Re: zfs mirror pool online but drives have read errors

First, make and test some backups with whatever you prefer,
zfs send, rsync, tar...

The zfs got errors, pool had enough read redundancy for
scrubs / silvering to work, drives are failing underneath.

Can also be tried in other box to confirm in isolation...

Test full drive read with...
dd if=/dev/drive of=/dev/null bs=1m conv=noerror

Test full drive write with...
dd if=/dev/random of=/dev/drive bs=1m

With camcontrol you also try standard and manufacturer
drive based sanitize / security erase / initialize functions.

These write operations may map out drive bad sectors
so the next dd read and write tests may be clean.

Then trust the drive again or not.

> It did repair 54M on the last scrub, I did another scrub today and again
> repairs are needed (only 128K this time).

More fails.

> smartctl does see the errors (but still says SMART overall-health
> self-assessment test result: PASSED ):

It's an estimation. Hardware fail is unpredictable.

> -zfs doesn't remove the drives because...

Likely.

> -Both drives are unreliable...

Likely.

> Could more expensive ssd's made a difference here ?

Maybe not, people try to hammer on cheap USB sticks
ZFS arrays thousand of write cycles for fun.

https://www.youtube.com/watch?v=7z526m1jvls

zfs set atime=off compression=on ... saves life too.

> 1200TBW

/2TB = 600 cycles

> "zfs send > imgfile"
> what would have have happened here if more and
> more read errors would occur ?

zpool failmode=wait
If the send completed ok, the image should be good.
Users are supposed to test their backups anyway.