Backup Central

Dave Kempe <dave < at > solutionsfirst.com.au>
wrote the following on Thu, 29 Sep 2005 09:02:49 +1000
Hi,
I am trying to get rdiff-backup 1.0.1 going on cygwin. I will have a
windows package built shortly, but I would love it if windows-> windows
backups worked.
the initial backup works ok, but after the session stats are printed at
the end I get an error similar to the one below. Thats fine, but
subsequent backups crash with the error below before copying any files.
I think os.fsync is not available on cygwin python2.4 and is erroneously
turned on...

Let me know if there any other information you need to help me out - is
there some way I can turn off fsyncing on cygwin?

rdiff-backup needs some way to order writes. For instance, when
updating it sometimes needs to make sure a file is fully written
before it is moved. Many optimized file systems will mess with the
order of these operations for better performance (at least this is
what the ext2/3 people told me).

So without fsync, how can rdiff-backup be guaranteed everything is
written in the right order?

It would be easy to add a check and then just disable fsyncing if it's
not available, but I don't think this would be entirely safe.


--
Ben Escoto

Ben Escoto wrote:
It would be easy to add a check and then just disable fsyncing if it's
not available, but I don't think this would be entirely safe.



since we are using windows, we have come to accept we don't live in an
ideal world I think we are prepared to accept not entirely safe.
Perhaps a warning at the start of the backup might be sufficient :)

dave

David Kempe <dave < at > solutionsfirst.com.au>
wrote the following on Sat, 15 Oct 2005 12:37:57 +1000

since we are using windows, we have come to accept we don't live in
an ideal world I think we are prepared to accept not entirely
safe. Perhaps a warning at the start of the backup might be
sufficient :)

That shouldn't be too hard, but is there really no better solution?
Why doesn't windows have fsync? I mean, it must have some system call
that does exactly the same thing, but cygwin just doesn't map it?

If there's no way to do it then I'll just add a test to fs_abilities,
and display that warning you mentioned if the filesystem doesn't
support fsync.


--
Ben Escoto

Ben Escoto wrote:
That shouldn't be too hard, but is there really no better solution?
Why doesn't windows have fsync? I mean, it must have some system call
that does exactly the same thing, but cygwin just doesn't map it?

From my reading it seems that the python2.4/cygwin version I have got
should support it - there was fsync bugs in older versions.
How do i do a test to see if it works - do you have a simple test script
I can run? Disabling it is not optimal i agree, but a warning would be
sufficient I think.

thanks

dave

David Kempe <dave < at > solutionsfirst.com.au>
wrote the following on Sun, 16 Oct 2005 13:50:07 +1000

From my reading it seems that the python2.4/cygwin version I have
got should support it - there was fsync bugs in older versions. How
do i do a test to see if it works - do you have a simple test script
I can run? Disabling it is not optimal i agree, but a warning would
be sufficient I think.

I remember an fsync bug in python that caused some problems for me,
but I thought the issue was that it was just missing from the os
module. Maybe there is a different bug in cygwin's fsync.

Anyway, for a simple script, try this:

echo testing > fsync-testfile
python -c 'import os; os.fsync(os.open("fsync-testfile", os.O_RDONLY))'
rm fsync-testfile

if it works, python won't say anything.


--
Ben Escoto

Ben Escoto wrote:
if it works, python won't say anything.



That didn't work for some reason.

I think this is the same error rdiff-backup gives:

Traceback (most recent call last):
File "<string>", line 1, in ?
OSError: [Errno 13] Permission denied

google doesn't show anything obvious as to why this is happening. For
now, disabling it with a test might be the way to go, with a warning...

dave