blog /2026-07-28
EngineeringHow Condensa uses Changed Block Tracking to make VMware exits boring
Big-bang migrations fail because they treat moving a data center as one event. Condensa treats it as a stream: CBT-based replication in the background, cutover per tenant when you're ready.
Awanio Engineering · 2 min read
The scariest sentence in a migration plan is "and then we switch everything over on Saturday night." Condensa exists to delete that sentence.
The problem with big-bang migrations
Moving hundreds of VMs off VMware in one window means: copy terabytes of disk images, hope the copy finishes before Monday, and hope nothing changed while you copied. Every hour of copying widens the gap between source and destination, and the rollback plan usually amounts to "restore from backup."
Replication as a stream, not an event
Condensa leans on Changed Block Tracking (CBT) — the same mechanism backup vendors use against ESXi. The flow looks like this:
- Full sync, once. The initial replication copies each VM's disks to the destination — a KubeVirt-native environment — while the VM keeps running on VMware.
- Incremental sync, continuously. After the full copy, CBT tells Condensa exactly which blocks changed since the last pass. Each subsequent sync only moves the delta, so the destination trails the source by minutes, not days.
- Cutover, per tenant. When a tenant's maintenance window arrives, the final delta is applied and the workload starts on the destination cluster. Cutover time is bounded by the last increment, not by total disk size.
Transfers ride on resumable TUS uploads, so a dropped link resumes where it stopped instead of restarting a multi-terabyte copy.
Why this changes the risk profile
- The rollback plan is "do nothing." Until final cutover, the source VM is untouched and authoritative.
- The window shrinks to the delta. A tenant with a busy database might need minutes of downtime; a quiet web tier needs seconds.
- Migration becomes schedulable. Instead of one heroic weekend, operators cut over tenant by tenant across weeks — with normal staffing.
The same pipeline handles sources beyond ESXi: Nutanix, Sangfor, and Hyper-V workloads follow the same replicate-then-cutover pattern into KubeVirt.
If you're planning an exit, the Condensa documentation covers prerequisites and the installation flow — or read how it fits the wider VMware exit strategy on the main site.