Help with filing a [maybe] ZFS/mmap bug.

[Andriy Gapon]

on 14/11/2013 17:55 Steven Hartland said the following:
> ----- Original Message ----- From: "Andriy Gapon" <avg at FreeBSD.org>
> 
>>>> I was able recreate the problem on a 10-BETA-something-or-other
>>>> recently (I'd only been using 9 up until then).  Andriy's patches
>>>> didn't make a difference.  I haven't heard anything since reporting
>>>> back to him.
>>>
>>> I've pretty much confirmed mmap support is causing the corruption when
>>> running rrdcached as since rebuilding with mmap disabled I've had no
>>> further corruption.
>>
>> Well, this is not a _proof_, of course...
> 
> Proof that mmap is causing the corruption, not that ZFS mmap support
> is the point at fault,  as the code in rrdtool takes a different path
> so could still be an application error, but without mmap enabled I've
> seen zero corruption so...

On some days I wore a green t-shirt and sometimes it rained.  On other days I
wore a red t-shirt and it never rained (so far, of course).
Is this a _proof_ that my red t-shirt acts as a rain repellant? :-)
The correlation I won't deny.

>>> @George when you got corruption what did the files look like? I ask as
>>> here I see lots of zeros as through the file size was correct but pretty
>>> much blanked.
>>
>> Steve, could you please provide a little bit more of description of the
>> corruption that you got.  Lengths of those zeroed regions, their offsets (modulo
>> page size).
>> Anything that could establish a pattern (if any exists).
> 
> They all seem to be corrupted to the same file which when dumped using
> hexdump -C <file> results in:
> 00000000  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
> *
> 00001000  00 00 00 00 00 00 f8 ff  00 00 00 00 00 00 f8 ff  |................|
> *
> 000a8000  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
> *
> 000af000  00 00 00 00 00 00 f8 ff  00 00 00 00 00 00 f8 ff  |................|
> *
> 00113de0

It seems like the file has lots of page-sized zeroed regions in it.
I am not familiar with the rrd format...  Could this result from some natural
causes (like absence of data)?  ALthough, the first page being all zeros is
quite suspicious.

In either case, it looks very different from what George reported.

-- 
Andriy Gapon