From: Stelian P. <st...@po...> - 2002-08-18 13:49:19
|
On Thu, Aug 15, 2002 at 01:31:07AM -0400, Mark W. Krentel wrote: > > But its great to see the other people on this list > > answer the Linus FUD at my place. Thanks guys! > > I'm not sure if I'm dispelling the FUD or adding to it. It seems that > the more tests I run, the more confused I get. :-( > > The original problem I reported in ext2 with files written without > O_TRUNC doesn't happen in later 2.4.x kernels. It's very repeatable > in RH 7.1 (2.4.2 kernel), but I can't reproduce it in RH 7.3 (2.4.18). > So that must have been a bug that has been fixed. Bonus points to > someone who can point to something in the kernel changelog showing > what was wrong and when it was fixed. There were so many things fixed (not only but especially in the VM area) between 2.4.2 and 2.4.18 kernels that I don't think there is a single fix... > So, I'm still not clear on a few fundamental questions. I'll try to answer them then. > (1) Is ext2/ext3 dump on a Linux 2.4.x kernel viable for an idle but > mounted file system? That is, finish some writes, close all file > descriptors, sync, wait 2-3 minutes (or 2-3 hours), would you expect > dump to see the correct version of all files? After issuing a flush command to the block device, and there are no more writes to the disk, then yes, I'd expect dump fiability to be 100%. If it doesn't, then it is a kernel bug or limitation. And you should report this to the kernel developers. > (2) Granted that there are normal risks with dump on a live file > system, do you interpret Linus's message as implying that these > risks are higher in Linux than in other Unixes? No. All Linus says is that dumping an active filesystem is not guaranteed to work. And that recent changes in the cache area can make dump work even less on active filesystems... > I've previously claimed that the 2.4 kernels make it impossible for > dump to see the correct version of some files. Theoretically, that > would make dump pretty much deprecated for a mounted file system. > Do you agree, or am I spreading FUD? You are somewhat spreading FUD because dump was never meant to be used on mounted filesystem, on any kernel. > (3) Is ext3 any more vulnerable to these cache issues than ext2? I have no idea on this matter. > I realize that these issues are more about the kernel than about dump, > and whatever problems remain can't be fixed in the dump code. But they > do affect dump quite heavily. I agree. However, there are mechanisms (like the snapshots etc) which, once they will be ready, will guarantee a 100% reliability even on mounted filesystems. > Anyway, from a practical view, the only file that restore -C has > reported as different has been /etc/dumpdates (of course). So, I've > decided to stay with dump until it proves to me that it no longer > works. I like dump and I hope that day never comes. This is exactly what I said, you are in the 99% fiability area. Stelian. -- Stelian Pop <st...@po...> |