Located the problem. An old fixme/too.
Not so easy to fix though :-(
The strategy currently used for searches is correct, but sub-optimal.
Searches can be very complex or very simple.
a simple flat query:
x search charset utf-8 text "charsettest" 1
means:
search for substring "charsettest"
AND
search for message id 1
The current strategy is to handle all search-keys separately and combine
the results afterwards. This approach scales - in terms of correctness -
very well even for very complex search trees (a AND b or C or D not E
and F) but leads to very poor performance for simple statements like above.
In the above example this leads to:
search *all* messages in the mailbox for the substring
search for the message with the specified message-id
merge both results.
Which is very suboptimal to say the least.
It would be much better to re-order the search-keys on the same level
where possible (left-hand and right-hand of AND operators). Scan for
cheap searches like sequence sets (like message ids or UIDS), and use
those to speed up the expensive search-keys (like body searches) through
a feed-forward mechanism.
Introducing this for left-hand/right-hand arguments of AND operators
seems safe and if done right, and without side-effects.
Let's see if I can come up with a solution.
--
________________________________________________________________
Paul J Stevens pjstevns @ gmail, twitter, skype, linkedin
* Premium Hosting Services and Web Application Consultancy *
www.nfg.nl/[email protected]/+31.85.877.99.97
________________________________________________________________
_______________________________________________
DBmail mailing list
[email protected]
http://mailman.fastxs.nl/cgi-bin/mailman/listinfo/dbmail
