[rt-users] Re: RT attachment corruption on RedHat 9

[Graeme Pyle]

Thanks Iain. I'll let you know if I make any further discoveries. I 
really appreciate your help!

Graeme

Iain Price wrote:
> Graeme Pyle wrote:
> 
>> Howdy guys,
>>
>> I've been reading your postings from a few weeks back relating to 
>> attachment corruption on RedHat 9.
>>
>> Have you managed to nail-down the problem yet? I'm having exactly the 
>> same problems you're experiencing, and also only on RH9 - my Gentoo 
>> box is fine. LANG=C doesn't completely fix the problem.
>>
>> Seems like I should bite the bullet and nuke the RH9 box...
>>
>> Thanks,
>> G
> 
> 
> Hi, firstly as mentioned later, there is an issue with mod_perl 2 that 
> ships with redhat 9 choping form submits off prematurely.  I recommend 
> you install apache 1.3 as modperl 1 (from source probably safest) 
> however there is also this LANG=C problem.  I've not entirely nailed 
> this one down yet, certainly setting LANG=C everywhere has some positive 
> effect... however (at the last point in my testing) attachments would 
> work correctly after i 'watched' a crontab restart the task at 8pm one 
> evening, but theres a restart every morning at 3am for the web server 
> (old issue with RT clogging up with session data i think, dunno if its 
> still a problem) but anyway hours later the attachments are corrupted 
> again.
> 
> The modperl bug symptom is that large (>10k) attachments just are 
> corrupt/truncated.   This other LANG=C bug corrupts only binary (8bit / 
> topbit set chars) attachments, of any length.  If you have the same 
> problem as me, run
> for a in 0 1 2 3 4 5 6 7 8 9 a b c d e f; do for b in 0 1 2 3 4 5 6 7 8 
> 9 a b c d e f; do printf \\x$a$b; done; done >stupid.doc
> which may not be the cleanest way to do this but it creates a 256 byte 
> file of every char code.  The LANG=C issue corrupts it to exactly 384 
> bytes (128 7bit chars and 2*128 bytes of top bit chars)
> 
> Since theres no difference  between my 8pm test cron job and my 3am one 
> for restarting apache/RT i must presume some external force is reverting 
> the system to a non LANG=C style operation (although /proc/pid/environ 
> still shows everything as i expect).  Its really not clear what this is 
> at the moment, a scheduled task or maybe just something about children 
> forking in certain ways.  My next intention is to be to write an 
> automated (every 15 min) attachment test (yuk) and see if the break 
> occurs predictably, and if this sheds light on the cause.
> 
> That all said, with the death of RHL anyway, we'll be migrating our 
> boxes here away (SuSEwards it seems) from RH and even if we weren't I 
> was about to recommend removing RH9 from production for at least 6 more 
> months.... goodness knows what else is lurking under the hood (theres 
> something about threading in RH9 also that breaks iPlanet LDAP server i 
> think)...
> 
> If you want to hang around i might find something out eventually (i'm 
> off work this week) but my proposal to my boss is to reinstall this box 
> early on in our migration process and maybe i can slip out of having to 
> fix this fun.
> 
> If you go at it before me, good luck, i would be interested to know 
> anything you discover no matter how trivial it seems, atm i know a lot 
> about the problem but not entirely what causes it to happen, if i can 
> get enough info together i might be able to hypothesise and test (which 
> is always better than blindly stumbling around with automated poking 
> tools :P)
> 
> Iain
> 
> 
> 
> 
> 
> 
>