Re: Replace /lib and /usr/lib on a server

[Alvin Oga <aoga@mail.Linux-Consulting.com>]

On 15 Jul 2005, Felix Natter wrote:

> it is my task to create a complete backup+restore method of our woody-based
> web- and mailserver. The problem is: how can I restore /lib and
> /usr/lib/*.so* in a running system? Whenever I tried it (tar --overwrite)
> the system halted (IIRC a statically compiled tar didn't help).

you can NEVER overwrite a file in a running system, unless it is an
identical file
	- you can replace a running system's kernel but that too
	is tricky and fun biz

the problem also apllies to all binaries in /bin /sbin /usr/sbin /usr/bin 
and eanything else in the sytem

if you want to restore into a running system ... and you can do it
properly, you'd be an instant millionaire ??

----

best way to do a "complete backup and restore"
	- rsync the master system into the slave box except for ip# and
	hostname ( the salve system is alway ready to take over since
	the last backup )
		- backup too often and the original problem that
		cause the master to die will propagate and kill
		the slave too ( ie.. you'd need to know why your master
		died in the first palce )

		- rsync is dumb ... it will propagate failures

	- do overlapping full and overlapping incremental backups
	just in case yesterdays full backup failed and you dont notice
	it till 2 weeks or months later when the system dies

	- gazillion "do it this way or that way" suggestions
	depending on who you ask


- big questions ...
	- how time is lost if the "backup and/or restore" didnt work
		- backups should be 100% automated and self correcting
		and self checking  ( no time to run .. just a few hours
		to setup properly )

		- soemthing else is seriously wrong if you have to
		"restore"

	- how much $$$ is lost
		$$$ lost due to crashes should be more than say 3x-5x 
		of the costs of all the preventative backup and restore
		methodology

	- how long to get back to "normal operations"
		( should be under 5 min )

c ya
alvin