If you're enabling mmap on a platform where it is intentionally not enabled by SQLite, then you probably need to get in and thoroughly verify that it's going to work correctly. You'll need to enable SQLITE_MMAP_READWRITE, obviously, but I don't think there's anything in there which makes the guarantee that SQLite will _only_ use memory-mapped access to the file.
[I think it might be reasonable to have a SQLITE_MMAP_NO_FILEIO type flag for this case, which would force a memory-mapped file to be accessed only via mmap.] -scott On Mon, May 2, 2016 at 12:19 PM, Mikael <mikael.trash at gmail.com> wrote: > Aha, great! So what I was asking for was already in the box, and a > mmap_size of say 2^63 will be fine then! > > > One reason that I asked for this was that I want to use it on OpenBSD, and > there, mmaping in Sqlite is disabled altogether, in the absence of a > unified buffer cache (UBC) in the OS. > > If I just force it on (by hacking the build script), as long as mmap_size > always is 2^63, will Sqlite access the file via memory accesses only, and > never using fread/fwrite which would lead to undefined behavior because of > the absence of a UBC? > > Thanks! > > On Tuesday, 3 May 2016, Scott Hess <shess at google.com> wrote: > > > The existing mmap functionality only maps the actual blocks associated > with > > the file. So if your file is 16kb and your mmap_size is 1GB, only 16kb > is > > used. Unless you add data to the file, then the mmap area grows, > > obviously. > > > > -scott > > > > > > On Mon, May 2, 2016 at 2:01 AM, Mikael <mikael.trash at gmail.com > > <javascript:;>> wrote: > > > > > Dear Dr. Hipp & list, > > > > > > What about making Sqlite's memory mapping adapt to the current database > > > size, in increments of say 100MB? > > > > > > The at least 48 bits (256TB) of addressing space that modern 64bit > > > architectures give per process is not suffering any risk of depletion, > as > > > long as the space not is used wastefully, which would be the case now > as > > > today in the absence of an incremental setting, to guarantee that a > > > database never will grow outside of the mmap size, a developer is > tempted > > > to set mmap_size to a value so high that it guaranteedly never will be > > > reached e.g. 1TB, and that way an application could be almost 100% > > wasteful > > > with address space, and that way a process would get a constraint of > max > > > 200 or so databases. > > > > > > Can Sqlite user code implement this by itself already somehow? > > > > > > This would also be useful to do memorymapped-only IO on an OS that not > > has > > > a unified buffer cache, such as OpenBSD, where memory mapping is > disabled > > > altogether for this reason currently. > > > > > > Looking forward to your response, > > > > > > Thanks, > > > Mikael > > > _______________________________________________ > > > sqlite-users mailing list > > > sqlite-users at mailinglists.sqlite.org <javascript:;> > > > http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users > > > > > _______________________________________________ > > sqlite-users mailing list > > sqlite-users at mailinglists.sqlite.org <javascript:;> > > http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users > > > _______________________________________________ > sqlite-users mailing list > sqlite-users at mailinglists.sqlite.org > http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users >
