Yes - every folder that I then go into after wiping the indexes, requires a full resync between client (Eudora) and dovecot of the headers (not downloading the message bodies). The hard drive on server works pretty hard for a while to resync up especially if there are a fair few messages, but after that initial performance hit the performance is good. Obviously it only reindexes the folders as they are opened.
It could just be a slow client, but I doubt it given the level of hard drive activity that goes on on the server. At a guess, it would be re-indexing perhaps, 30 mails/second, and it takes a while with 3000 messages in a (trash) folder. I hate to think how long it would take with 100 clients all resyncing at the same time after an index wipe or migration from an existing system to a new dovecot installation...
reuben
At 04:36 p.m. 5/12/2004, you wrote:
On 3.12.2004, at 07:19, Reuben Farrelly wrote:
What I would muchly like to see soon, is a means of forcing the re-indexing to occur at a convenient time, rather than when the Maildir is first read. Ideally also as part of this, a way of manually checking the validity of the index files and then recreating them as required.
It'd make testing of the -test releases much easier, as I could wipe all the indexes and recreate them straight away without having the performance hit when the users first re-read their mailboxes.
Is the performance hit so large with 1.0-test anymore? 0.99.x would re-read all messages and cause lots of I/O, but 1.0-tests try to delay reading as long as possible.