor.

Once the two system VMs are running, verify that you're able to create new guest networks or VPCs. These networks should create a virtual router.

Configuration

Service offerings

Deployment planner

There are a few deployment techniques that can be used. These are set within a compute offering and cannot be changed after it's been created (really? can we change it via API?). The options are:

Deployment planner	Description
First fit	Placed on the first host that has sufficient capacity
User dispersing	Evenly distributes VMs by account across clusters
User concentrated	Opposite of the above.
Implicit dedication	requires or prefers (depending on planner mode) a dedicated host
Bare metal	requires a bare metal host

More information from CloudStack's documentation on Compute and Disk Service Offerings.

Enable SAML2 authentication

Enable the SAML2 plugin by setting saml2.enabled=true under Global Settings.

Set up SAML authentication by specifying the following settings:

Setting	Description	Example value
saml2.default.idpid	The URL of the identity provider	https://sts.windows.net/c609a0ec-xxx-xxx-xxx-xxxxxxxxxxxx/
saml2.idp.metadata.url	The metadata XML URL	https://login.microsoftonline.com/609a0ec-xxx-xxx-xxx-xxxxxxxxxxxx/federationmetadata/2007-06/federationmetadata.xml?appid=c5b8df24-xxx-xxx-xxx-xxxxxxxxxxxx
saml2.sp.id	The identifier string for this application	cloudstack-test.my-organization.tld
saml2.redirect.url	The redirect URL using your cloudstack domain.	https://cloudstack-test.my-organization.tld/client
saml2.user.attribute	The attribute to use. If you're not sure what's available, look at the management logs after a login attempt.	For Azure AD: http://schemas.xmlsoap.org/ws/2005/05/identity/claims/emailaddress

Restart the management server. To allow a user access, create the user and enable SSO. The user's username must match the value that's obtained from the saml2.user.attribute field.

Bugs

SAML Request being rejected by Azure AD

If you are using Azure AD, you may have issues authenticating because the SAML request ID that's generated might begin with a number. When this happens, you will get an error similar to: AADSTS7500529: The value '692rv91k6dgmdas33vr3b2keahr4lqjv' is not a valid SAML ID. The ID must not begin with a number.. For more information, see: https://github.com/apache/cloudstack/issues/5548

Users cannot login via SSO

Users that will be using SAML for authentication will need to have their CloudStack accounts created with SSO enabled. There seems to be a bug with the CloudStack web UI where a user's SAML IdPID isn't settable (it gets set to a '0'). A work-around would be to create and authorize users via CloudMonkey.

The steps on adding a new user are:

1. Create the user: create user firstname=First lastname=User email=user1@ucalgary.ca username=user1@ucalgary.ca account=RCS state=enabled password=asdf
2. Find the user's ID: list users domainid=<tab> filter=username,id
3. Authorize the user: authorize samlsso enable=true entityid=https://sts.windows.net/c609a0ec-xxx-xxx-xxx-xxxxxxxxxxxx/ userid=user-id
4. Verify that the user is enabled for SSO: list samlauthorization filter=userid,idpid,status

When authorizing a user, the entityid must be the URL of the identity provider. The end slash is also mandatory.

Enable SSL

With a reverse proxy doing SSL termination

My proof of concept version has CloudStack sit behind a Traefik reverse proxy that does SSL termination. This is by far the easiest solution since Traefik handles all the SSL renewals automatically and you don't need to tinker with the cloudstack management server. The console proxies are also proxied through Traefik by specifying a consoleproxy.url.domain (Eg. *.cloudstack-test.example.com which is converted to the IP-address version once the system VM is online). Note that consoles require both HTTP (port 80) and WSS (port 8080) to be reverse proxied using this method.

