cyrus-murder problems with database corruption in the frontend/master

[Derrick J Brashear]

On Wed, 6 Apr 2005, João Assad wrote:

>>>> cyrus/mupdate[12614]: IOERROR: mapping /var/lib/imap/mailboxes.db file: 
>>>> Cannot allocate memory
>>> 
>> 
>> Resource limited memory, or are you really running out of memory?
>> 
>> Letting processes continue running in the face of an mmap failure needs to 
>> be re-examined I guess.

[]

> I've noticed that once I get the mmap error, the system will still run 
> without spitting db errors for anywhere from a few mins to a few hours. Also, 
> I never get more than one mmap error before the db becomes unnusable.

that's not shocking.  i'd still like to know why mmap fails, but, failing 
that, and since you're already getting corruption, can you in mupdate.c 
replace fatal with a function which does:

void fatal(const char *s, int code)
{
   assert(0);
   exit(code); /* shut up GCC */
}

and then give us a backtrace from the core which you will then get?