Description
System information
| Type | Version/Name |
|---|---|
| Distribution Name | FreeBSD |
| Distribution Version | 14-STABLE |
| Kernel Version | 14.3-STABLE |
| Architecture | amd64 |
| OpenZFS Version | zfs-kmod-2.2.7-FreeBSD_ge269af1b3 |
Describe the problem you're observing
Consider a pool with all top level vdevs being either single disks or mirrors.
Consider that some vdevs have been added some time (and some data) after the pool was originally created.
When we attach a disk to one of the newer top-level vdevs (either via zpool attach or zpool replace), ZFS initiates a resilver that involves all blocks that have even been written.
This is because spa_vdev_attach calls vdev_dtl_dirty(t = DTL_MISSING, txg = TXG_INITIAL).
It should be sufficient to pass txg equal to the top-level vdev's creation time vdev_crtxg as the vdev cannot have any blocks written before that.
The problem can be worked around to a degree using sequential reconstruction (e.g., zpool replace -s) but it would be nice to avoid extra work in either case.