[Tfug] disk usage mystery..

[JD Rogers]

On Tue, Apr 3, 2012 at 3:55 PM, JD Rogers <rogersjd at gmail.com> wrote:
> John, nice find. Thanks for passing it along. I finally read through
> those bug reports.
>
> I'm still not sure if my issue is related because I don't have
> anything of size in /tmp and /tmp is a tmpfs mount in ram anyway. I

Turns out I'm a liar. I was thinking of another laptop with SSD that I
set up using tmpfs for /tmp, but in any case, I just tried using tmpfs
and I I still run out of space after a day, so it clearly isn't
anything in /tmp.

It gives me an idea though.. I may try mounting various directories
like /home or /var onto another filessystem and test which one runs
down, which might help narrow things down.

> even tried rebooting and "never opening a gnome terminal" until the
> disk filled, so it doesn't seem to matter. Even weirder, the disk was
> full a couple times in the past week and then suddenly my space
> returned without restarting x and killing pulseaudio, so clearly
> something can occasionally release the used space, but I haven't made
> any progress in narrowing down what. In the last week, the disk usage
> even hovers at 500MG for a day or so before suddenly spiking and
> filling the remaining GB. Its almost like whatever is broken is slowly
> healing. :-)
>
> I'm tempted to give up, move on, and just wait another few months
> until I buy a new laptop and then do a fresh install. But I'd love to
> at least be able to track down the culprit and file a bug report.
> Frustrating.
>
> Hope you have better luck with CTRL-End.
>
> On Sun, Mar 25, 2012 at 9:23 AM, John Gruenenfelder
> <jetpackjohn at gmail.com> wrote:
>> I've got an update for you that may help in tracking down the ultimate
>> culprit of your file system filling bug...
>>
>> While scouring the Internet with Google to find a solution to the
>> problem that libvte derived terminals (roxterm, gnome-terminal, etc.)
>> are treating End and CTRL-End as the same key (End), I found a bug
>> report stating that gnome-terminal (again, via libvte) was constantly
>> writing to /tmp.  Remembering your plight, I took a look.
>>
>> Some of the suggestions in the following bug report threads may help:
>>
>> https://bugs.launchpad.net/ubuntu/+source/vte/+bug/865082
>> https://bugzilla.gnome.org/show_bug.cgi?id=631685
>>
>> In short, one of the temp files libvte stores in /tmp contains the
>> entire scroll-back buffer and it is writing to it constantly.  Like
>> most transient temp files, it is opened for read/write access and then
>> deleted from the file system (but remains open as long as the creating
>> process is alive).  Given how much sensitive information can
>> potentially be displayed in a terminal, I certainly hope they've got
>> the security concerns nailed down.  As somebody who does not use a
>> stupendously large buffer size, I naturally assumed that it was stored
>> in RAM.  For that matter, even for a giant buffer I still would have
>> assumed that it was kept in memory.
>>
>> Anyway, the second link above suggests a very useful way for
>> monitoring file system access.  You can use the kernel's inotify
>> system and a simple Python command to have all of this information
>> streamed to a terminal or file:
>>
>>  python -mpyinotify /tmp
>>
>> You will need Python's inotify bindings (in Debian, this is the
>> python-pyinotify package).  It's fairly easy to read and even
>> colorized.  Obviously, don't run this in a terminal that will be
>> hammering /tmp.  Regular xterm will work fine.  Similarly, if you dump
>> the output to a file, make sure you are not storing that file in the
>> same place you are monitoring.  By default, this seems to recursively
>> monitor the path you specify.  I created some directories a couple of
>> levels deep and touched a file inside and Python was still notifying
>> me of the access.
>>
>> I hope this makes solving your problem a little easier.
>>
>>
>> P.S. I still haven't found a solution to my End=CTRL-End problem with
>> libvte.  It works fine with xterm, though.  I map CTRL-Home/End to
>> jump to the beginning/end of a buffer in EMACS so it's not a show
>> stopper, just annoying.  The devs are at least aware of it, via an
>> Ubuntu bug report, but the last activity was a year and a half ago...
>> :(
>>
>>
>> --
>> --John Gruenenfelder    Systems Manager, MKS Imaging Technology, LLC.
>> Try Weasel Reader for Palm OS  --  http://weaselreader.org
>> "This is the most fun I've had without being drenched in the blood
>> of my enemies!"
>>         --Sam of Sam & Max
>>
>> _______________________________________________
>> Tucson Free Unix Group - tfug at tfug.org
>> Subscription Options:
>> http://www.tfug.org/mailman/listinfo/tfug_tfug.org