dump-users

[Dump-users] restore: Incremental tape too high

[Tom Y. <mad...@te...>]

i'm getting this when i try to restore a level 1 dump on top of a level 0.

i have a server (CentOS linux) with software RAID, which i want to replace 
with a newer server.  to get a clean dump, i sync'ed the discs then broke 
the mirrors (mdadm -f, mdadm -r), e2fsck'ed the detached mirror, and did a 
level 0 dump of it with the "-u" flag.

once the dump had finished, i edited /etc/dumpdates on the live mirror to 
reflect the time at which the mirrors were split (09:26:53), rather than 
the slightly later time the dump started (09:34:43), as i needed a later 
incremental to reflect all changes since the mirrors were broken, rather 
than since the start of the dump.

once the mirrors had resynced, i did a level 1 incremental.

"restore tvf" confirms that the level 0 dates from 09:34:43, while the 
level 1 dates from 09:26:53.

when i restore the level 0 dump to the new box, which runs OK, then try to 
restore the level 1 on top, i get the error "restore: Incremental tape too 
high".

am i right in thinking that the error is trying to tell me that the 
incremental "pre-dates" the full backup?  if so, is there any way to tell 
restore to ignore that and just do the restore?

thanks for any light anyone can shed.


-- 

       Tom Yates  -  http://www.teaparty.net

Re: [Dump-users] restore: Incremental tape too high

[Stelian P. <st...@po...>]

Hi,

On Tue, Jun 08, 2010 at 07:45:41PM +0100, Tom Yates wrote:

> i'm getting this when i try to restore a level 1 dump on top of a level 0.
[...]
> am i right in thinking that the error is trying to tell me that the 
> incremental "pre-dates" the full backup?

Yes: (in restore.c:
	if (hdr.dumpdate != dumptime)
		errx(1, "Incremental tape too %s",
			(hdr.dumpdate < dumptime) ?  "low" : "high");

> if so, is there any way to tell 
> restore to ignore that and just do the restore?

I'm not sure you can do this without modifying the source code
(commenting out the above test).

The next time I suggest you use the -T option of dump to set the
dump time to the date when the mirrors were broken.

Stelian.
-- 
Stelian Pop <st...@po...>

Re: [Dump-users] restore: Incremental tape too high

[Tom Y. <mad...@te...>]

On Thu, 10 Jun 2010, Stelian Pop wrote:

> Hi,
>
> On Tue, Jun 08, 2010 at 07:45:41PM +0100, Tom Yates wrote:
>
>> i'm getting this when i try to restore a level 1 dump on top of a level 0.
> [...]
>> am i right in thinking that the error is trying to tell me that the
>> incremental "pre-dates" the full backup?
>
> Yes

it's extremely useful just to know i'm thinking about the right error - 
thank you for that.

> : (in restore.c:
> 	if (hdr.dumpdate != dumptime)
> 		errx(1, "Incremental tape too %s",
> 			(hdr.dumpdate < dumptime) ?  "low" : "high");
[...]
> I'm not sure you can do this without modifying the source code
> (commenting out the above test).

i would be very happy to do that, and if this crops up again i may well do 
that.  i thought about it myself, but couldn't be sure i wasn't breaking 
some other important test at the same time, which uncertainty you have 
removed for me.

> The next time I suggest you use the -T option of dump to set the
> dump time to the date when the mirrors were broken.

the man page is seriously unclear that -T can be used in that way, but now 
i know that it can, i'm completely sorted for next time.

in the event, i did some find-based trickery to confirm that there existed 
no file which (a) had been modified in the eight-minute window between 
breaking and dumping, and (b) had not since been re-modified, so i decided 
it was safe for me to revert /etc/dumpdates and do a new incremental.

but if my window had been longer or i hadn't thought to do that test, your 
input would have completely saved me.  thank you!


-- 

       Tom Yates  -  http://www.teaparty.net