Thread (33 messages) 33 messages, 2 authors, 2020-07-08

Re: Assemblin journaled array fails

From: Michal Soltys <hidden>
Date: 2020-06-16 11:11:24

On 6/9/20 11:51 PM, Michal Soltys wrote:
On 20/06/09 20:36, Song Liu wrote:
quoted
On Tue, Jun 9, 2020 at 2:36 AM Michal Soltys [off-list ref] wrote:
quoted
On 6/5/20 2:26 PM, Michal Soltys wrote:
quoted
On 6/4/20 12:07 AM, Song Liu wrote:
quoted
The hang happens at expected place.
quoted
[Jun 3 09:02] INFO: task mdadm:2858 blocked for more than 120 
seconds.
quoted
quoted
quoted
[  +0.060545]       Tainted: G            E
5.4.19-msl-00001-gbf39596faf12 #2
[  +0.062932] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs"
disables this message.
Could you please try disable the timeout message with

echo 0 > /proc/sys/kernel/hung_task_timeout_secs

And during this wait (after message
"r5c_recovery_flush_data_only_stripes before wait_event"),
checks whether the raid disks (not the journal disk) are taking IOs
(using tools like iostat).
No activity on component drives.
To expand on that - while there is no i/o activity whatsoever at the 
component drives (as well as journal), the cpu is of course still 
fully loaded (5 days so far):

UID        PID  PPID  C    SZ   RSS PSR STIME TTY          TIME CMD
root      8129  6755 15   740  1904  10 Jun04 pts/2    17:42:34 mdadm 
-A /dev/md/r5_big /dev/md/r1_journal_big /dev/sdj1 /dev/sdi1 
/dev/sdg1 /dev/sdh1
root      8147     2 84     0     0  30 Jun04 ?        4-02:09:47 
[md124_raid5]
I guess the md thread stuck at some stripe. Does the kernel have
CONFIG_DYNAMIC_DEBUG enabled? If so, could you please try enable some 
pr_debug()
in function handle_stripe()?

Thanks,
Song
Massive spam in dmesg with messages like these:

[464836.603033] handling stripe 1551540328, state=0x41 cnt=1, pd_idx=3, 
qd_idx=-1
                 , check:0, reconstruct:0
[464836.603036] handling stripe 1551540336, state=0x41 cnt=1, pd_idx=3, 
qd_idx=-1
                 , check:0, reconstruct:0
[464836.603038] handling stripe 1551540344, state=0x41 cnt=1, pd_idx=3, 
qd_idx=-1
                 , check:0, reconstruct:0
<cut>
So what should be the next step in debugging/fixing this ?
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help