I used a similar trick in XBean and it worked well. The problem I
was having was, as you are starting the server different problems can
occur, some are fatal and some are not. In my original code I had
system exits spread thought the boot code, which made it competely
untestable. As soon as I introduced an FatalStartupException, I
could test the code. My main method simply call the boot code,
catches a fatal exception, prints the message, and exits with the
exit code within the exception class. IIRC, I set it up so the user
needs to pass a flag to get the exception stack trace to print.
Anyway, I suggest we introduce a logical exception hierarchy for
startup errors, which capture the most common problems such as a bind
error. Then the boot code can decide (based on user flags maybe) if
the error should be treated as fatal.
-dain
On Nov 30, 2005, at 3:52 PM, John Sisson wrote:
Currently if the start of a GBean fails, an error with a stack
trace is written to the log.
Sometimes the cause of the GBean failure is a common user error
that does not require a stack trace for the user to rectify the
problem. An example of this is where a port is already in use and
we get a BindException. All the user really needs to be told is
that Jetty or Tomcat was attempting to use port x and it was
already in use.
Since BindExceptions are a common problem that new users may run
into the first time they start Geronimo on their system, seeing a
stack trace IMHO is not a good impression. I have heard many
people comment that they like Geronimo's startup because it doesn't
output heaps of unncessary messages. It would be nice if it still
operated that way when users make a common mistake.
Could we introduce a new exception class e.g.
GBeanLifecycleException that can be optionally used by GBeans (such
the JettyConnector shown below) and change the code in the catch
block of GBeanInstanceState.attemptFullStart(..) to check if the
exception is an instance of GBeanLifecycleException and if so, call
the method on the exception that returns whether the stack trace
should be logged.
Comments or other ideas on how we could we improve the current code
GBean code so that the start of a GBean can "optionally" result in
a message being logged without a stack trace being logged.
An example of one of the stack traces the user would currently see:
09:08:35,375 ERROR [GBeanInstanceState] Error while starting; GBean
is now in the FAILED state: objectName="geronimo.server:J2EEAppl
ication=null,J2EEModule=org/apache/geronimo/
Jetty,J2EEServer=geronimo,j2eeType=GBean,name=JettyWebConnector"
java.net.BindException: Address already in use: JVM_Bind
at java.net.PlainSocketImpl.socketBind(Native Method)
at java.net.PlainSocketImpl.bind(PlainSocketImpl.java:331)
at java.net.ServerSocket.bind(ServerSocket.java:318)
at java.net.ServerSocket.<init>(ServerSocket.java:185)
at org.mortbay.util.ThreadedServer.newServerSocket
(ThreadedServer.java:391)
at org.mortbay.util.ThreadedServer.open(ThreadedServer.java:
477)
at org.apache.geronimo.jetty.connector.JettyConnector.doStart
(JettyConnector.java:203)
at
org.apache.geronimo.gbean.runtime.GBeanInstance.createInstance
(GBeanInstance.java:936)
at
org.apache.geronimo.gbean.runtime.GBeanInstanceState.attemptFullStart(
GBeanInstanceState.java:325)
<SNIP>
Thanks,
John
