mdadm raid soft lock-ups ubuntu kernel 4.13.0-36

[Adam Hamsik]

Hi,

we're running Ubuntu 16.04.4, mdadm - v3.3 and Kernel 4.13.0-36.
We have created raid10 using 22 960GB SSDs [1] . The problem we're
experiencing is that /usr/share/mdadm/checkarray
(executed by cron, included in a mdadm pkg) results in (soft?)
deadlock - load on the node spikes up to 500-700 and all I/O operations
are blocked for a period of time. We can see traces liek these [2] in
our kernel log.

e.g. it ends up in static state like

test at os-node1:~$ cat /proc/mdstat
Personalities : [linear] [multipath] [raid0] [raid1] [raid6] [raid5]
[raid4] [raid10]
md1 : active raid10 dm-23[9] dm-22[8] dm-21[7] dm-20[6] dm-18[4] dm-19[5]
dm-17[3]
                    dm-16[21] dm-15[20] dm-14[2] dm-13[19] dm-12[18]
dm-11[17]
                    dm-10[16] dm-9[15] dm-8[14] dm-7[13] dm-6[12] dm-5[11]
dm-4[10] dm-3[1] dm-2[0]
      10313171968 blocks super 1.2 512K chunks 2 near-copies [22/22]
[UUUUUUUUUUUUUUUUUUUUUU]
      [===>.................]  check = 19.0% (1965748032/10313171968)
finish=1034728.8min speed=134K/sec
      bitmap: 0/39 pages [0KB], 131072KB chunk
unused devices: <none>

and the only solution is to hard reboot the node. What we found out is that
it
doesn't happen on idle raid, we have to generate some significant load
(10 VMs running fio[3] with 500GB HDDs.) to be able to reproduce the issue.

Anyone ever experienced similar issues? Do you have any suggestions how to
better trouble shoot this issue and maybe identify if disks or software
layer
is responsible for this behaviour

[1] http://www.samsung.com/us/dell/pdfs/PM1633a_Flyer_2016_v4.pdf
[2] https://gist.github.com/haad/09213bab1bc30a00c7d255c0bc60897b
[3] https://github.com/axboe/fio





Regards
Adam.

Adam Hamsik
00421 904 937 495
adam.hamsik at chillisys.com
haad at netbsd.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/kernel-team/attachments/20180608/1832cf7a/attachment.html>