Project hosts

What can change on a project host and when

Know which host edits are online, which require restart, and which require deprovision.

The rule of thumb

Some host settings are policy and can change immediately. Other settings change the underlying provider machine and need a restart or full deprovision. Treat host edits as infrastructure changes, not normal project settings.

Changes that can happen while running

Disk enlarge supports online growth for provisioned GCP and Nebius hosts using persistent storage. Ephemeral storage cannot be resized this way. Keep backups current and check both provider disk size and usable filesystem capacity afterward. If the disk grows but filesystem growth reports a warning, follow the warning's recovery instructions before treating the extra space as usable.

Access policy, per-project RAM cap, shared-pool tier, and many metadata or billing policy settings are also host record changes. They do not by themselves recreate the provider machine.

Changes that interrupt running work

Switching spot and standard pricing can be requested while a host is running, but the effective machine changes only after restart.

For managed cloud hosts, stop the host before changing CPU, RAM, machine type, or GPU selection. Apply the permitted change, then start the host and verify that it reports the requested configuration. Self-hosted connector changes follow the connector's own capabilities.

Check the UI's restart/reprovision warnings before applying a change and plan for interruption of the projects running on that host.

Changes that require deprovision

Moving a host between region or zone requires deprovision. Region and zone are provider placement decisions; CoCalc cannot mutate a running VM into another region. Back up projects, drain or move workloads, deprovision, then provision again in the new location.

Practical checklist

  1. Check whether projects are running on the host.
  2. Check whether the change is disk, pricing, instance shape, region, or zone.
  3. Back up projects before restart or deprovision work.
  4. Warn users about interruption when the change requires restart.
  5. Warn users about provider-resource deletion when the change requires deprovision.

Agent notes

For GCP and Nebius persistent storage, check online disk and filesystem growth. For spot/standard changes, expect restart. Stop a managed cloud host before editing its machine shape. Region/zone and disk-type/storage-mode changes require deprovision. Check current backups and interruption effects first.