buffindexed: expireover needs _very_ much memory

[Katsuhiro Kondou]

In article <19991013063212.B25866 at CIS.FU-Berlin.DE>,
	Heiko Schlichting <inn-bugs at fu-berlin.de> wrote;

} Hmm. I never (re)defined OV_BLOCKSIZE and therefore this value should always
} defined to 8192. Maybe my interpretation on 'len 16384' or 'len 8192' in the
} above output is wrong. It seems to be gip->len, not the blocksize. Should I
} expect this to be the same for all buffers?

No, the size may differ depending on where mmap is used.
That's system dependent.  I misunderstood your result :<

} 'noexpireover' for news.daily and tried expireover by hand today. But if
} you think it could help to clear/reinitalize the overview buffers, I'll

I don't think your buffers are corrupted from seeing your
info, but can you recreate all buffers to see if it'll be
ok?  If the problem still exists, I'll tend to modify the
code not to use mmap for some case.

} do that after I found how to rebuild the overview database from the CNFS
} spool. I know that rebuilding overview was possible in INN 1.x & 2.2 with
} expireover -a but this program does not have an "-a" option in INN 2.3 and
} I didn't succeed in finding it's replacement.

To recreate overview data, you can do it with 'makehistory'.
'-I -O -x -l 0' would be useful for you.  '-T' may needed
if you don't have enough space for pathtmp.  For buffindexed,
you need to recreate all buffers with 'dd' before doing that.
-- 
Katsuhiro Kondou