On Thu, 2004-04-15 at 00:23, Michael A. Peters wrote: > On Wed, 2004-04-14 at 16:13, Michael A. Peters wrote: > > > > > > 2. The O_DIRECT patch/hack that allows rpm (actually, the underlying > > > Berkeley DB) to work correctly. > > > > That's scary - a kernel patch needed for a user space library to work > > properly? I'll have to look that one up and see what that's about - but > > I initially don't like the sound of that. > > OK - O_DIRECT looks like a performance tune more than anything, and > isn't _necessary_ so to speak. > > I was running vanilla rpm 4.3 on an LFS system with BerkeleyDB 3.x > (whatever the stable 3 was) and has no issues - so I suspect the > O_DIRECT patch wouldn't have been a necessity to have. But not having > run RH9/FC1 I can't say for sure. Just that other systems with rpm and > BDB 3 and 4 and rpm ran fine w/o the patch - so unless RH/Fedora messed > up BDB then that wouldn't have been an issue. The RHL9/FC1 kernels disabled O_DIRECT. If bdb tries to use it, bad things can happen as it doesn't take into consideration the alignment rules necessary when using O_DIRECT. I think this actually got fixed in later versions of bdb, but I don't know if those fixes made it into the RHL9/FC1 versions of bdb. Dave
