[Samba] Problem with AIO random read

[mikeliu]

Hello Volker,

Thanks for kindly feedback :)

> When the system is stalled, how many threads does the smbd 
>have active? /proc/pid/tasks will tell you.

At the beginning ...
pidstat -t|grep smbd
11:41:16 AM     0      5388         -    0.00    0.00    0.00    0.00     1 
smbd
11:41:16 AM     0         -      5388    0.00    0.00    0.00    0.00     1 
|__smbd
11:41:16 AM     0      5390         -    0.00    0.00    0.00    0.00     1 
smbd
11:41:16 AM     0         -      5390    0.00    0.00    0.00    0.00     1 
|__smbd
11:41:16 AM 65534      5423         -    0.03    0.26    0.00    0.28     1 
smbd
11:41:16 AM 65534         -      5423    0.02    0.20    0.00    0.22     1 
|__smbd
11:41:16 AM 65534         -      5593    0.00    0.00    0.00    0.00     1 
|__smbd
11:41:16 AM 65534         -      5594    0.00    0.00    0.00    0.01     1 
|__smbd
11:41:16 AM 65534         -      5595    0.00    0.00    0.00    0.00     1 
|__smbd
11:41:16 AM 65534         -      5599    0.00    0.00    0.00    0.00     1 
|__smbd

When the throughput drop to zero ...
pidstat -t|grep smbd
02:37:45 PM     0      5388         -    0.00    0.00    0.00    0.00     1 
smbd
02:37:45 PM     0         -      5388    0.00    0.00    0.00    0.00     1 
|__smbd
02:37:45 PM     0      5390         -    0.00    0.00    0.00    0.00     0 
smbd
02:37:45 PM     0         -      5390    0.00    0.00    0.00    0.00     0 
|__smbd
02:37:45 PM     0      5423         -    0.03    0.34    0.00    0.38     0 
smbd
02:37:45 PM     0         -      5423    0.03    0.26    0.00    0.29     0 
|__smbd


>Maybe it does not happen anymore under strace, but could you 
>strace -f that smbd while its running into the issue? 

When I executed strace on smbd,the error did not occur.(Timing ?)
For log.smbd,the highest debug level I could reproduce this error is log
level 5.


> Ah, btw, you don't happen to have sendfile/recvfile active? 
> I think those don't play well with aio.
# testparm -sv 2>/dev/null|grep -E 'receivefile|sendfile'
        min receivefile size = 0
        use sendfile = No
 
This symptom did not occur in Samba-4.0.13,but it occur after upgrading to
Samba-4.0.14.
So I use Samba-4.0.13 and patch this one "bug #10298: Reduce smb2_server
processing overhead".
This symptom can reproduce again.

Regards, 
Mike



--
View this message in context: http://samba.2283325.n4.nabble.com/Problem-with-AIO-random-read-tp4677191p4677287.html
Sent from the Samba - General mailing list archive at Nabble.com.