[Pvfs2-users] corrupted output

[Chris Poultney]

Hi Sam-

I've tested on Fedora 12 (kernel 2.6.32.26-175) and Ubuntu 10.10 (kernel 
2.6.35-24) with the same result on both.

Michael Moore has also been helping with this - I haven't been copying 
the group as it's been a lot of details. But I will when we figure out 
the issue!

-crispy


On 12/20/2010 11:31 AM, Sam Lang wrote:
>
> Hi Chris,
>
> This looks like it might be an issue with writev in the kernel.  Are you running RHEL5 by chance?
> -sam
>
> On Dec 15, 2010, at 8:39 AM, Chris Poultney wrote:
>
>> Sure - it's attached. I ran exactly the code I posted earlier.
>>
>> -crispy
>>
>>
>> On 12/14/2010 07:25 PM, Sam Lang wrote:
>>>
>>> Hi Chris,
>>>
>>> Would you be willing to run your test with strace and post the output?  That will help us figure out what read/write system calls PVFS is getting from your app.
>>>
>>> Thanks,
>>> -sam
>>>
>>> On Dec 14, 2010, at 4:29 PM, Chris Poultney wrote:
>>>
>>>> All-
>>>>
>>>> I'm having a problem where files written to a PVFS volume overwrite the beginning of the file after some output has already been generated, as if lseek() had been called mid-write. The overwriting behavior does not show up on a regular ext3 volume. Sample code is attached. I'd like to know if anyone can duplicate this behavior, because it looks like a bug.
>>>>
>>>> Essentially what happens is this: given a large (1024+ character) string, I write some portion of the string using a C++ ostream, then write another line:
>>>>
>>>> out<<   st.substr(0, N);
>>>> out<<   "done";
>>>>
>>>> If N<   1024, everything is fine. If N>= 1024, "done" is written at the beginning of the file, overwriting what was there before.
>>>>
>>>> I'd love to figure this out!
>>>>
>>>> Cheers,
>>>> -crispy
>>>>
>>>> <pvfstest.cpp>_______________________________________________
>>>> Pvfs2-users mailing list
>>>> Pvfs2-users at beowulf-underground.org
>>>> http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
>>>
>> <strace.out>
>