Skip to main content
Add worker hosts running zwrm-agent to distribute VMs beyond a single server. The control plane schedules workloads across all healthy hosts.

​
Adding a host

Install the agent on each worker host: 
curl -fsSL https://releases.zwrm.eu/zwrmd/install.sh | sudo bash -s -- \
    --agent \
    --control-plane-url=https://cp.example.com:8080 \
    --region=eu-central \
    --datacenter=dc1
The agent registers automatically and begins sending heartbeats. Verify with: 
zwrm host list

​
Scheduling

The control plane uses a placement scheduler to decide where new VMs run. Two strategies are available: 
[scheduler]
  strategy = "spread"    # distribute evenly across hosts
  # strategy = "pack"    # fill hosts before using new ones
StrategyBehavior
spreadDistributes VMs evenly across all healthy hosts. Maximizes fault tolerance.
packFills hosts to capacity before using the next one. Minimizes host count.
Both strategies respect region and datacenter constraints when specified in the deployment.
Apps with local volumes are pinned to the host where the volume was created, regardless of scheduler strategy. See Volumes for details.

​
Host health monitoring

The control plane monitors host health:
  • Check interval: Every 15 seconds
  • Offline threshold: 30 seconds without a heartbeat
  • On offline: Host is marked offline, its executor is deregistered, and all running machines are marked failed
When the agent reconnects and resumes heartbeats, the host is automatically re-registered.

​
Host management

# List all hosts with status and capacity
zwrm host list

# Get detailed info for a host
zwrm host status <host-id>

​
Host drain

Drain a host to gracefully migrate its workloads before maintenance: 
# Start draining a host
zwrm host drain <host-id>

# Check drain progress
zwrm host drain-status <host-id>

# Cancel a drain
zwrm host undrain <host-id>

# Return a host from maintenance
zwrm host activate <host-id>

​
Drain process

  1. Mark the host as draining (no new VMs scheduled)
  2. For each VM on the host:
    • Select a target host (excluding the draining host)
    • If the VM has volumes, transfer the ext4 image to the target via gRPC streaming
    • Destroy the original VM
    • Start a replacement VM on the target host
    • Wait for health check to pass
  3. If any migration fails, rollback: destroy the replacement, restore original volume mounts
Volume transfers are one-time copies, not replication. After drain, the volume lives on the new host and subsequent deploys pin to that host.

​
Image distribution

When the scheduler places a VM on a remote agent host, the agent pulls the app image from the control plane’s built-in image registry. Images are content-addressed (SHA256) and cached locally on each agent with LRU eviction. See Host Agent for cache configuration.