bugs in expire? corrupt history?

[Paul Theodoropoulos]

At 03:23 PM 01/14/2000 , Mark Hittinger wrote:

> > -rw-rw-r--   1 news     news     1518704679 Jan 12 08:16 history
> > -rw-rw-r--   1 news     news     2082078872 Jan 12 08:16 history.pag
> > -rw-rw-r--   1 news     news     1461649725 Jan 14 08:27 history
> > -rw-rw-r--   1 news     news     2135831624 Jan 14 08:27 history.pag
>
>                                    2147483647
>
>Those .pag file sizes are so close to a 32bit signed maxint that it looks
>like some calculation is overflowing.  What kind of OS is this on and
>what is the native int size?  Are you using tagged or non-tagged history?

this is solaris 7, SPARC, 64 bit kernel, compiled with largefiles support, 
tagged-hash history. i went to tagged hash last week, after the server 
started going into swap-thrash hell during expire. the box has 384M ram. 
being an ultra 5, i can only add another 128M, unfortunately, so that would 
be too tiny a bandaid to make a significant difference...

>Could you have a conflict with innd's md5 routine and some system library
>md5 routine?

could be, but i wouldn't know where to look. it's a pretty straightforward 
install, many of the GNU programs are installed, but that's it.

>Start saving the contents of history.dir before expire so you can see
>what the dbzagain estimates would be.

will do.

>With large files turned on in Solaris and a 3.5 gig history file, my
>.hash file is around 500meg and my .index is around 700meg.  (So far
>so good).

i believe i was getting similar ratios under non-tagged as well.
-----------------------------------------------------------------
Paul Theodoropoulos                    Advanced TelCom Group Inc.
Senior Unix Systems Administrator      Internet Services Division
paul at atgi.net                          Santa Rosa, California, US
Work: http://www.atgi.net         Play: http://www.anastrophe.com