On Thu, Jan 24, 2013 at 12:58 PM, Gerald Combs <gerald@xxxxxxxxxxxxx> wrote:
> On 1/23/13 7:46 PM, Jeff Morriss wrote:
>> Hi Gerald,
>>
>> The buildbot's been generating a lot of fuzz failure bugs recently--none
>> of which I've been able to reproduce in any of the supported trunks
>> (Valgrind is silent too).
>>
>> Any idea what's going on?
>>
>> The last fuzz failure log says this (which doesn't make sense):
>>
>>> /home/wireshark/menagerie/menagerie/9460-test.pcap: (-nVxr) (-nr) OK
>>>
>>> ls: cannot access
>>> /dev/shm/buildbot/clangcodeanalysis/menagerie-fuzz/fuzz-*.pcap: No
>>> such file or directory
>
> In order to improve performance and reduce the I/O load I've been using
> /dev/shm (which uses tmpfs) as the temporary directory for fuzzing.
> However, it looks like the contents of /dev/shm disappear and reappear
> periodically. The buildbots complain about missing files but when I
> check they're there. I'm not sure what's causing it. Maybe a cron job or
> daemon periodically clobbers and un-clobbers /dev/shm?
>
> I moved fuzzing to its own mount point. Hopefully Buildbot will stop
> complaining.
Another thing perhaps worth trying is to invert the loop over passes
with the loop over files. Instead of fuzzing each file once and then
starting over, it would do the first file n times, then the second
file n times, etc. which would mean that each file is hot in cache for
all but the first pass.
Cheers,
Evan