We have never had any version of Bacula in which Differential
backups were incomplete.
That's really good to know, thanks for the confirmation.
Running with a FD that is more recent
that the Dir and the SD or a Dir and SD that are not identical is
a big problem. Until that is fixed it doesn't make much sense to
speculate about any odd behavior.
I have addressed that yesterday by rolling out the rawhide backports Alex Domoradov linked to (thanks, Alex!). So everything is now 5.2.13 except for some Debian-derived systems whose FDs are 5.2.6, which sounds like it wouldn't be a problem.
I'm going to force a full dump on all clients over the next day or so, and then monitor from there. It will be a couple of weeks with the current configuration before I will be able to conclude that things are OK.
Also, if you are doing
something very unusual in your FileSet or using multiple FileSets
in your backup, there could be a configuration problem or perhaps
even an undiscovered bug,
I have multiple FileSets (one per OS variant, essentially), but any given client only ever uses one. I'm hoping that's not "very unusual".
but the first thing to do is correct any
possible version problems you have.
I'm curious; does the Dir have programmatic access to the FD's version number (I know it's in "status client" but that might just be text)? Seems like if this is as vital as you describe, it would be better to error out if it was newer.
Cheers,