My dynamic traefik configs to make this happen looks like the following:

http:
  routers:
    cloudstack:
      rule: Host(`cloudstack-test.example.com`)
      service: cloudstack-poc
      entrypoints:
        - http
      middlewares:
        - https-redirect

    cloudstack-https:
      rule: Host(`cloudstack-test.example.com`)
      service: cloudstack-poc
      entrypoints:
        - https
      tls:
        certresolver: letsencrypt

    cloudstack-pub-ip-136-159-1-100:
      rule: Host(`136-159-1-1.cloudstack-test.example.com`)
      service: 136-159-1-100
      entrypoints:
        - https
      tls:
        certresolver: letsencrypt

    cloudstack-pub-ip-136-159-1-100-ws:
      rule: Host(`136-159-1-1.cloudstack-test.example.com`)
      service: 136-159-1-100-ws
      entrypoints:
        - httpws
      tls:
        certresolver: letsencrypt
		
  services:
    cloudstack-poc:
      loadBalancer:
        servers:
          - url: "http://172.19.12.141:8080"

    136-159-1-100:
      loadBalancer:
        servers:
          - url: "http://136.159.1.100"

    136-159-1-100-ws:
      loadBalancer:
        servers:
          - url: "http://136.159.1.100:8080"

  middlewares:
    https-redirect:
      redirectscheme:
        scheme: https

Without a reverse proxy

If you need to serve CloudStack without a reverse proxy, you would need to obtain your SSL certificates and convert it into PKCS12 format. The conversion is done using the keytool utility.

## Combine Files
# cat key.key servercert.crt intermediate.crt root.crt > combined.crt

## Create keystore
# openssl pkcs12 -in combined.crt -export -out combined.pkcs12

## Import keystore
# keytool -importkeystore -srckeystore combined.pkcs12 -srcstoretype PKCS12 -destkeystore /etc/cloudstack/management/combined.pkcs12 -deststoretype pkcs12

Edit /etc/cloudstack/management/server.properties with:

https.enable=true
https.keystore=/etc/cloudstack/management/combined.pkcs12
https.keystore.password=<enter the same password as used for conversion>

For more details, see: https://www.shapeblue.com/securing-cloudstack-4-11-with-https-tls/

Open-ended questions

Console Proxy via DNS name?

Is it possible to set up console proxy to use a specific DNS name rather than directly using the public IP address that the console proxy was assigned?

This is possibly an issue if the console proxy and web services need to sit behind a reverse proxy (yes, I know, they should have a public IP already).

Supposedly, you can use '*' which will be replaced by the IP address of the proxy. Eg. *.example.com will be converted to 136-159-1-2.example.com

Enable SSL on Console Proxy

Set consoleproxy.sslEnabled = true. But you also need to set the SSL cert for this to work. Where do you do this?

Compute offerings with 'unlimited' CPU cycles?

Compute offerings require a MHz value assigned. Why is this? Can we just assign a VM entire cores?

- If you read the docs, CPU (in MHz) only has an effect if CPU cap is selected. In all other cases, the value here is something akin to 'cpu shares'.

- if you put in a huge number like 9999, deployment would fail though.

How to implement showback?

Is there a way to implement showback based on resources consumed by account?

Monitoring resources?

Is there a way to monitor resource usage by account, node? Any good way to push VMs into a CMDB like ServiceNow?

NetApp integration?

Is it possible to do guest VM snapshots by leveraging NetApp?

Tasks

Re-add existing KVM bare metal host

Once a host has been added to CloudStack, the CloudStack agent will have generated some public/private keys and configured itself to talk to the management node. If you need to remove and re-add a host, you will need to clean up the agent before re-adding it back to CloudStack again. Based on my experience, I had to do the following:

1. Before removing the host from CloudStack, drain it of all VMs. virsh list should be empty. If not and you've removed the host from the management server already, manually kill each VM with virsh destroy.
2. systemctl stop cloudstack-agent
3. rm -rf /etc/cloudstack/agent/cloud*
4. systemctl stop libvirt
5. rm -rf /var/lib/libvirt/qemu
6. Edit /etc/cloudstack/agent/agent.properties and remove the keystore passphrase, any UUIDs and GUIDs, cluster/pod/zone, and the host.
7. Restart with systemctl start cloudstack-agent (libvirt should come up automatically as it's a dependency). Ensure that it comes up OK.

You may then re-add the host back to CloudStack.

