To those moving to Jaunty: Do not use the new ext4
dotancohen at gmail.com
Thu Mar 12 09:57:09 GMT 2009
> So we would see the same behavior with ext3 if the time between syncs
> was the same?
Yes, in fact, there are other filesystems in which this is the case.
> I later read Theodore Ts'o (one of the developer) comment 45 an
> explanation why and how:
> But I still cant understand, no matter if you depending on the cache
> or the libraries you using don't use flush() the right way isn't a
> journaling file system supposed to only register complete transactions
> and by that guarantee (of course you cant guarantee that, but it
> should be minimal) that the files would not be destroyed/truncated?
I am do not really understand all the gotchas of filesystem-level
access, but it seems to me that you would see quite a lot of locked up
processes if a "complete transaction" entailed waiting until the
actual write. The process just needs to know that the filesystem got
the data and will handle it from there. Assuming that the filesystem
will be responsible with that data seems reasonable for most cases.
For critical cases, the process need not assume and can force a write
to disk. Which KDE does not.
More information about the kubuntu-users