> Memory usage tops out at 2,066,287,984 for me.
That's pretty much all there is for a normal 32bit Windows process (2GB
in user mode, a litte of which may be taken by user mode parts of the
OS, the other 2GB reserved for Kernel mode). I hear 32bit Linux
programmers get another GB of user mode memory, but I've never done
anything memory-intensive on Linux.
> Because of the way you have structured your SAVEPOINTs, the statement
> log (used to ROLLBACK TO a prior savepoint) must add at least one new
> page for each of your 500K UPDATEs.
So a RELEASEd savepoint can still take a page in the log? Is there an
approved way to work around this?
Since this happens rarely in the original context I'm thinking about
making it a temporary database as described at
https://www.sqlite.org/inmemorydb.html. When most of the cache is unused
(as it usually is) that would be (roughly) as fast as an in-memory db,
right? If so, is there any way to influence where in the filesystem the
empty temporary database is created? And, just to make sure, a temporary
database would also get an on-disk statement log, right?
Greetings,
Brendan E. Coughlan
_______________________________________________
sqlite-users mailing list
[email protected]
http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
