When using Azure Spot VM Elastigroups, the health check settings include an option for VM State. This health check monitors VM provisioning and power states to determine if the VM is healthy or unhealthy and to take recovery actions when necessary.

Azure does not have an equivalent of AWS EC2 instance-level health checks.

The VM State health check works by fetching the VM provisioning state every 30 seconds. If there is a change, the power state is checked to prevent hitting Azure rate limits.

Health check logic

• VM is considered unhealthy if:

  • The VM failed to provision or remains stuck in the creating state beyond the defined grace period.

  • The VM is in deallocating state.

  • The VM is in deallocated state.

• VM is considered healthy if:

  • The VM provisioning is successful.

When a VM enters deallocating or deallocated states—both potential signs of an Azure-initiated interruption—Elastigroup triggers a recovery, regardless of whether auto-healing is enabled or disabled.

Use cases

• Provisioning failure: For example, if a VM image is corrupted, the VM fails to provision and is therefore marked unhealthy.

• Provisioning delay: If a custom data script runs too long, the VM remains in provisioning for an extended period, resulting in the VM being marked unhealthy.

Additional details about Azure VM states:

• Provisioning State: Reflects the status of control-plane operations initiated by the user. These states are separate from the power state. Common states include:

  • Create

  • Update

  • Delete

  • Deallocate

• Power State: Represents the last known operational state of the VM. Common states include:

  • Starting

  • Running

  • Stopping

  • Stopped

  • Deallocating

  • Deallocated

Outcome

The VM State health check ensures that Elastigroup can proactively identify and replace VMs that fail provisioning or get interrupted by Azure before they affect workload reliability.