From: Brian M. <bri...@gm...> - 2010-07-13 21:31:11
|
On Tue, Jul 13, 2010 at 1:12 PM, Les Mikesell <les...@gm...> wrote: > On 7/13/2010 11:53 AM, Brian Mathis wrote: >> I am trying to use BPC over a WAN, and the connection I'm using seems >> to be unstable. The backup is dying every few hours and giving a >> "Child exited prematurely" message. I can see a definite pattern in >> my bandwidth usage going up and down. >> >> In a situation where the connection drops like this, does BPC/rsync >> pickup where it left off, or does it try to start the whole thing over >> again? If I point the backup at an LVM snapshot instead of the real >> files, will that allow it to pickup where it left off more easily? > > I think incrementals are discarded on failures but the files copied in a > partial full are saved. However saving them may not save that much time > since the full runs will verify the contents against the source anyway. > If you are going through NAT or stateful firewalls some intermediate > network device might be timing out and dropping the connection in long > idle times. Assuming you are using rsync over ssh, you might try > enabling ssh keepalives by setting ServerAliveInterval in > /etc/ssh/ssh_config. > > -- > Les Mikesell > les...@gm... I've looked at my sar charts (ksar) and there are no obvious out of memory conditions, and nothing in the log files. I can also clearly see the network usage dropping when it crashes (also shown on sar). I've updated the SSH config with the ServerAliveInterval directive, so I'll have to see if that helps. I originally thought it was an unstable internet connection, but I actually have 2 backups running at the same time, 1 incoming and 1 outgoing, and only one of them drops at a time, so it's not the whole connection. One of the backups seems to crash regularly at about the 6 hour mark, and the other at the 2 1/2 hour mark. The only correlation is that the backup that lasts longer is on a server with more RAM, but again there's no indication of out of memory. |