Building RPMs

To build the RPM packages from scratch, you'll need to install a bunch of dependencies and then run the build script. For more information, see:

• https://docs.cloudstack.apache.org/en/4.15.2.0/installguide/building_from_source.html#building-rpms-from-source

# yum groupinstall "Development Tools"
# yum install java-11-openjdk-devel genisoimage mysql mysql-server createrepo
# yum install epel-release

# curl -sL https://rpm.nodesource.com/setup_12.x | sudo bash -
# yum install nodejs

# cat <<EOF > /etc/yum.repos.d/mysql.repo
[mysql-community]
name=MySQL Community connectors
baseurl=http://repo.mysql.com/yum/mysql-connectors-community/el/$releasever/$basearch/
gpgkey=http://repo.mysql.com/RPM-GPG-KEY-mysql
enabled=1
gpgcheck=1
EOF
# yum -y install mysql-connector-python

enable powertools

# yum install jpackage-utils maven

# git clone https://github.com/apache/cloudstack.git
# cd cloudstack
# git checkout 4.15

# cd packaging
# sh package.sh --distribution centos8

Usage server

Install cloudstack-usage. Start it and restart the management server. Set enable.usage.server=true in global settings.

Question: Where is the collected data located? Can we visualize it in the UI?

Tools

CloudMonkey

Get started:

• Download from: https://github.com/apache/cloudstack-cloudmonkey/releases/tag/6.1.0
• Documentation at: https://cwiki.apache.org/confluence/display/CLOUDSTACK/CloudStack+cloudmonkey+CLI

When you first run CloudMonkey, you will need to set the cloud URL:

$ cmk
> sync
> set url http://172.19.12.141:8080/client/api
> set username admin
> set password password

The settings are then saved to ~/.cmk/config.

Run sync to fetch all the available API calls that your account can use. You can then tab complete any available commands.

Cheat sheet

What	Command
Change output format	set display table|json
Create compute offering	create serviceoffering name=rcs.c2 displaytext=Medium cpunumber=2 cpuspeed=750 memory=2048 storagetype=shared provisioningtype=thin offerha=false limitcpuuse=false isvolatile=false issystem=false deploymentplanner=UserDispersingPlanner cachemode=none customized=false
Add a new host	add host clusterid=XX podid=XX zoneid=XX hypervisor=KVM password=**** username=root url=http://bm01

Automate deployments

There is an example script on how to automate a basic zone deployment at: https://github.com/apache/cloudstack-cloudmonkey/wiki/Usage

Limitations

List of limitations I've come across during testing.

• You cannot unmount an ISO image from a running VM. Why?

Troubleshooting

When you run into issues, check the logs in /var/log/cloudstack/. There's typically a stacktrace which gets generated whenever you encounter an error.

Can't create shared network in a advanced zone using Open vSwitch

Whenever I try creating a shared network in an advanced zone that is using OVS, the step fails with: "Unable to convert network offering with specified id to network profile".

Stack trace shows that the OVS guest network guru isn't able at designing the network because the zone isn't capable of handling this network offering.

