I have been working on getting some of my customers interested in using
Avalon and Fortress here in Japan. So far there has been a lot of interest.
One comment that came up, which I have also thought about as well, is why
Fortress continues to attempt to initialize additional components after one
has failed.
For example I have an application which defines a DataSource component
which is then looked up by 8 other components. When things are configured
correctly, all works great. But if there is any problem with the
configuration
of the DataSource component, Fortress will kick out an impressive chain of
stack traces that fills a couple hundred lines in a log file. The log output
gets more impressive the deeper in the dependency tree an error occurs.
Fortress will correctly print a stack tract for the problem in the
DataSource
component. But it will then continue on and attempt to configure each of the
other components as well. All 8 of the components which depend on the
DataSource will also print out stack traces when they fail to look up the
DataSource.
As I recall, ECM behaved this way as well. My question is there a reason
why the container should not be giving up on the first failure and exiting?
It seems like it should be the responsibility of an individual component to
catch an exception during initialization if it should not stop the container
for completing its initialization.
If there are reasons for doing this what would you think of the idea of
adding a way to configure Fortress so that the container's initialization
will immediately fail if any of its components' initialization fails. I
would
also like to be able to configure things so that the exception message
which causes initialization to fail is displayed in the console/log. But be
able to avoid showing large stack traces.
The stack traces have been fine when I have installed applications myself.
But I have been running into problems lately trying to explain them away
to customers who simply make a typo in the config file while deploying
an application themselves.
I can of course work on this myself, but I wanted to understand the issues
which led to the way things are currently implemented.
Cheers,
Leif
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
