getting rid of all-machines.log

[Kapil Thangavelu]

if we stopped redirecting the log on linux in our upstart jobs, log start
would rotate the per agent/job log files for us.



On Thu, Aug 14, 2014 at 11:49 AM, Matt Rae <matt.rae at canonical.com> wrote:

>
> On Thu, Aug 14, 2014 at 8:23 AM, Nate Finch <nate.finch at canonical.com>
> wrote:
>
>> The front page of 12factor.net says "offering maximum portability
>> between execution environments" ... that's exactly what I'm going for.
>>
>> We're going to support Windows. Windows does not have rsyslog or
>> logrotate.  We *have* a solution which is cross platform.
>>
>> My main concern is that if we use rsyslog on linux and something else on
>> Windows, now we're supporting and maintaining two things, when they could
>> both be the same.  Why support two things when you can support just one?
>>
>>
> I'm sure writing to stdout and redirecting to a logfile that gets watched
>> by rsyslog is the linux devop standard.  But it has also proved annoyingly
>> complex for us to configure and maintain. We've wasted several developer
>> *weeks* of time getting it to work right.
>>
>> Why would we continue to maintain the rsyslog stuff when it has been such
>> trouble for us?  Just because it's the standard, doesn't mean it's the
>> right tool for this specific job.
>>
>>
> I talked with Ian about this a little this morning, and my suggestion was
>> that we give users a way to turn off our in-app log rotation and let them
>> set it up themselves external to the app (with rsyslog or whatever they
>> like).... I just don't want to have to configure it or support it.
>>
>
> Yeah this sounds like a good suggestion. Maybe we can solve the problem in
> a cross platform way but still have options to log to stdout/syslog in case
> a user wants to integrate with an existing logging system themselves. A lot
> of applications have options of logging to stdout or to syslog depending on
> which you prefer.
>
> Principle of least astonishment and taco bell programming [2] are a couple
> ideas that might be good to keep in mind when deciding how we do it.
>
> [1] http://en.wikipedia.org/wiki/Principle_of_least_astonishment
> [2] http://widgetsandshit.com/teddziuba/2010/10/taco-bell-programming.html
>
>
>>
>>
>> -Nate
>>
>>
>>
>>
>>
>>
>> On Thu, Aug 14, 2014 at 9:04 AM, Gustavo Niemeyer <
>> gustavo.niemeyer at canonical.com> wrote:
>>
>>> As a side note, MongoDB offers a "capped collection" mechanism with
>>> the semantics that you can insert rows at will, and it rolls around
>>> automatically by replacing oldest entries with the newest ones. This
>>> tends to be a very convenient way to do structured logging, both on
>>> the writing and on the reading side. Using that as a general rsyslog
>>> might be unwise, as there are a number of details to get right and the
>>> volume may get wild depending on the application, but at least for the
>>> juju implementation itself it might be quite convenient. Imagine being
>>> able to ask "please tell me the output of the last run of the
>>> db-relation-joined hook on unit db/3".
>>>
>>> On Wed, Aug 6, 2014 at 11:24 AM, John Meinel <john at arbash-meinel.com>
>>> wrote:
>>> > all-machines.log is where we aggregate the messages from all
>>> > machines/units/etc.
>>> > It is likely to get big, which is why we want log rotate, but if you
>>> want to
>>> > be able to "juju debug-log" and actually get the feed of everything
>>> that is
>>> > going on, that needs to be *somewhere*. And yes, we'd like to switch to
>>> > something like log stash instead, but until we get there we do still
>>> need
>>> > it.
>>> > John
>>> > =:->
>>> >
>>> >
>>> >
>>> > On Wed, Aug 6, 2014 at 4:13 PM, Tim Penhey <tim.penhey at canonical.com>
>>> wrote:
>>> >>
>>> >> On 06/08/14 16:11, Nate Finch wrote:
>>> >> > all-machines.log seems both redundant and a ticking time bomb of
>>> disk
>>> >> > space usage.  Do we really need it?  Can we drop it and maybe later
>>> >> > schedule some time to use something like logstash that is both more
>>> >> > featureful and is cross platform compatible (unlike rsyslog)?
>>> >>
>>> >> not yet...
>>> >>
>>> >> debug-log uses all-machines.log, we cannot get rid of it at this
>>> stage.
>>> >>
>>> >> We can't drop it until a replacement is in place.
>>> >>
>>> >> Tim
>>> >>
>>> >>
>>> >> --
>>> >> Juju-dev mailing list
>>> >> Juju-dev at lists.ubuntu.com
>>> >> Modify settings or unsubscribe at:
>>> >> https://lists.ubuntu.com/mailman/listinfo/juju-dev
>>> >
>>> >
>>> >
>>> > --
>>> > Juju-dev mailing list
>>> > Juju-dev at lists.ubuntu.com
>>> > Modify settings or unsubscribe at:
>>> > https://lists.ubuntu.com/mailman/listinfo/juju-dev
>>> >
>>>
>>> --
>>> gustavo @ http://niemeyer.net
>>>
>>> --
>>> Juju-dev mailing list
>>> Juju-dev at lists.ubuntu.com
>>> Modify settings or unsubscribe at:
>>> https://lists.ubuntu.com/mailman/listinfo/juju-dev
>>>
>>
>>
>> --
>> Juju-dev mailing list
>> Juju-dev at lists.ubuntu.com
>> Modify settings or unsubscribe at:
>> https://lists.ubuntu.com/mailman/listinfo/juju-dev
>>
>>
>
> --
> Juju-dev mailing list
> Juju-dev at lists.ubuntu.com
> Modify settings or unsubscribe at:
> https://lists.ubuntu.com/mailman/listinfo/juju-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/juju-dev/attachments/20140814/749a19b5/attachment.html>