Re: [atomic-devel] cron job for logrotate on containers

[Daniel J Walsh <dwalsh redhat com>]

On 12/17/2014 10:32 AM, Lokesh Mandvekar wrote:
> On Tue, Dec 16, 2014 at 12:15:24PM -0500, Andy Goldstein wrote:
>> On Dec 16, 2014, at 12:12 PM, Colin Walters <walters verbum org> wrote:
>>
>>> On Tue, Dec 16, 2014, at 11:10 AM, Lokesh Mandvekar wrote:
>>>> I have a script which would run logrotate on all running docker containers.
>>>> Plan is to have this installed via the docker rpm into /etc/cron.daily. Posting here 
>>>> for further discussion as per Colin's suggestion before I include it in any rpms.
>>>>
>>>> ---
>>>> #!/bin/sh
>>>>
>>>> for id in $(docker ps -q); do
>>>>    docker exec $id logrotate -s /var/log/logstatus /etc/logrotate.conf >/dev/null 2>&1
>>>> done
>>>> exit 0
>>> I don't like that this presumes I have logrotate inside my containers.  If I have
>>> configured my containers to be microservices (potentially using a custom "thin"
>>> base instead of the present "fat" base) using remote logging, then I don't need this.
>>>
>>> I get that this is trying to address a real problem, but I feel like with this
>>> we're encouraging people to do the wrong thing.
>>>
>>> I also think cron is just a hack =/  There are race conditions here if
>>> the container exits before the exec runs, and I know we ignore that
>>> with >/dev/null, but still.
>>>
>>> This seems like something to add to *documentation* around container
>>> logging as an option if you're using "containers as VMs" model.  But
>>> not something to turn on by default for every user of the Docker RPM.
>>>
>> +1, this should be optional and disabled by default
> We could add this to a 'docker-logrotate' optional subpackage along with a
> doc describing issues. Would that work for all?
>
I think this is the easiest method, then users who want logrotate can
simply install the package.