>>>>> On Tue, 15 Dec 2020 11:31:46 +0100, Andrea Venturoli said:
> 
> Hello.
> 
> I'm backing up a FreeBSD server where Cyrus IMAP runs and I'm using 
> Full/Differential/Incremental scheme.
> 
> As you probably know, Cyrus IMAP uses hardlinks to avoid saving 
> duplicate messages (and probably for other things).
> 
> 
> 
> I tried a test recovery yesterday and I was flooded with messages like 
> the following:
> > Error: create_file.c:351 Could not hard link 
> > /zroot/brestore/temp/var/spool/imap/.zfs/snapshot/bacula/user/xxxx/SomeFolder/12610.
> >  -> 
> > /zroot/brestore/temp/var/spool/imap/.zfs/snapshot/bacula/user/xxxx/34966.: 
> > ERR=No such file or directory
> 
> What I *think* happened here is that the user moved the message to 
> SomeFolder and Cyrus implemented this by creating a new hard link, later 
> removing the original message file from Inbox.
> All the errors above came up reading from the Full volume: no errors 
> were present when reading the Differential or Incremental ones, but I'm 
> not sure this is a rule.
> I verified some of these files are currently present in the original 
> data, but not on the restored data!!!
> 
> How does Bacula support hard-links? Is this expected to work? Are there 
> any options I should set?
> 
> 
> 
> Could it be that Bacula:
> _ does not restore the "original" file, since it was present at the 
> moment of the Full backup, but gone later (when the last incremental was 
> taken);
> _ doing so cannot create the linked file?

Possibly, but creating or removing a hard link should change the ctime on the
inode and cause the file and its links to be backed up again in the next
Differential/Incremental.

What is the fileset definition?

Perhaps you can track down a minimal set of jobs that trigger the problem by
using the "Select backup for a client before a specified time" option.  Start
with a time just after the end of the Full backup.  Does that restore without
errors?  If so, then increase the time to include more and more
Differential/Incremental backups until the problem occurs.

__Martin


_______________________________________________
Bacula-users mailing list
Bacula-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bacula-users

Reply via email to