Glossary

Failback

Failback is the return of workloads and data from a recovery site to the original site after an outage has been resolved. It is planned rather than urgent, and it often takes longer than the failover before it, because every change made during the outage has to be copied back first.

The four steps of a failback

  1. Fix the cause of the outage and confirm the original site is healthy.
  2. Replicate in reverse, sending every change made at the recovery site back to the original while users keep working.
  3. Cut over in a short window once both sides are nearly identical, pausing writes while the last changes are applied.
  4. Restore normal protection, including replication back to the recovery site and backups.

The last step is the one most often skipped. Until it is done, the organization runs without a standby.

What decides how long it takes

If the original data is intact, only the changes made during the outage travel back, which is hours to days of transfer. If the site was destroyed, or ransomware was the cause, the original storage may need rebuilding from nothing, sometimes in a clean environment first. Moving petabytes over a network can take weeks, which is why large recoveries sometimes ship drives.

Writes caught in the gap

With asynchronous replication, the last writes before the failure may never have reached the recovery site. If the original site turns out to hold them, someone has to decide which version of those records is correct. The size of that gap is set by the recovery point objective. Sites that run active-active avoid most of it, because the repaired site rejoins and resynchronizes.

Why failback is rarely tested

Recovery tests usually end once service runs at the alternate site, and the return trip is left for another day that seldom comes. The result is a failback procedure that has never been timed, run for the first time by a tired team after a real incident. Including the reverse replication and the final cutover in disaster recovery testing, even on a single system, shows how long the return takes and whether the original site accepts the data back.