optimal_io_size is large causing LVM LV alignment inconsistency

Question

I'm trying to create a LVM VG and LV, but the LV is not aligned.

Here's my starting position:

% lsblk -t /dev/sdd
NAME   ALIGNMENT MIN-IO   OPT-IO PHY-SEC LOG-SEC ROTA SCHED       RQ-SIZE  RA WSAME
sdd            0   4096 33553920    4096     512    1 mq-deadline      60 128   32M
├─sdd1         0   4096 33553920    4096     512    1 mq-deadline      60 128   32M
├─sdd2         0   4096 33553920    4096     512    1 mq-deadline      60 128   32M
└─sdd3         0   4096 33553920    4096     512    1 mq-deadline      60 128   32M

Note the ALIGNMENTs are all 0 (correct).

On sdd2, I create a VG test and then a LV align-me:

% sudo vgcreate --pvmetadatacopies 2 --vgmetadatacopies 2 test /dev/6TBd1p2
  Physical volume "/dev/6TBd1p2" successfully created.
  Volume group "test" successfully created
% sudo lvcreate -L 64g -n align-me test
  Logical volume "align-me" created.

However, ALIGNMENT for test-align--me is wrong:

% lsblk -t /dev/sdd
NAME               ALIGNMENT MIN-IO   OPT-IO PHY-SEC LOG-SEC ROTA SCHED       RQ-SIZE  RA WSAME
sdd                        0   4096 33553920    4096     512    1 mq-deadline      60 128   32M
├─sdd1                     0   4096 33553920    4096     512    1 mq-deadline      60 128   32M
├─sdd2                     0   4096 33553920    4096     512    1 mq-deadline      60 128   32M
│ └─test-align--me        -1   4096        0    4096     512    1                 128 128   32M
└─sdd3                     0   4096 33553920    4096     512    1 mq-deadline      60 128   32M

I also see the following log message (repeated 4 times):

kernel: device-mapper: table: 254:6: adding target device sdd2 caused an alignment inconsistency: physical_block_size=4096, logical_block_size=512, alignment_offset=0, start=33553920

I've calling vgcreate with --dataalignmentoffset 4k and also --dataalignment with 1m and 4m but the results are the same.

I'm running:

LVM version:     2.02.182(2) (2018-10-30)
Library version: 1.02.152 (2018-10-30)
Driver version:  4.39.0

Answer 1

Symptoms

1. The lsblk -t OPT-IO value was really high (pointed out by frostschutz)

2. When running smartctl on the disk, I was seeing:

   Read Device Identity failed: scsi error unsupported field in scsi command
   

Cause

It seems that -1 is returned if an optimal IO value can't be read.

As pointed out by Steve Dee:

• 33553920 / 512 (logical sector size) = 65535
• -1 represented as a 16 bit value = 65535. (2¹⁶ = 65536)

This 33553920 value is then used by pvcreate/vgcreate to align the first PE (pe_start) as shown by:

sudo pvs -o +pe_start --units b

Work-around

Passing --dataalignment 1m to vgcreate will have pe_start = 1048576B = 1MiB.

This will ensure that pe_start is aligned with a disk sector, but the (incorrect) misalignment message will still be printed.

Root cause fix

Disabling UAS on the drive had the OPT-IO value return to 0 (consistent with my other disk drives). It also allowed smartctl to be run on the drive.

Passing --dataalignment 1m is not needed if this fix is applied.