Re: [Savannah-hackers-public] disk-usage monitoring on vcs.sv.gnu.org

[Assaf Gordon]

Hello Karl,

On Jan 25, 2015, at 17:00, Karl Berry <address@hidden> wrote:

> As long as you're doing a du of everything -- when looking for disk hogs
> (wasn't that what we were talking about initially?),

Yes, that's the goal.

> you need to know
> the exact directory.  I find GNU du's --separate-dirs option the most
> useful approach.

Do you mean each repository directory and other 'important' directories on the 
disk, or literally every directory on the disk separately (as --separate-dirs 
does)?

For example, each git repository directory contains several sub-directories 
(e.g. "hooks", "objects", "refs", etc.) - wouldn't it be more helpful to treat 
it as one item (the repository directory + its sub-directories) ?

If we want to track individual repositories, how about something like this (not 
tested):

    for SRC in /srv/git /srv/hg /srv/bzr /source /web ; do
        find "$SRC" -maxdepth 1 -type d -print0 \
            | LC_ALL=C sort -z \
            | sed -z 1d \
            | xargs -0 du --bytes --human --summarize
    done

> On another diskfull front: frontend's disk has been 100% full for a
> while, for the same reason as before.  I'm wasting my time dealing with
> it again.  Can nagios be configured to send a report, say, hourly when
> any partition is, say, >98% (= requires action)?  If not, a trivial
> shell script that just runs df would do the job.

I'm not well familiar with Nagios, but I'll give it a try.

- Assaf