Is the header block always the first block inserted? If so you can find it by
ordering the blocks by idnr and just taking the first one. No extra field
necessary..
I don't think there would be a performance difference at all between the two
methods and it would keep the table structure simpler, which to me is better.
Just a thought.
-Micah
On Wednesday 09 June 2004 05:00 pm, Aaron Stone wrote:
> Off the top of my head, there's a two step process here. First is that the
> top level, messgae-oriented functions will need an extra flag. Second is
> that the inner physmessage functions will need that flag, too.
>
> db_insert_message_block() and db_insert_message_block_physmessage().
>
> I don't think we should wait on 2.0 for this. We can tweak performance in
> the point releases. We should add whatever columns we'll need to support
> the functionality, though.
>
> Aaron
>
> Paul J Stevens <[EMAIL PROTECTED]> said:
> > Hi Ilja,
> >
> > The wiki looks pretty much empty so I just started playing around.
> >
> > I've been testing with a messageblks.is_header field so db_fetch_headers
> > will not retrieve the whole message before parsing.
> >
> > Looks like we may get something like a factor 2 improvement in searching
> > headers.
> >
> > I still have to figure out how to modify message insertion to use such a
> > field. Perhaps you or Aaron have some bright ideas.
> >
> > For now I've modified the database by hand using group by queries to
> > locate messageblk_idnr for the header block and a shell-script to
> > generate update queries.
> >
> > queries used:
> >
> > alter table messageblks add is_header int(1) not null default 0;
> >
> > update the messageblk table:
> >
> > #> mysql --skip-column-names -B -e "select messageblk_idnr from
> > messageblks group by physmessage_id" dbmail |awk 'BEGIN {
> > printf("\nupdate messageblks set is_header=1 where messageblk_idnr in
> > ("); } { if(NR % 200 == 0) { printf("\nupdate messageblks set
> > is_header=1 where messageblk_idnr in ("); i=0; } else {
> > printf("%s,",$1); }}'|sed 's/,$/);/' | mysql dbmail
> >
> >
> > #> mysql -B -e "select messageblk_idnr from messageblks group by
> > physmessage_id" dbmail |awk '{ if(NR % 50 == 0) { printf("\nupdate
> > messageblks set is_header=1 where messageblk_idnr in ("); i=0; } else {
> > printf("%s,",$1); }}'|sed 's/,$/);/'
> >
> > Ilja Booij wrote:
> > > Paul J Stevens wrote:
> > >> Hi all,
> > >>
> > >> I'm looking to optimize _ic_search without resorting to drastics.
> > >>
> > >> I noticed dbmsgbuf.c retrieves the full message from the database in
> > >> two functions. In both cases first the physmessage_id is queried using
> > >> the msgid, followed by a query on the messageblks table using the
> > >> physmessage_id. Two queries where one suffices. A simple but effective
> > >> improvement I think. Query syntax (left join ... using ...) checked on
> > >> mysql/postgres (dunno about oracle or ansi-sql).
> > >>
> > >> I also fixed some of the FIXME's ...
> > >>
> > >> Ilja?
> > >
> > > Looks good. I'v committed the patch.
> > >
> > >> PS. Of course searching still sucks major :-( until we get some real
> > >> header caching...
> > >
> > > true..
> > >
> > > Ilja
> > > _______________________________________________
> > > Dbmail-dev mailing list
> > > Dbmail-dev@dbmail.org
> > > http://twister.fastxs.net/mailman/listinfo/dbmail-dev
> >
> > --
> > ________________________________________________________________
> > Paul Stevens mailto:[EMAIL PROTECTED]
> > NET FACILITIES GROUP PGP: finger [EMAIL PROTECTED]
> > The Netherlands________________________________http://www.nfg.nl
