How to Set Up High Availability and Failover in Apache CloudStack

Configure the server's out-of-band management interface

Start on the physical server's management controller. On Dell hardware this is normally iDRAC; other vendors may provide iLO, IPMI, or another BMC interface. Enable IPMI-over-LAN or the equivalent remote power-management feature and assign the management controller a stable IP address.

If you are using Dell iDRAC, the exact menu varies by generation, but the setting is typically under iDRAC Settings → Connectivity → Network → IPMI Settings.

Configure Host HA for the KVM cluster

Open Infrastructure, select your KVM cluster, and open its HA configuration. Choose the KVM Host HA provider for the cluster. The provider performs host health checks and uses neighboring KVM hosts to help determine whether workload disks are still active before CloudStack attempts recovery or fencing.

Enable HA for the cluster and participating hosts

After selecting the provider, enable Host HA at the cluster level and make sure the intended hosts are included. Host HA is opt-in: a provider must be configured and HA must be enabled before CloudStack manages the hosts through the HA state machine.

Configure OOBM for each KVM host

Open the details page for the first KVM host and configure Out-of-band Management. For an IPMI 2.0 BMC, use CloudStack's IPMITOOL driver and enter the BMC address, port, username, and password. Repeat this configuration for every host that will participate in Host HA.

Enable and verify out-of-band power management

After saving the OOBM configuration, enable it for the host. CloudStack must be able to read the server's power state before Host HA can reliably recover or fence that host. Repeat the process for the second KVM host and any additional participating hosts.

If you want to test IPMI independently of CloudStack, run the command from the Management Server. Install ipmitool first if it is not already installed.

sudo apt update
sudo apt install -y ipmitool

sudo ipmitool -I lanplus \
  -H 192.168.10.11 \
  -U cloudstack \
  -a chassis power status

The -a option prompts for the BMC password instead of placing it directly in your shell command. Replace the example BMC address and username with the values from your environment.

Create a compute offering for HA workloads

Open Service Offerings → Compute Offerings and create a new offering for the workloads that should use CloudStack instance HA. Choose the CPU, memory, CPU speed, and other resource limits that make sense for your lab or application.

Enable high availability on the compute offering

In the compute-offering form, enable the High Availability option before saving. Instances deployed with this offering will be marked for CloudStack instance HA.

Deploy an instance with the HA compute offering

Create a new instance using the template prepared in Guide 05. During the compute-offering step, select the HA-enabled offering you just created. Finish the normal deployment workflow and wait for the instance to reach the Running state.

Test the HA workflow and verify recovery

Confirm which KVM host is currently running the HA-enabled instance. Before testing anything, verify that the second host is Up, has spare CPU and memory capacity, can access the same NFS primary storage, and has working OOBM.

In a lab, simulate an unexpected host failure from outside CloudStack—for example by powering off the compute server through its BMC. Then monitor the host HA state, events, alerts, and the HA-enabled instance from the CloudStack UI.

Do not expect recovery to happen at one fixed number of minutes. Host HA performs multiple checks before it decides that a host can be safely recovered or fenced, and the timing depends on the HA configuration and the failure scenario. CloudStack is intentionally conservative because it must avoid starting a second copy of an instance while the original host might still be writing to the same disks.