Il giorno lun 22 apr 2019 alle ore 19:40 Reindl Harald <h.reindl@xxxxxxxxxxxxx> ha scritto: > you hardly can avoid the read on the source disks anyways Reading from source disks is not an issue. > in case of RAID10 i would just pull half of the disks (you notice when > it was the wrong "pair") and move on with rebuild on half of the array > with new ones, in case everything goes wrong the rmeove ones hold the > complete dataset > > for other than RAID1/RAID10 you have no other option than go ahead and do it A normal raid rebuild is done online, during the rebuild the server is online and data are changed on all disks. If something should fail, one of the disk i've previously removed would be automatically out-of- sync and would hold old data. In example: i remove one disk from a raid1 and I add a brand new disk. The raid is degraded but still running during the rebuild. Now, as the server is online, users will change "file1" contents, then the source disk fails. The rebuild is stopped, I have to start from the previous disk that i've removed, but "file1" stored on that disk is different (or unexistant). Thus, I'm out of game. This is what I don't want, that's why i've said to do the rebuild with the array unmounted. Noone will be able to change data on it, thus one of the older disk that i'm replacing will be perfectly valid in case of failure during a rebuild. But the server is very old, the mdadmin version is very old. Can I sart a recent live distro and being able to rebuild the existing array (one disk one by one) by using a recent mdadmin and obviously still bein able to boot the new array with the new disks or the new mdadmin will mess the old array in some way ?
