[Freeipa-devel] [PATCH] 0014 Add final debug message in installers

[John Dennis]

On 04/16/2012 04:15 PM, Rob Crittenden wrote:
> John Dennis wrote:
>> On 04/16/2012 01:31 PM, Rob Crittenden wrote:
>>> John Dennis wrote:
>>>> On 04/13/2012 06:25 AM, Petr Viktorin wrote:
>>>>>> When the utility sets logging to console, the extra log message gets
>>>>>> printed out there. I agree this isn't optimal.
>>>>>> Attached patch removes the console log handler before logging the
>>>>>> result.
>>>>>>
>>>>>
>>>>> I read through log_manager, and found that I can do this more cleanly
>>>>> with remove_handler.
>>>>>
>>>>> John, is this a good use of the API?
>>>>
>>>> The problem you're trying to correct is that under some circumstances a
>>>> log message is emitted twice to the console, right?
>>>>
>>>> Removing the console handler fees like the wrong solution, it's a pretty
>>>> big hammer, you have to know when to remove it and once removed any
>>>> expectation that messages will appear on the console will be untrue.
>>>>
>>>> Can you give me a brief explanation as to why you're getting duplicate
>>>> messages on the console. I wonder if there isn't a better way to handle
>>>> the problem which isn't so invasive and potentially hidden.
>>>>
>>>> Or is the issue you don't want a console handler at all? If that's the
>>>> case then maybe we should provide a configuration that does not create
>>>> one, that way it will be explicit and obvious there is no console
>>>> handler.
>>>>
>>>> FWIW, the current configuration of logging is historical dating back to
>>>> the beginning of the project. When I added the log manager the intent
>>>> was to fix deficiencies in logging, not to modify the existing behavior.
>>>> I wasn't clear to me the existing configuration was ideal. If you're
>>>> hitting problems because of the existing configuration perhaps we should
>>>> look at the historical configuration and ask if it needs to be modified.
>>>
>>> This patch is standardizing logging the final disposition of a number of
>>> commands. Currently is must be inferred.
>>>
>>> The problem is that some commands have had this disposition added but
>>> open no log files so some error messages are being displayed twice and
>>> in log format.
>>>
>>> I think the easiest solution would simply be to scale this back to only
>>> those commands that open a log file at all.
>>
>> You say "command" but command is a loaded word :-) I presume you mean
>> command line utility and not an IPA Command object. The reason I'm
>> drawing this distinction is because the environment Commands execute in
>> are not supposed to change once api.bootstrap() has completed.
>>
>> Who or what is aggregating the final disposition of a number of
>> commands? The reason I ask is because the log_mgr object is global and
>> deleting a handler from a global resource to satisfy a local need may
>> have unintentional global side effects. How is the aggregation
>> accomplished?
>>
>>
>
> This patch is in the context of the command-line utilities like
> ipa-server-install, ipa-client-install, etc. and not the ipa tool. Some
> of them initialize the api, some do not.

This is exactly the type of problem I'm concerned about, a conflict over 
who owns the logging configuration.

These multiple potential "owners" of a global configuration is what lead 
me to introduce the "configure_state" attribute of the log manager, to 
disambiguate who is in control and who initialized the configuration. In 
fact bootstrap() in plugable.py explicitly examines the configure_state 
attribute of the log_manager and if it's been previously configured (for 
example because the api was loaded by a command line utility that has 
configured it's own logging) it then defers to the logging configuration 
established by the environment it's being run in (in other words if a 
command line utility calls standard_logging_setup() before loading the 
api it wins and the logging configuration belongs to the utility, not 
the api).

All of this is to say that isolated code that reaches into a global 
configuration and takes an axe to it and chops out a significant part of 
the configuration that someone else may have established without them 
knowing about it feels wrong.

I suspect there is some deeper problem afoot and unilaterally chopping 
out the console handler is not addressing the fundamental problem and 
likely introduces a new one. But I really need to look at the specific 
instances of the problem, maybe if I dig through the history of this 
patch review more carefully I'll discover it, but if someone wants to 
chime in an point me at the exact issue that would be great :-)

> As I said, the intention of my original request was to log the
> disposition of an install command. I think this patch goes too far and
> updates all admin utilities.
>
> rob


-- 
John Dennis <jdennis at redhat.com>

Looking to carve out IT costs?
www.redhat.com/carveoutcosts/