Context

In v1.7, Longhorn announced the first version of v2 volume online rebuilding feature. It works but not as perfect as the v1 part, especially in the efficiency perspective: v2 only supports full rebuilding now. It cannot validate then reuse any intact snapshots or valid chunks, which caused lots of unnecessary data transfer during rebuilding.

Improvements

1. The snapshot Checksum feature not only helps figure out corrupted/bit-rot replicas but also makes intact snapshot reusage possible. [FEATURE] v2 volume supports snapshot checksum #8666
   • To support the fast rebuilding workflow, Longhorn and SPDK should somehow automatically calculate and store checksum once a snapshot is created.
2. We can reuse the intact snapshots in failed replicas when Longhorn can verify them.[IMPROVEMENT] v2 volume supports delta replica rebuilding based on snapshot checksum #8771
   • The control plane needs to check and compare snapshot checksum between healthy replicas and reused failed replicas, then skip data transfer for the intact snapshots and directly add into the rebuilding replica snapshot chain (by changing the lvol parents back and force). This feature is called fast rebuilding in v1. [FEATURE] Support v2 volume delta replica rebuilding based on snapshot checksum: data plane and control plane #9488
   • Some lvols may contain some intact chunks but not all. Longhorn requires a new SPDK API to verify and skip the intact part then rebuild the remaining. [FEATURE] Support v2 volume delta replica rebuilding based on snapshot checksum: SPDK part #5573