I've been thinking about this for a while, and especially with the query time escalation code, have found that the queries which do full header searches are incredibly slow (for the obvious reason that they cannot use the index).
I have two thoughts here. First, our imap search algorithm is based on merging multiple independent queries. We should do some planning so that a search that includes a free-text headervalue query does all of the other queries first, finds the possible result set, then chunks out the headervalue scan. For example, I often do a search limited by date: SEARCH SINCE SINCE 1-Feb-2007 FROM dbmail This currently breaks into two queries that search all messages, the latter of which has a "WHERE headervalue LIKE '%dbmail%'", and takes about 25 seconds and boatloads of CPU when running against my Inbox of 30,000 messages. Restricting it to the most recent few thousand would make it much much faster. Second thought is http://sphinxsearch.com/ Aaron _______________________________________________ Dbmail-dev mailing list Dbmail-dev@dbmail.org http://twister.fastxs.net/mailman/listinfo/dbmail-dev
