Deployment Requirements
Plan Management and appliance nodes separately. Agree the hypervisor version and workload profile for your release with MTG. A platform appearing in the wizard does not confirm support for every version or workload.
Choose a Management Deployment
Management supports two deployment options:
-
The Management ISO VM supplies the host operating system, console setup and controls to start and stop Management.
-
The Management container runs with Docker, Docker Compose or rootful Podman on an operator workstation or operator-created Linux VM. You supply and maintain the host, runtime, networking, time and persistent storage.
Both options use the same Management application and need 2 CPUs, 4 GiB RAM and 100 GB persistent storage. For containers, reserve that capacity in the engine’s storage and mount a persistent volume at /var. Host and container-runtime overhead is additional. Choose a native image variant for the engine; the ISO uses AMD64.
For container deployments, check the minimum Docker, Docker Desktop, Compose and Podman versions.
Management ISO VM
| Resource | Allocation |
|---|---|
CPU |
2 virtual CPUs, AMD64 |
Memory |
4 GiB; configure 4096 MiB where the hypervisor uses MiB |
Persistent storage |
One dedicated 100 GB state disk |
Boot |
Management ISO attached and first in boot order for every boot |
Network |
Reachable Management address, DNS resolution and a reachable NTP server |
The state disk holds persistent host settings, Management data, credentials, images and logs. Allow space for bundle imports and operations. Keep the disk access-controlled; it contains secrets.
Appliance Nodes
Choose Standalone for one node or Three nodes for three nodes. Standalone has no node redundancy. Each appliance node needs UEFI firmware and three separate writable disks. Configure UEFI before the first boot and put the intended OS disk first in the boot order. Secure Boot may remain disabled.
| Per appliance node | Minimum | Recommended |
|---|---|---|
Virtual CPUs |
4 |
8 |
Memory |
16 GiB |
32 GiB |
OS disk |
135 GB |
135 GB |
Database disk |
100 GB |
200 GB |
Data disk |
200 GB |
500 GB |
Configure at least 16384 MiB for the minimum memory allocation. Disk capacities are decimal GB. Workload volume and retention can require more storage than these starting values. Management discovers eligible disks and requires one distinct disk for each role. It cannot use a single disk for several roles.
Minimum and recommended sizes use the same appliance service defaults. The recommended size provides more capacity for workload and recovery activity; it does not select a separate tuning profile or guarantee a particular transaction rate.
Network and DNS Plan
Record these inputs before opening the installation wizard:
| Input | Requirement |
|---|---|
Management identity |
A public DNS name resolving to its reachable host address; HTTPS port, default |
Management internal address |
An RFC1918 address that does not overlap the Management bridge, Kubernetes networks or networks Management needs to reach. It is not the host’s LAN address. |
Cluster and nodes |
A lowercase cluster name and node names; one static IPv4 address per appliance node. Separate node DNS names are not installation inputs. |
Shared appliance subnet |
All appliance nodes and their gateway on one subnet. The installation wizard accepts a subnet prefix from 1 to 29. |
Application endpoint |
A free load-balancer IPv4 address in the node subnet and an application hostname resolving to that address. |
Name and time services |
One to four DNS server IPv4 addresses and one to four NTP server addresses or hostnames for appliance nodes. Management host setup uses its own network and one NTP source. |
Administrative API connections |
Management derives Kubernetes API URLs as |
Proxies |
HTTP/HTTPS proxy addresses, authentication if the proxy requires it, and the required |
Management must reach every appliance node. Operator workstations must resolve and reach Management and the application endpoints they use. Configure your own DNS records for disconnected environments instead of depending on the wizard’s public sslip.io fallback names. Refer to Ports and protocols when configuring firewalls.
The application load-balancer address serves ERS and remains a required input. Initial Kubernetes configuration uses the first node’s IP; Management retains all node IPs with the cluster credentials. For later connections, it can select a reachable node after authenticating to it. This does not guarantee that an interrupted operation continues automatically after a node fails.
Release and offline Inputs
Prepare these files and network services:
-
The Management ISO and checksum, or the Management container image for your chosen deployment option.
-
A compatible appliance release bundle. Retrieve it through a configured registry or transfer its OCI archive for browser upload.
-
The matching platform-specific Talos boot artifact and checksum linked by Management after you submit the installation plan. Transfer these into the isolated environment if its browser cannot reach the download service.
-
Required DNS and NTP services inside the deployment network, plus any HSM, application or customer integration endpoints.
-
A location outside Management for the generated artifact bundle, backups and release records.
The Management ISO contains Management and its internal services. It does not make the appliance release bundle available automatically. Offline installation still needs all of the separate release and boot inputs. See Make bundles available and Bundle types.