You misunderstand. I'm not saying that amandad should answer multiple requests, I'm saying that the tape server part (planner etc.) should understand that e.g. fileserv and amanda-server in my example is actually the same host, and make only one request at a time. The information needed to do this is easy to find - in fact, planner has probably looked it up already; it's bound to do gethostbyname() at some stage to be able to contact the client at all.No way would I call this a bug. Because of the business AMANDA is in, answering multiple concurrent requests could be dangerous.
There's no reason that your backup software shouldn't know every machine by one, and only one name, no matter how many names your users know it by.
I disagree completely.
Well, according to that logic, Amanda should use IP addresses only, not host names.There are places in IT to obscure what's going on, this ain't one of them.
Anyhow, I think this is a bug no matter what your view on using diffeent aliases for the same host is, simply because Amanda allows it.
Possibly.If you're worried about your indexes being broken when you move a disk, how often do you really move a disk? Compare that to how often you restore, or better, how often you test your restores. You might be chasing a ghost, Scooby.
The point is not that I move disks a lot or that I am worried about the indexes as such, though. It's more of a general policy we have: Never refer to hosts using their "real" names, especially in configuraton files - always use "function" or "service" based aliases. This saves a lot of work in the long run, and also makes disaster recovery a lot easier and faster. What if my server fails in such a way that I have to move its shared disks to a differerent (backup) host? I'd rather not have to trace all config files that might refer to the volumes via the server name in order to get the net back into a functional state then. (Not that running backup is a top priority under such circumstances, mind you.)
The weird bit is that planner does display a "host busy" message under other circumstances. I sometimes run it by hand in order to (partly) test the effect of config changes, and on a couple of events I've managed to get it to exit in an unclean manner so that amandad is left running on the client. If I restart planner after that, I get a message saying that the host is busy, not that connection timed out.It might make sense to also be prepared to do non-index recoveries. (Be prepared, test both monthly!) If you really want to do this, run multiple amdumps per day, and alter the disklist in between. Make sure one is done before the next starts. IIRC, The timeout is because inetd doesn't start a second amandad for the second connection request. So it simply goes unanswered.
This might be a good thing to put some code into: On amandad startup, if it detects another amandad running, it returns some sort of "HOST BUSY" message. Probably easier to do in 2.5 with the NOP feature. > -----Original Message----- > From: Toralf [mailto:toralf@;kscanners.com] > Sent: Thursday, October 17, 2002 1:40 PM > To: Anthony Valentine > Cc: Amanda Mailing List > Subject: Re: Host name aliases in disklist > > > > On Thu, 2002-10-17 at 03:03, Toralf Lund wrote: > > > > Hi, > > > > your assumption about the hostnames is correct. > > > > > > > > What you are doing is one of the big NO!NO!'s of amanda. > > > Just like I suspected ;-/ > > > > > > > > > > Never list a single host in multiple concurent amanda runs, or, > > > > in your case, with different names in a single amanda > > configuration. > > > > amandad can only handle one request at the same time. > > > Yes, I knew that. Which is why amdump/planner should do a host > > lookup and > > > base it's decision on IP addresses. > > > > > > > > Only solution is to use only one hostname per physical host in > > disklist. > > > I'd say the failure to handle aliases is a *BUG*! > > > > I don't think he is saying that it's the aliases that are a problem, > > but > > rather having the same host in the disklist twice as two separate > > aliases. > The problem seems to be that Amanda doesn't understand the concept of > aliases, i.e. that different hostnames may refer to the same physicl > host. > > > > > So, if that is true, you could have an entry for either > amanda-server > > or > > fileserv, but not both only because they are the SAME host. > > > > Amanda can only handle one connection at a time, which would explain > > why > > amanda-server will time-out while fileserv is backed up > fine, and then > > vise versa. > It's still a bit surprising to me that I get a timeout.�I would > normally expect a "host busy" message. > > > > > > > > > > No other way out there. > > > Again, this is really a shame. It causes extra work & a > whole lot of > > mess > > > in the indexes etc. if you move disks around a bit. > > > > > > - Toralf > > -- > > UNIX was not designed to stop you from doing stupid things, > > because that would also stop you from doing clever things. > > >
