partiioning: sfdisk silently failing on i/o errors

[Olivier Gayot]

We got multiple private reports of partitioning failing with the sfdisk command failing silently when exit status 1.

Example:

        New situation:
        Disklabel type: gpt
        Disk identifier: $UUID_REDACTED

        Device Start End Sectors Size Type
        /dev/sda1 2048 4095 2048 1M BIOS boot
        /dev/sda2 4096 67106815 67102720 32G Linux filesystem
        An error occured handling 'disk-sda': ProcessExecutionError - Unexpected error while running command.

Another example:

        New situation:
        Disklabel type: gpt
        Disk identifier: $UUID_REDACTED

        Device Start End Sectors Size Type
        /dev/sdc1 2048 2203647 2201600 1G EFI System
        /dev/sdc2 2203648 1465145343 1462941696 697.6G Linux filesystem
        An error occured handling 'disk-sdc': ProcessExecutionError - Unexpected error while running command.
        Command: ['sfdisk', '--no-reread', '/dev/sdc', '--no-tell-kernel']

In all the reports so far, the kernel seems to report some sort of i/o errors with the relevant drive.

Looking for things such in Syslog.txt or InstallerJournal.txt just before sfdisk.

* ubuntu kernel: I/O error, dev $device, sector 0 op 0x1:(WRITE) flags [...]
* ubuntu kernel: critical medium error, dev $device, sector $sector op 0x0:(READ) [...]
* ubuntu kernel: critical medium error, dev $device, sector $sector op 0x1:(WRITE) [...]

I wonder if there is a flag we can pass to sfdisk to be more verbose. The manpage mentions the following variables:

       SFDISK_DEBUG=all
           enables sfdisk debug output.

       LIBFDISK_DEBUG=all
           enables libfdisk debug output.

       LIBBLKID_DEBUG=all
           enables libblkid debug output.

       LIBSMARTCOLS_DEBUG=all
           enables libsmartcols debug output.

       LOCK_BLOCK_DEVICE=<mode>
           use exclusive BSD lock. The mode is "1" or "0". See --lock for more details.

We can experiment and see what's manageable from the logs standpoint