[MacLoggerContest] Journalling

[Jonathan G0DVJ]

Interesting discussion ...

Are we mixing up two meanings of Journalling?  Maybe John mentioned it 
in the general sense of keeping a journal record of QSO data so that 
nothing is lost.  i.e. at the application level.  There is also the OSX 
specific meaning of Journalling which was introduced in Panther and 
which is a system-wide volume configuration aspect under admin user 
control at disk set-up time, and not the application.   Not sure if 
Jack is alluding to this in his posting?

I agree that the simplest way seems to be to write then flush from the 
application's viewpoint.  However maybe this is another area where we 
should just state what the user experience should be and leave it to 
Don to decide how best to implement things to achieve it, like Jack 
rightly pointed out when I mentioned multi-threading in an earlier 
post!

I think we all agree with John's original point about losing nothing 
from the committed log - Most other systems I have used cannot 
guarantee that the current QSO still being entered (i.e. not 
completed/committed) won't be lost, but that is the most one can lose.

73,
Jonathan.
--

On Mar 2, 2005, at 6:44 am, Jack Brindle wrote:

>
> On Feb 28, 2005, at 1:56 PM, John Bastin wrote:
>
>> Automatic journaling, so even if you have a power failure, NOTHING is 
>> lost.
>
> This grabbed my attention - I'd like to drill into it a bit to 
> understand exactly what is being asked for and why. More importantly, 
> I want to understand the current need for journalling, because I don't 
> think I understand it properly now.
>
> Under MacOS X information written to files may be immediately flushed 
> to disk. When writing a log file, the data may be appended to the log 
> file and saved to disk immediately. As I understand journalling, the 
> information is written to the journal file, then to the log file. In 
> this case a power failure before the journal write would lose the 
> entry, while one in-between the two writes will simply cause a 
> journal-to-log file update on restart. But, the information that would 
> be appended to the log file could have been handled in place of the 
> journal write, taking care of the whole thing at once. In both cases, 
> power failures before the first write completes causes the entire 
> entry to be lost, while a failure after the first write completes just 
> causes the operator to be rather unhappy.
>
> It seems that the simplest way to handle things would be to write the 
> log entry to the file and immediately flush the file to disk. Is this 
> too simple? What am I missing?