Live migration

Live migration

Move running droplets between compatible nodes within a region.

Live migration transfers a running droplet to another node in its region while preserving its resource identity, attached replicated volumes, and network configuration. The operation requires compatible compute and reachable storage.

Eligibility

Both nodes must be healthy, in the same region, and compatible with the droplet's CPU profile and hypervisor configuration. The destination must satisfy placement requirements and have capacity for the full droplet. Attached replicated volumes must be accessible from the destination.

Droplets with local-only disks or exclusive passthrough devices cannot use this workflow. A cold move requires a planned stop. Cross-region moves use backup and recovery.

Start a migration

Open Droplet → Placement → Migrate, select a destination or allow Infra to choose one, and review the compatibility checks. The preview shows source and destination nodes, reserved memory, volume reachability, and policy constraints.

Select Start migration. The returned operation locks conflicting actions such as resize, stop, or delete. Metrics and journal access remain available. A second placement change returns operation_conflict until the lock is released.

Transfer lifecycle

01ValidateCheck compatibility
02ReserveHold target capacity
03TransferCopy tracked pages
04CommitTransfer execution
05Clean upRelease source state
Cancellation supportedSource remains authoritative before commit.
Cancellation unavailableInspect committed placement and cleanup.

Migration validates compatibility, reserves destination capacity, transfers tracked memory, commits execution, and cleans up source state. Cancellation is supported before commit.

Each phase reports its own progress. Transferred bytes do not predict a completion time.

Validation checks source execution, destination compatibility, storage access, and current policy. Reservation then holds destination resources while the source continues running. Both phases permit cancellation.

During transfer, Infra copies memory and tracks changed pages. Cancellation can still leave the guest running on its healthy source while temporary target state is cleaned up.

Commit transfers execution and network ownership. Cancellation is unavailable from that boundary through cleanup because independent source and destination execution would be unsafe. Cleanup removes source reservations and temporary state after authoritative placement is established.

Page tracking identifies memory changed during the transfer. A write-heavy guest can dirty memory faster than the migration path transfers it. Infra then reports non-convergence instead of treating transferred bytes as completion.

During cutover, execution can pause and network paths can reconverge. Applications must tolerate transient latency and retry interrupted requests. Use a load balancer and multiple replicas when an application requires continuity through host changes.

Observe progress

Open the migration row on the source node, destination node, or droplet. Review bytes transferred, remaining dirty memory, transfer rate, phase, and the journal. Progress is not a time estimate; repeated memory pages can increase transferred bytes without reducing remaining work at the same rate.

The droplet's observed node changes after placement commits. Cleanup can continue after the destination begins execution. The operation is Succeeded only after its required cleanup completes.

Failure and recovery

Before commit, a failed or canceled migration leaves the droplet on its source when the source remains healthy. Destination reservations are released after cleanup. The journal records whether the failure occurred before or after commit.

After commit, Infra reconciles placement against the active hypervisor and storage lease. It does not start a second copy to resolve an ambiguous result. An operation can remain Reconciling until authoritative state is available. Follow Troubleshooting if operator action is required.

Migrate a fleet

Use drain for host evacuation or rebalancing for capacity redistribution. Those workflows create individual migrations and report partial completion. Successful moves are not reversed when another member cannot migrate.

Migration controls documents bandwidth ceilings, convergence timeouts, and pause policy. These settings constrain an eligible transfer; they do not remove CPU, memory-backing, device, or storage requirements.