On 26/10/18 22:43, Diederik de Haas wrote: > Hi, > > I had 2 x 3TB RAID1 arrays (with 4 3TB drives), md0 consisting of sdb1 and > sdc1 and md1 consisting of sdd1 and sde1. > > My md0 was getting full, so I bought 2x8TB (sdf1 and sdg1) drives and thought > I could just add them so md0 would be 11TB in size. Apparently it doesn't work > that way and I just had 4 drives containing the same data and my md0 still was > only 3TB big. That would be a raid-10 you were thinking of - raid-0 an 8 and a 3, and then mirror your two 8,3 pairs. Personally, I wouldn't think that a good idea. > > So I figured that if I'd 'fail' and then 'remove' the 3TB drives from the > array and then enlarged the partitions/arrays to 8TB then I'd get md0 to 8TB > and then I could repurpose the 2x3TB drives. > That seemed to work, until I rebooted. Um - ouch. Did you do an madadm resize, followed by a partition resize? Because yes that should have worked. > > The issue is that mdadm still looks at sd[bc]1 for md0 instead of sd[fg]1 and > all 4 partitions have the same GUID That seems well weird. The whole point of a GUID is it is unique per device, so 4 partitions with the same guid sounds well wrong. However, looking at your output, I think you mean they all have the same *array* guid, which is correct - they're all the same array. What happens if you try to manually assemble sd[fg] into an array? b and c are both spares, so you might well get a working array from f and g. What happens if you remove sd[bc] from the computer? Will md0 re-appear? Oh - and you DID make sure that the resync onto sd[fg] was complete before you started messing about removing the other two drives? With drives that size a re-sync can take ages ... One last point. What do you get with "mdadm --version"? What's happening sounds suspiciously similar to a known bug with 3.3 or 3.4 - can't remember the details because we're now on (iirc) the verge of releasing 4.2. > > # cat /proc/mdstat > Personalities : [raid1] [linear] [multipath] [raid0] [raid6] [raid5] [raid4] > [raid10] > md1 : active raid1 sde1[0] sdd1[1] > 2929992704 blocks super 1.2 [2/2] [UU] > bitmap: 0/22 pages [0KB], 65536KB chunk > > md0 : inactive sdb1[1](S) sdc1[0](S) > 5859985409 blocks super 1.2 > > unused devices: <none> > > I've attached far more info about my drives/partitions/array in 'raid.status'. > I have no reason to think anything is wrong with md1, only included it for > completeness. > > So I'd like to know what I need to do in order for md0 to point to sd[fg]1 > partitions. Since those drives are way larger, I'm guessing I really need to > prevent some kind of syncing which I had when I first added the larger disks. > I've used and written data to those larger drives which I'd really like to > keep. > I _think_ that I could technically repartition and/or zero out the sd[bc]1 > partitions/drives and thereby 'fix' it, but I'd rather not do anything > destructive before getting more knowledgeable people's opinion. > And I'd also like learn the proper way to do it (and how I should've done it > to begin with). > > Is my initial idea at all possible with mdadm (combining 4 drives so that > 'small size' + 'large size' = total size, ie 11TB in my case)? > Or is the only (or best) way to create 2 different md devices and combine them > with LVM? > Combining "small size" with "large size" is eminently possible, as mentioned above. I just wouldn't do it. If you want to do something like that, you're better off combining your two newly redundant 3TB drives with those for md1, and creating a raid-10 or raid-6. Either of those will give you a 6TB array, or if you went to raid-5 you could have 9TB. Cheers, Wol
