[BBLISA] dump or Legato Networker?

[Daniel Hagerty]

Dean Anderson <dean at av8.com> writes:

> On Sun, 6 Jan 2008, Edward Ned Harvey wrote:
> > 
> > Is there any way to signal the Linux kernel to flush its filesystem write
> > cache, directly before performing the backup?
> 
> sync(8) should do this.  But maybe not on linux... sigh.

    Sync will not help you with backups of anything more than
negligible duration, and probably not even then.


    Some flavor of linux filesystem apparently has snapshot support
these days, enough that at least Ted Tso (one of the original ext2
people) was recently recommending it on LKML for a data safety problem
that a user was having.  The users problem was different than backups
per se, but the solutions offered ought to apply.


    On the subject of racing in dump vs that of userland backup
programs, note that the class of race conditions that dump is exposed
to is much broader than userland programs.  Userland programs race as
they enumerate the files and directories of the filesystem, where they
can miss concurrent adds, deletes, and renames.  They race on the
contents of files that are being actively written, and that's about
it.

    Dump is racing against the entire filesystem structure.  It can
generally work, but I've also seen specacular failures that are
impossible via the more straightforward userland route.  More than
once I've received a dump output log that was fifty megabytes of error
messages as dump tries to seek past the end of the disk.