While running an import to wcqs instances they slowly start falling off the network. At least one instance reports the following in dmesg, suggesting problems with disks:
[3572611.930811] INFO: task jbd2/dm-0-8:535 blocked for more than 120 seconds.
[3572611.937776] Not tainted 4.19.0-17-amd64 #1 Debian 4.19.194-2
[3572611.944134] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[3572611.952129] jbd2/dm-0-8 D 0 535 2 0x80000000
[3572611.952130] Call Trace:
[3572611.952132] __schedule+0x29f/0x840
[3572611.952134] ? bit_wait_timeout+0x90/0x90
[3572611.952134] schedule+0x28/0x80
[3572611.952135] io_schedule+0x12/0x40
[3572611.952136] bit_wait_io+0xd/0x50
[3572611.952137] __wait_on_bit+0x73/0x90
[3572611.952138] out_of_line_wait_on_bit+0x91/0xb0
[3572611.952139] ? init_wait_var_entry+0x40/0x40
[3572611.952164] jbd2_journal_commit_transaction+0x144f/0x1840 [jbd2]
[3572611.952169] kjournald2+0xbd/0x270 [jbd2]
[3572611.952171] ? finish_wait+0x80/0x80
[3572611.952173] ? commit_timeout+0x10/0x10 [jbd2]
[3572611.952175] kthread+0x112/0x130
[3572611.952177] ? kthread_bind+0x30/0x30
[3572611.952179] ret_from_fork+0x1f/0x40