On Dec 1, 2009, at 5:07 AM, Bernhard Rosenkraenzer wrote:

> Hi,
> we've been using rpm5 head on our build machines for about 2 weeks now -
> after increasing the number of locks in DB_CONFIG, it works very reliably,
> but now the build machines are running out of diskspace because the rpm
> database takes up several (>20) gigabytes of space.
> 
> The files that have always been there are about the size they used to be,
> but every rpm invocation creates a new log.000* files, each 10 MB in size.
> 
> I can see why the rollback features they bring along can be important to
> some people, but for desktop users who install lots of updates on
> relatively small disks (and for developers who build lots of stuff) this
> can be a real problem -- is there a way to limit the number/size of log.*
> to keep it within reasonable boundaries, short of running
> 
> cd /var/lib/rpm
> rm -f `ls |grep -vE (Packages|DB_CONFIG)`
> rpm --rebuilddb
> 
> whenever it gets too large?
> 

Here's the CLI commands needed to remove log files:

	cd /var/lib/rpm
	/usr/lib/rpm/bin/db_checkpoint -1
	/usr/lib/rpm/bin/db_archive -d

There's also db_recover, which will abort uncommitted database transactions.

(aside)
The db_recover utility likely "works" now that I've made the logging abstractions
necessary for syscall/scriptlet ACID "opt-in". untested atm.

All of the db_checkpoint/db_archive/db_revover mechanism is easily
automated directly in -lrpmdb.

What is harder is devising "policy" to attempt the RPM implementation that is
consistent with goals other than
	What is this "stuff"?!? Nuke it!

hth

73 de Jeff