So, with the latest replacement of disks in my RAIDZ2, I used zpool replace <pool> <old ID> /dev/sdx. Previously, while replacing with like-sized drives, it was not an issue (unless your replacement drives had “less space”).

But using the new 16TBs, I realised that ZFS decided to create one single honking 16TB partition (and a “partition #9” 8MB “buffer”), instead of matching the required 6TB and leaving empty space for future use, even when the pool had “autoexpand=off“.

So I should have replaced using a manually created partition instead of assigning the whole disk…

Sigh… Let’s see what we can do…

The simplest “hack”/”quick fix” I could think of was:

• force the device offline
  • zpool offline <pool> /dev/sdx
• forcibly shrink the partition, assuming:
  • you know both the starting sector (typically 2048) and ending sectors of the other “correctly sized”, existing RAIDZ2 partitions
    • use parted /dev/sdx unit s print to get the start and ending sectors of another working drive
  • run parted to shrink the first (and only) partition – note the trailing “s“ that tells parted the number is the ending sector instead of treating it as some other unit like MiB or GiB):
    • parted resize /dev/sdx 1 <ending sector>s
• force the whole drive into a degraded state by changing the drive and partition unique GUID:
  • run gdisk
  • select/enter /dev/sdx
  • key in “x” to enter the “extra functionality (experts only)” menu
  • key in “f” to randomise both the disk and partition unique GUID
  • key in “w” to quit
• (optionally) reboot (shutdown -R now) to ensure the system picks up the new information, or at least force a refresh (udevadm trigger)
• bring the device back online
  • zpool online <pool> /dev/sdx
  • zpool status should now show /dev/sdx as “DEGRADED”

2024/03/21 Update:

I got a few more new 16TBs; being the wiser, I needed to first create the same-sized partitions before specifying the replacement as the specific partition (instead of disk); so, instead of the above steps (which are unnecessary):

• create the partition (assuming brand new, empty drive): parted mkpart <name> 2048s <ending sector>s – note the trailing “s“ that tells parted the number is the ending sector instead of treating it as some other unit like MiB or GiB), where:
  • name is the name of the partition (I simply use the the ID_SERIAL from udevadm info --query=property /dev/sdx|grep "ID_SERIAL=")
  • ending sectors is the ending sector of the other partition(s) you are attempting to match the size for (use parted /dev/sdx unit s print to get the start and ending sectors of another working drive)

• replace the device with the partition
  • zpool replace <pool> <current ID shown> /dev/sdxn
• and resilvering should ensue (I don’t think there’s a way to avoid this, unless we have some zdb wizardry followed by a zpool scrub)
• if you get the “/dev/sdxn is part of active pool '<pool>'” error, you will have to clear the ZFS label info but resilvering will (still) be required
  
  • zpool labelclear -f /dev/sdxn
  • zpool replace <pool> <current ID shown> /dev/sdxn
  • resilvering should start