|
Re: [Networker] Questions on backup renamed directories attribute?
2012-08-10 17:51:21
On 2012-08-10 16:21, Dag Nygren wrote:
On Wednesday 08 August 2012 18:09:59 George Sinclair wrote:
Hi,
Three questions on the 'Backup renamed directories' attribute:
1. If you turn on the 'Backup renamed directories' attribute for a
client then under what circumstances *might* 'save -i' (backup command
attribute) not work as expected?
I wouldn't expect any other drawbacks/differences than slower backups because
of the
extra lookups. Have you seen differences?
Thanks for your response, Dag. I have 'Backup renamed directories'
enabled for a number of clients, but I've not actually tried
implementing 'save -i' at the same time. However, if you use NMC, then
as a test, first try turning the 'Backup renamed directories' option on
for a given NSR client resource. Next, enter 'save -i' in the 'Backup
command' field under Apps & Modules. Does NMC respond with this message ???:
Enabling Backup renamed directories may cause <save -i> specified in
'backup command' attribute not working properly, re-apply if this is
acceptable
If so then cancel, turn off the 'Backup renamed directories' option,
save and then come back in and now try setting the 'save i' value first
and then turn on the 'Backup renamed directories' option second, in that
order. Does it not complain in that case? Hmm ...
Could also simply use global directives with the save environment
directive: ignore
2. Why does enabling this attribute cause the save program to perform a
lookup in the CFI in order to determine if a directory have been renamed?
Normally, NW is looking at the ctimes of the files and comparing these
against the value stored in the media database for the affected save set
for the previous backup. I would think the CFI would only be read when
doing recovers, so if a directory is renamed, then its ctime will
change, and if NW knows that the ctime has changed then why not just
back up everything under there without having to check the client file
index?
This is because NTFS (or Microsoft) doesn't consider renaming a directory as
beeing a change in that directory (sic!) and doesn't modify the ctime of the
directory.
This is a grey area in POSIX and of course MS had to do it on a different way
than the rest of the world...
Well, I guess I was referring mostly to Unix in my question, but I
didn't specify OS, so I should have been more specific. You make an
excellent point, however, about the MS example. I hadn't thought about
that. But let's say were instead talking about maybe a Linux system,
wherein the OS *does* recognize ctime changes on directories. In that
case, would NW still read the CFI to determine if directories have been
renamed, or would it just look at the ctime of the directory, compare
that with the timestamp for the media database entry for the last backup
of the save set, and if it's later then just back up all the constituent
files carte blanche, thus saving it the trouble of even having to even
read the CFI?
The Help under NMC for 'Backup renamed directories' says: ... The save
program performs a lookup in the client file index to determine whether
a directory has been renamed. If a directory has been renamed, all of
the files and subdirectories under the directory will be backed up.
So I was thinking, therefore, that it didn't matter what OS was used on
the client, and NW would still read the CFI? The Admin guide also
mentions that enabling this feature will increase backup times but
doesn't specifically state anything about OS and/or the CFI needing to
be read, at least nowhere that I could find.
George
3. What happens if you have this attribute turned on, but you're writing
your data to a non-indexed pool?
Shouldn't make any difference.
Best
Dag
--
George Sinclair
Voice: (301) 713-3284 x210
- The preceding message is personal and does not reflect any official or
unofficial position of the United States Department of Commerce -
- Any opinions expressed in this message are NOT those of the US Govt. -
|
|
|
