users@glassfish.java.net

Re: GlassFish v3 logging one pager ready for review

From: Dies Koper <diesk_at_fast.au.fujitsu.com>
Date: Thu, 5 Feb 2009 10:35:38 +1100

Hi Carla,

I forgot one thing.
Would you consider issue 3466 for inclusion?

https://glassfish.dev.java.net/issues/show_bug.cgi?id=3466

The suggestion is to put the most important fields logged to the
server.log first:
<timestamp> <message level> <message> <thread ID> <logger> <APS version>

The main point being that to most users the message text is more
important than the thread ID, APS version, etc.

Thanks,
Dies


Dies Koper wrote:
> Hi Carla,
>
> I see that you have already received a great number of useful comments.
> I have a few that I believe were not mentioned yet.
>
> 1. Currently logged messages are preceded by a timestamp. Will you take
> care to use the timestamp that the error occurred (i.e. the log method
> was called), and not the time that the message is released from the
> queue and actually written to the log file?
>
> 2. Will there be an option to have logging done synchronously instead
> (or flush per message setting, configurable queue size of 0, etc.)?
> I'm thinking of two cases were that would be useful:
> - Lots of messages could be lost when a VM crashes. We'd like to have
> some control over how much could be lost (maybe start with a queue size
> of 0 and slowly increase it as we feel the need for a performance gain
> and feel confident about the server's stability).
> - In test applications we log to the server log and at the end of the
> test we check the log for a "PASS" message, and that no unexpected
> messages were logged. Asynchronous logging makes automation of such test
> difficult.
>
> 3. Finally, we think it is very important to have a limit on the queue
> size and when it is filled up, to have the log methods block until there
> is space enough again. (Your pager is not clear about what code is
> blocked. You are talking about the log methods, not the asynchronously
> running logger thread, right?). An unbounded queue would lead to
> out-of-heap errors eventually. In fact, it might even be desirable to
> log a WARNING message when the queue is filled up, to assist the admin
> in tuning the queue size.
>
> Thanks,
> Dies
>
>
> glassfish_at_javadesktop.org wrote:
>> The one-pager for GlassFish v3 logging is now available for review
>> and comment. The official review period is one week, so please post
>> your comments and feedback in the form of reply to this mail thread
>> by 30 Jan 2009. We will try to consider feedback that arrives after
>> that but we cannot guarantee we will be able to do so.
>>
>> You can find the one pager here:
>>
>> http://wiki.glassfish.java.net/attach/V3FunctionalSpecs/GFv3Logging-onepager.html
>>
>>
>>
>> Generic v3 functional specification page is:
>>
>> http://wiki.glassfish.java.net/Wiki.jsp?page=V3FunctionalSpecs
>>
>> Thank you in advance for your comments.
>>
>> Carla. [Message sent by forum member 'carlavmott' (carlavmott)]
>>
>> http://forums.java.net/jive/thread.jspa?messageID=327898