Common tasks

Common tasks

Procedures for capacity, migration, recovery, access, and cleanup.

Use these procedures for routine infrastructure changes. Open the linked feature page for prerequisites, configuration, and failure behavior.

Add compute capacity

  1. Open Fleet → Regions and select the constrained region.
  2. Check capacity by zone and node pool, including reserved headroom.
  3. Prepare a Linux host with compatible CPU features and KVM support.
  4. Enroll the node into the intended zone and pool.
  5. Keep it cordoned while validating storage, mesh connectivity, and agent health.
  6. Uncordon it and verify that a test droplet can start on the new capacity.

Adding a node does not move existing droplets. Run a reviewed rebalance if the fleet should redistribute workloads.

Maintain a node

  1. Cordon the node to prevent new placements.
  2. Preview its drain and inspect blockers for every droplet.
  3. Resolve incompatible CPU, storage, affinity, or capacity requirements.
  4. Start the drain and follow its child migration operations.
  5. Confirm there are no running droplets and no required storage replicas on the host.
  6. Perform maintenance, validate health, and uncordon the node.

A drained compute host can still hold storage replicas. Follow the storage pool procedure before powering it off if it serves both roles.

Expand application storage

  1. Open the volume and check its attachment and backup policy.
  2. Create a recovery point before changing a production filesystem.
  3. Increase the volume size and wait for the resize operation to succeed.
  4. Rescan the block device inside the guest if the new size is not visible.
  5. Expand the partition and filesystem using the guest operating system's tools.
  6. Verify free space from the application and check backup selection.

Infra expands the block device. It does not infer the correct filesystem command for an arbitrary guest image. Volume shrinking is unsupported.

Restore a deleted or damaged service

  1. Select a completed backup with the required disks.
  2. Restore into an isolated network.
  3. Validate guest boot, filesystem integrity, and application data.
  4. Reapply credentials, firewall policy, and application configuration.
  5. Switch load balancer backends or DNS after verification.
  6. Retain the previous resources until rollback is no longer required.

Restores create new resource IDs. References in deployment tooling do not update automatically.

Grant an operator project access

Open Organization settings → Teams, select the team, and add a project binding with the Project Operator role. Confirm that the team does not also inherit a broader organization role. Use Roles and teams to review effective permissions, then test access with a separate member session.

Investigate an unreachable droplet

Check the droplet state, node heartbeat, guest console, interface addresses, firewall rules, and route configuration in that order. Compare host traffic counters with guest traffic. If the node is unreachable, use the node failure procedure before attempting a restart elsewhere.

Remove a project

Export the resource inventory and audit records required by retention policy. Delete workloads, detach and remove volumes, delete network dependencies, and expire retained backups when permitted. Release reserved addresses and remove service accounts. The project deletion review lists remaining blockers; organization nodes are outside project deletion.

Replace an application version

Follow Rolling service replacement to pin a new template revision, create overlap capacity, verify readiness, drain previous backends, and retain a compatible rollback path. The procedure separates guest creation from application admission so an accepted infrastructure operation does not send traffic to an uninitialized application.

Withdraw network access

Follow Firewall rollout and access revocation to inspect combined allow rules, preview changing selectors, and verify the applied revision on every target. Include IPv6 and established sessions in the review when the change responds to a credential exposure or compromised backend.