2021-09-28 16:36:26,416 DEBUG [c.c.a.ApiServer] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) CIDRs from which account 'Acct[76a1585d-1bf6-11ec-a3c5-8f3e88f01ab1-admin]' is allowed to perform API calls: 0.0.0.0/0,::/0
2021-09-28 16:36:26,439 DEBUG [c.c.u.AccountManagerImpl] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) Access granted to Acct[76a1585d-1bf6-11ec-a3c5-8f3e88f01ab1-admin] to [Network Offering [7-Guest-DefaultSharedNetworkOffering] by AffinityGroupAccessChecker
2021-09-28 16:36:26,517 DEBUG [c.c.n.g.BigSwitchBcfGuestNetworkGuru] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) Refusing to design this network, the physical isolation type is not BCF_SEGMENT
2021-09-28 16:36:26,521 DEBUG [o.a.c.n.c.m.ContrailGuru] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) Refusing to design this network
2021-09-28 16:36:26,524 DEBUG [c.c.n.g.NiciraNvpGuestNetworkGuru] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) Refusing to design this network
2021-09-28 16:36:26,527 DEBUG [o.a.c.n.o.OpendaylightGuestNetworkGuru] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) Refusing to design this network
2021-09-28 16:36:26,530 DEBUG [c.c.n.g.OvsGuestNetworkGuru] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) Refusing to design this network
2021-09-28 16:36:26,536 DEBUG [c.c.n.g.DirectNetworkGuru] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) GRE: VLAN
2021-09-28 16:36:26,536 DEBUG [c.c.n.g.DirectNetworkGuru] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) GRE: VXLAN
2021-09-28 16:36:26,536 INFO  [c.c.n.g.DirectNetworkGuru] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) Refusing to design this network
2021-09-28 16:36:26,539 INFO  [c.c.n.g.DirectNetworkGuru] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) Refusing to design this network
2021-09-28 16:36:26,543 DEBUG [o.a.c.n.g.SspGuestNetworkGuru] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) SSP not configured to be active
2021-09-28 16:36:26,546 DEBUG [c.c.n.g.BrocadeVcsGuestNetworkGuru] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) Refusing to design this network
2021-09-28 16:36:26,549 DEBUG [o.a.c.e.o.NetworkOrchestrator] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) Releasing lock for Acct[76a0531f-1bf6-11ec-a3c5-8f3e88f01ab1-system]
2021-09-28 16:36:26,624 DEBUG [c.c.u.d.T.Transaction] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) Rolling back the transaction: Time = 172 Name =  qtp1816147548-400; called by -TransactionLegacy.rollback:888-TransactionLegacy.removeUpTo:831-TransactionLegacy.close:655-Transaction.execute:38-Transaction.execute:47-NetworkOrches
trator.createGuestNetwork:2572-NetworkOrchestrator.createGuestNetwork:2327-NetworkServiceImpl$4.doInTransaction:1502-NetworkServiceImpl$4.doInTransaction:1450-Transaction.execute:40-NetworkServiceImpl.commitNetwork:1450-NetworkServiceImpl.createGuestNetwork:1366
2021-09-28 16:36:26,667 ERROR [c.c.a.ApiServer] (qtp1816147548-400:ctx-291672d1 ctx-3f19296a) (logid:83c45c2a) unhandled exception executing api command: [Ljava.lang.String;@69a8823d
com.cloud.utils.exception.CloudRuntimeException: Unable to convert network offering with specified id to network profile
        at org.apache.cloudstack.engine.orchestration.NetworkOrchestrator.setupNetwork(NetworkOrchestrator.java:739)
        at org.apache.cloudstack.engine.orchestration.NetworkOrchestrator$10.doInTransaction(NetworkOrchestrator.java:2634)
        at org.apache.cloudstack.engine.orchestration.NetworkOrchestrator$10.doInTransaction(NetworkOrchestrator.java:2572)
        at com.cloud.utils.db.Transaction$2.doInTransaction(Transaction.java:50)
        at com.cloud.utils.db.Transaction.execute(Transaction.java:40)
        at com.cloud.utils.db.Transaction.execute(Transaction.java:47)
...

Possible answer

The guest network was set up with GRE isolation. This however isn't supported with KVM as the hypervisor (see this presentation). After re-creating the zone with the guest physical network set up with just VLAN isolation, I was able to create a regular shared guest network that all tenants within the zone can see and use.

To make the shared network SNAT out, I created another shared network offering that also has SourceNat and StaticNat.

$ cmk list serviceofferings issystem=true name='System Offering For Software Router'
$ cmk create networkoffering \
name=SharedNetworkOfferingWithSourceNatService displaytext="Shared Network Offering with Source NAT Service" traffictype=GUEST guestiptype=shared conservemode=true specifyvlan=true specifyipranges=true \
serviceofferingid=307b14d8-afd1-43ea-948c-ffe882cd5926 \
supportedservices=Dhcp,Dns,Firewall,SourceNat,StaticNat,PortForwarding \
serviceProviderList[0].service=Dhcp serviceProviderList[0].provider=VirtualRouter \
serviceProviderList[1].service=Dns serviceProviderList[1].provider=VirtualRouter \
serviceProviderList[2].service=Firewall serviceProviderList[2].provider=VirtualRouter \
serviceProviderList[3].service=SourceNat serviceProviderList[3].provider=VirtualRouter \
serviceProviderList[4].service=StaticNat serviceProviderList[4].provider=VirtualRouter \
serviceProviderList[5].service=PortForwarding serviceProviderList[5].provider=VirtualRouter \
servicecapabilitylist[0].service=SourceNat servicecapabilitylist[0].capabilitytype=SupportedSourceNatTypes servicecapabilitylist[0].capabilityvalue=peraccount

Using this network offering, I was able to create a shared network in the advanced networking zone that has a NAT service which is visible to all accounts. The only issue with this approach is that there isn't a way to create a port forwarding for a specific VM because the account that owns this network is 'system'.