[1772] in athena10
Re: tmp-cleaner for Linerva
daemon@ATHENA.MIT.EDU (Quentin Smith)
Wed Mar 18 21:32:40 2009
Date: Wed, 18 Mar 2009 21:31:41 -0400 (EDT)
From: Quentin Smith <quentin@MIT.EDU>
To: Evan Broder <broder@mit.edu>
cc: Nelson Elhage <nelhage@mit.edu>, Greg Price <price@mit.edu>,
linerva@mit.edu
In-Reply-To: <49C18AC0.3050309@mit.edu>
Message-ID: <alpine.DEB.1.10.0903182130440.1549@vinegar-pot.mit.edu>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII; format=flowed
That seems pretty poor - if I leave a screen session for 2 days, I expect
it to still be there when I get back, including temporary files. I'm not
sure what the best metric is - would "pgrep $user" be too lenient?
--Quentin
On Wed, 18 Mar 2009, Evan Broder wrote:
> Looks like they don't get special treatment, certainly - people who show
> up in `w -h` get a free pass.
>
> - Evan
>
> Quentin Smith wrote:
>> As written now, does it preserve the files of people who have screen
>> sessions running?
>>
>> --Quentin
>>
>> On Wed, 18 Mar 2009, Nelson Elhage wrote:
>>
>>> I guess we may want to change the parameters for the tmp cleaner. The
>>> default is
>>>
>>> # The values here are minimum numbers of days. Negative means mtime,
>>> # positive means atime.
>>> my %dirs = ("/tmp" => 2,
>>> "/var/tmp" => 3);
>>>
>>> My gut reaction is to want something slightly longer, although I'm not
>>> sure. Much longer might not with this f2swp business, for instance.
>>>
>>> (We may want to make that configurable in debathena-tmp-cleaner
>>> through a file in /etc, as well, at some point. Doing that as long as
>>> we're deploying that might not be a bad plan)
>>>
>>> - Nelson
>>>
>>> On Wed, Mar 18, 2009 at 05:54:33PM -0400, Greg Price wrote:
>>>> Sounds good to me. We ran all the way out of space just now, so it's
>>>> past time we dealt with this.
>>>>
>>>> Nearly all of the huge files I deleted in today and yesterday's
>>>> episodes belonged to users who were not logged in, so this cleaner may
>>>> well be sufficiently thorough, anyway.
>>>>
>>>> Greg
>>>>
>>>>
>>>> On Wed, Mar 18, 2009 at 04:52:43PM -0400, Nelson Elhage wrote:
>>>>> Lacking a definite better plan, I say we set up the (Deb)Athena tmp
>>>>> cleaner. If we need something more aggressive, we can move to
>>>>> tmpreaper or something that doesn't necessarily respect logged-in
>>>>> status. But the Athena cleaner is something people are used to and we
>>>>> know works reasonably well in comparable environments.
>>>>>
>>>>> I'll hold off on doing so until I get at least one ack from another
>>>>> maintainer.
>>>>>
>>>>> - Nelson
>>>>>
>>>>> On Tue, Mar 17, 2009 at 07:26:45PM -0400, Greg Price wrote:
>>>>>> We should install a tmp-cleaner on Linerva. Today we nearly hit 50G.
>>>>>>
>>>>>> linerva / tmp / andersk 17:41 (Anders Kaseorg)
>>>>>> Filesystem 1K-blocks Used Available Use%
>>>>>> Mounted on
>>>>>> /dev/mapper/linerva-tmp
>>>>>> 51606140 47324116 1660968 97% /tmp
>>>>>> linerva / tmp / price 18:01 (Gregory N. Price)
>>>>>> # find /tmp -maxdepth 1 -size +1G -atime 3
>>>>>>
>>>>>> finds seven files. None of the owners are logged in. I'm
>>>>>> going to
>>>>>> remove them with /bin/saferm (like debathena-tmp-cleaner
>>>>>> uses.)
>>>>>> Then we can deal non-urgently.
>>>>>> linerva / tmp / price 18:01 (Gregory N. Price)
>>>>>> (pausing for objections...)
>>>>>> linerva / tmp / price 18:05 (Gregory N. Price)
>>>>>> Done.
>>>>>>
>>>>>> vinegar-pot:~# df /tmp/
>>>>>> Filesystem 1K-blocks Used Available Use%
>>>>>> Mounted on
>>>>>> /dev/mapper/linerva-tmp
>>>>>> 51606140 40478324 8506760 83% /tmp
>>>>>>
>>>>>> We should still set up a real tmp-cleaner.
>>>>>>
>>>>>> One obvious choice is debathena-tmp-cleaner, which as written removes
>>>>>> files in /tmp older than 2 days and in /var/tmp older than 3.
>>>>>>
>>>>>> But it refuses to remove files owned by users who are logged in, no
>>>>>> matter how long ago the file was last touched. Since the history of
>>>>>> -tmp-cleaner is that ghudson wrote it in July, I suspect that this
>>>>>> was
>>>>>> motivated by the cluster situation where a logged-in user "owns" the
>>>>>> machine and may as well do as they please in /tmp/. For Linerva we
>>>>>> may want to vary this behavior.
>>>>>>
>>>>>> It might also be nice to use upstream software. Any recommendations?
>>>>>>
>>>>>> Greg
>>>
>
