Re: The read data is wrong from raid5 when recovery happens

[Guoqing Jiang <guoqing.jiang@linux.dev>]

On 5/30/23 10:30, Xiao Ni wrote:
> On Tue, May 30, 2023 at 10:23 AM Guoqing Jiang <guoqing.jiang@linux.dev> wrote:
>>
>>
>> On 5/30/23 10:11, Xiao Ni wrote:
>>>> May I ask if these processes write the same file with same offset? It's
>>>> insane if they do... If not, this cound be a problem.
>>> They write to different files. One process writes to its own file.
>> How big is the capacity of your array? I see the script write 100G file
>> first, then create
>> different files with 3GB size. So you probably need a array with 200G array.
>>
>> ./01-test.sh:dd if=/dev/zero of=/tmp/pythontest/file1bs=1M count=100000
>> status=progress
>>
>> Thanks,
>> Guoqing
>>
>>
> I used ssd disks and without --size. So the raid is almost 1TB

I can't reproduce it after disable write 100G file1 and decrease 
FILE_SIZE to 30
to fit my array, and iostats is on, just FYI.

localhost:~/readdata #cat /sys/block/md126/queue/iostats
1
localhost:~/readdata #./01-test.sh
mdadm: stopped /dev/md126
umount: /dev/md126: no mount point specified.
mdadm: Defaulting to version 1.2 metadata
mdadm: array /dev/md126 started.
mke2fs 1.43.8 (1-Jan-2018)
/dev/md126 contains a ext4 file system
        last mounted on /tmp/pythontest on Mon May 29 22:21:18 2023
Creating filesystem with 784896 4k blocks and 196224 inodes
Filesystem UUID: 2eb67cc5-67d0-432e-9b66-d3bc3c472d93
Superblock backups stored on blocks:
        32768, 98304, 163840, 229376, 294912

Allocating group tables: done
Writing inode tables: done
Creating journal (16384 blocks): done
Writing superblocks and filesystem accounting information: done

Mount point already exists: /tmp/pythontest
Clearing mount point!
mdadm: set sdd faulty in md126
mdadm: hot removed sdd from md126
5.15.0-rc4-59.24-default
Mon May 29 22:31:27 EDT 2023
dd start at : Mon May 29 22:31:27 EDT 2023
Mon May 29 22:31:27 EDT 2023
dd ended at : Mon May 29 22:31:27 EDT 2023
mdadm: added /dev/sdd
Mon May 29 22:31:30 EDT 2023
stresstest start  at : Mon May 29 22:31:30 EDT 2023
Test started!
start
[...]
start
Rebuild ended, testing done!
stresstest end at : Mon May 29 22:32:30 EDT 2023
Mon May 29 22:32:30 EDT 2023


Thanks,
Guoqing