Opera Memory Fix



Hey, so some heap-shot action has revealed some memory culprits, and
Opera was being one (never dumping the file contents of the index
after indexing) I committed r4210 which appears to fix it on my
install, but I want to make sure that this doesn't crash anyone, or
break the backend in different opera versions. So, please help me
mobilize and test this patch so we can look at a release!

p.s. Also, our in-memory hashtable of Guid's for the FSQ is immense,
it would be trivial to change the LuceneNameResolver to use a sqlite
db to store that data on-disk and out of memory, freeing a few megs. I
don't know a ton about the FSQ in general, and almost nothing about
the uid mapping, if this class is super-dynamic then maybe its a bad
call, but it appears that its a big build at startup, then static.
Could be a big memory saver.

Anyways, ill probably have some more awesome profiling discoveries by
the end of today, but wanted to get people testing the Opera patch so
we don't hold up the release, and get input on the FSQ thing in
general (not for this release though).


-- 
Cheers,
Kevin Kubasik
http://kubasik.net/blog


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]