Hello all, we are using rsync to produce a replica of our filesystem. it's very important for us to be sure that the files replicated on the remote backup host are left in a consistent state as we may need to boot services from those files (configs, logs, etc)! right now we use a 'greedy approach'... we stop all important services (dhcpd, squid, sendmail, etc) just to minimize the possibility of copying to the backup a file that was being written. Could we be over doing it ? Are we exagerating ? An example: if we do not stop sendmail were rsync'ing can we be sure that the queues on the replica machine are in a consisten state ? same for dhcpd leases or squid's cache ? We would really like to leave all those important services up but we are really affraid of producing a corrupted replica... can anyone comment on this ? is rsync imune to this kind of problem ? thank you very much. Joao -------------- next part -------------- HTML attachment scrubbed and removed
* Message by -Joao Miguel Ferreira- from Sun 2008-03-30:> we are using rsync to produce a replica of our filesystem. > > it's very important for us to be sure that the files replicated on the > remote backup host are left in a consistent state as we may need to boot > services from those files (configs, logs, etc)! > > right now we use a 'greedy approach'... we stop all important services > (dhcpd, squid, sendmail, etc) just to minimize the possibility of copying to > the backup a file that was being written.[...]> We would really like to leave all those important services up but we are > really affraid of producing a corrupted replica...Use filesystem snapshots. Linux supports them via LVM. Solaris and FreeBSD offer them too. -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 835 bytes Desc: not available Url : http://lists.samba.org/archive/rsync/attachments/20080330/0cee6a9a/attachment.bin
On Sun, 2008-03-30 19:39:12 +0100, Joao Miguel Ferreira <joao.miguel.c.ferreira@gmail.com> wrote:> right now we use a 'greedy approach'... we stop all important services > (dhcpd, squid, sendmail, etc) just to minimize the possibility of copying to > the backup a file that was being written. > > Could we be over doing it ? Are we exagerating ?I don't think that rsync could do that, but there are chances... Some years ago, I hacked a small preload library to wrap all the usual file-changing syscalls and replicating them over the net to a different box. That box doesn't have anything running except the data-accepting daemon matching that preload lib. Unfortunately, that lib got lost (commercially written) and wasn't ever ment to be open-sourced. However, the idea is there and it wasn't all that hard to implement it. Maybe a better aproach would be to just omit the receiving daemon and use the NFS protocol instead. Maybe using FUSE these days instead of a preload lib would also be a nice imprevement... MfG, JBG -- Jan-Benedict Glaw jbglaw@lug-owl.de +49-172-7608481 Signature of: What we do for ourselves dies with us. What we do for the second : others and the world remains and is immortal. (Albert Pine) -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 189 bytes Desc: Digital signature Url : http://lists.samba.org/archive/rsync/attachments/20080330/e805355a/attachment.bin
On Sun 30 Mar 2008, Joao Miguel Ferreira wrote:> > An example: if we do not stop sendmail were rsync'ing can we be sure that > the queues on the replica machine are in a consisten state ?I don't know how sendmail organizes things, but at least with exim there aren't any extra indexes or such that need to be consistent with the spool.> same for dhcpd leases or squid's cache ?dhcpd leases should not be a problem. I'd not even consider syncing a squid cache. After all, it's a cache, not important data... Paul Slootman