Hi Michael,
That is correct. However, AppenderSkeleton.doAppend is synchronized which prevents
simultaneous access to CyclicBuffer. Thus, SMTPAppender is safe even in a
multi-threaded environment.
The configuration related methods such as setOption, activateOptions etc. are not
synchronized either. This is deemed safe because the PropertyConfigurator and the
DOMConfigurator act on new instances of appenders and these appenders are available
only after they are configured because the category where they are attached to is
synchronized during configuration. A locked category cannot log due to the
synchronization in the Category.callAppenders method.
However, before creating and configuring the new appender instance, these
configurators close all attached appenders to a category mentioned in the config file.
One possible bug is that a configurator closes an appender while the appender is in
the middle of an append operation. See
http://nagoya.betaversion.org/bugzilla/show_bug.cgi?id=509 for an example. The
solution is to have the close method in appenders as synchronized.
I think the assumptions made about how appenders are accessed and configured should be
better documented. The fact of the matter is that appenders are thread-safe (with the
exception of bug 509) even if they are not synchronized all over the place. I hope
this clarifies the matter. Ceki
At 10:09 06.02.2001 +0100, you wrote:
> Hello,
>
> I looked through the code of the SMTPAppender, and it seems like there is no
>synchronization their. For example, in the append() method, adding element to the
>CyclicBuffer is not synchronized. CyclicBuffer itself has no internal synchronization
>either.
>
> Michael
>
----
Ceki Gülcü e-mail: [EMAIL PROTECTED] (preferred)
av. de Rumine 5 [EMAIL PROTECTED]
CH-1005 Lausanne
Switzerland Tel: ++41 21 351 23 15
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
