Re: Xfs Access to block zero exception and system crash

[Sagar Borikar <sagar_borikar@pmc-sierra.com>]

[-- Attachment #1: Type: text/plain, Size: 2363 bytes --]



Eric Sandeen wrote:
> Sagar Borikar wrote:
>   
>> Sagar Borikar wrote:
>>     
>>> Copy is of the same file to 30 different directories and it is
>>>       
>> basically
>>     
>>> overwrite.
>>>
>>> Here is the setup:
>>>
>>> It's a JBOD with Volume size 20 GB. The directories are empty and this
>>> is basically continuous copy of the file on all thirty directories.
>>>       
>> But
>>     
>>> surprisingly none of the copy succeeds. All the copy processes are in 
>>> Uninterruptible sleep state and xfs_repair log I have already attached
>>>       
>>> With the prep. As mentioned it is with 2.6.24 Fedora kernel.
>>>       
>> It would probably be best to try a 2.6.26 kernel from rawhide to be sure
>> you're closest to the bleeding edge.
>>
>> <Sagar> Sure Eric but I reran the test and I got similar errors with
>> 2.6.24 kernel on x86. I am still confused with the results that I see on
>> 2.6.24 kernel on x86 machine. I see that the used size shown by ls is
>> way too huge than the actual size. Here is the log of the system
>>
>> [root@lab00 ~/test_partition]# ls -lSah
>> total 202M
>> -rw-r--r--  1 root root 202M Jul  4 14:06 original ---> this I sthe file
>> Which I  copy.
>> drwxr-x--- 65 root root  12K Jul  6 21:57 ..
>> -rwxr-xr-x  1 root root  189 Jul  4 16:31 runall
>> -rwxr-xr-x  1 root root   50 Jul  4 16:32 copy
>> drwxr-xr-x  2 root root   45 Jul  6 22:07 .
>>     
>
> It'd be great if you provided these actual scripts so we don't have to
> guess at what you're doing or work backwards from the repair output :)
>   
Attaching the scripts with this mail.
>   
>> dmesg log doesn't give any information. Here is XFS related
>> info:
>>
>> XFS mounting filesystem loop0
>> Ending clean XFS mount for filesystem: loop0
>> Which is basically for mounting XFS cleanly. But there is no exception
>> in XFS. 
>>     
>
> and nothing else of interest either?
>   
Not really. That's why it was surprising. Even after setting the 
error_level to 11
>   
>> Filesystem has become completely sluggish and response time is increased
>> to 
>> 3-4 minutes for every command.  Not a single copy is complete and all
>> the copy processes are sleeping continuously. 
>>     
>
> And how did you recover from this; did you power-cycle the box?
>   
There was no failure. Only the processes were stalled. System was 
operative.
> -Eric
>   

[-- Attachment #2: copy --]
[-- Type: text/plain, Size: 50 bytes --]

#! /bin/sh

while [ 1 ]

do
cp -f $1 $2
done






[-- Attachment #3: runall --]
[-- Type: text/plain, Size: 189 bytes --]

#! /bin/sh

for i in 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
do
	
mkdir -p testdir_$i	
./copy testfile testdir_$i &
rm -Rf testdir_$1/testfile
./copy testfile testfile_$i &
done