Hi,
we have found some issues of squatter 3.4.2 which doesn't allow us to move from 2.5.x to 3.4.2 on production.
- In 2.5.x, if you had "search_engine: squat" but no squat file in a mailbox, that would be searched via cyrus.index file.
In 3.4.x, searches returns no result if you have ""search_engine: squat" but no squat file, even for a simple subject query.
This makes it hard during a live upgrade, if you want to remove all old squat files and start from scratch with the new squatter.
Unless you can confirm that old 2.5.x squat files will be used by 3.4.x, and in this case one can just start a new full squattering to overwrite old files. - The new version of squatter is much smarter, create smaller files and is faster during searches.
The problem is that non squattered mails (recent ones) get all returned in a search command, resulting in a mix of matched mails (through squatter index) and recent mails not checked for a match.
I looked into the code and try to follow the flow via gdb during an imap search, and infact there is a piece of code ("add_unindexed" in search_squat.c) that will add all the unindexed emails all together, without filtering them through the default search engine (as 2.5.x did).
I tried to figure out how to do it but it looks cumbersome to me.
Maybe someone can help me to achieve this into the code?
BTW, we discovered 2.5.x squatter contains a big bug, creating huge squat files or sometimes going into "memory exhausted".
The main problem is that it does not consider mime types, indexing every single part, be it text or binary and treating them as text.
You can imagine that the index file is filled with useless data about images or binary attachments, and that's ther reason why it became so slow nowadays, stealing a lot of disk space.
The main problem is that it does not consider mime types, indexing every single part, be it text or binary and treating them as text.
You can imagine that the index file is filled with useless data about images or binary attachments, and that's ther reason why it became so slow nowadays, stealing a lot of disk space.
We have a patch letting it skip any part that is not "text/*" if anyone is interested: it's now finally faster and cheaper!
Thanks for any help,
Gabriele
Sonicle S.r.l. : http://www.sonicle.com
eXoplanets : https://gabrielebulfon.bandcamp.com/album/exoplanets
