On Wed, Dec 08, 2021 at 05:21:51PM +0000, Rory Campbell-Lange wrote:

On 08/12/21, Roman Mamedov (rm@romanrm.net) wrote:

quoted

On Wed, 8 Dec 2021 12:40:14 +0000
Rory Campbell-Lange [off-list ref] wrote:

quoted

We're trying to upgrade the disks in a btrfs pair, and I have successfully replaced one of them using btrfs replace. I presently have 

Label: 'btrfs-bkp'  uuid: da90602a-b98e-4f0b-959a-ce431ac0cdfa
	Total devices 2 FS bytes used 700.29GiB
	devid    2 size 2.73TiB used 1.73TiB path /dev/mapper/cdisk4
	devid    3 size 2.73TiB used 1.75TiB path /dev/mapper/cdisk2

I'd like to get rid of cdisk2 and replace it with a new disk.

However I'm unable to mount cdisk4 (the new disk) in degraded mode to allow me to similarly replace cdisk2 as I previously did for cdisk3. Is this because some of the data in only on cdisk2? If so I'd be grateful to 
know how to ensure the two disks have the same data and to allow cdisk2 to be replaced.

Looks like you need to ensure everything is RAID1 first:

You definitely need these two:

quoted

  btrfs balance start -dconvert=raid1,soft /bkp
  btrfs balance start -mconvert=raid1,soft /bkp

but not this one:

quoted

  btrfs balance start -sconvert=raid1,soft /bkp

It might warn you about operating on system chunks, but I believe this still
needs to be done. 

I wasn't able to run system chunks (-s) on btrfs 4.20.1-2 (debian) without
forcing it:

    ERROR: Refusing to explicitly operate on system chunks.
    Pass --force if you really want to do that.

-m implies -s.  In normal use, there is never a reason to have different
profiles for metadata and system, so balance refuses to do it.

-s is useful to set up some scenarios for developer testing (though even
that could be achieved with -mconvert=...,vrange=... without needing a
special option).

The manual could be clearer on this.

Happily everything worked fine after running the data and metadata balances.

Thanks very much indeed for the advice.

Rory