rpm 5.x in Th

[Tomasz Pala]

On Sun, Sep 23, 2012 at 18:17:38 -0400, Jeffrey Johnson wrote:

>> Try running replicated postgresql master node on failing drive and share
>> your results!
> 
> Why?

Just try it.

> The issue(s) involved with maintaining databases
> consistently with package manager upgrades are non-trivial
> to solve with a perl script like yours.

My script does (well, should - it was 1-minute typing and might be
error-prone) exactly what shall be done. It doesn't alter any
database more than every other legitimate invocation, as it doesn't use
any hacks except reversing order of transactions performed.
In term of databases it's not rollback - but contrary to RDBMSes here this
doesn't matter; primary difference is that standalone database has
strictly specified and controlled I/O vectors, while rpm in properly
used system cannot track data, but metadata only; in *sql it's not normal
usage-scenario when someone replaces blob using filesystem tools, so the
assumption of data consistency is solid.

> So write a perl sc riot that permits upgrading postgresql
> masters now that you have solved --rollback with one line of perl.

I ain't solved --rollback, just replaced it with something usable.

-- 
Tomasz Pala <gotar at pld-linux.org>