This all seems quite reasonable to me.
I will note that I do not like the current output of the progress bar
startup monitor. I would prefer that it state how many configurations
will be started, and then list each configuration together with its
startup time, one configuration per line. I think this would work fine
both foreground and background.
Thanks,
david jencks
On Nov 20, 2005, at 10:24 PM, John Sisson wrote:
I am reviewing Dave's startup patches ( GERONIMO-1166 ) and have a
couple of ideas for comment.
[1]. I propose we provide a geronimo.sh file that is modelled on
Tomcat's catalina.sh file (
http://svn.apache.org/repos/asf/tomcat/container/tc5.5.x/catalina/src/
bin/ ), as a large number of users would already be familiar with its
syntax and environment variable naming conventions and it would be
good if we had some consistency across Apache products.
geronimo.sh would support options such as:
jpda start - Start Geronimo under JPDA debugger
run - Start Geronimo in the current window (same as Dave's
proposed -foreground)
start - Start Geronimo in a separate window (same as Dave's
proposed -background)
Also make startup.sh consistent with Tomcat's startup.sh and move the
redirection logic and foreground/background logic to geronimo.sh.
If we are consistent with Tomcat it means that if an option isn't
passed to geronimo.sh (e.g. start) then usage information will be
printed to the terminal. If users invoke startup.sh, it in turn
invokes geronimo.sh with the start option (consistent with Tomcat).
I am happy to make these changes if I have no objections.
[2]. File name used for redirected output when using startup.sh
-background
Currently the patch redirects output to a startupProgress.log file.
I am thinking the file should be renamed to geronimo.out (consistent
with Tomcat's catalina.out) since it may contain more than startup
messages over the life of the process.
[3]. Improving format of progress messages in redirected output when
using startup.sh -background
For the startup output to not appear garbled in the file that output
is redirected to (due to the carriage returns generated by
ProgressBarStartupMonitor) we probably need a modified version of
ProgressBarStartupMonitor that outputs a line when a configuration is
starting/started (without the update thread that updates the line
approx every 500ms that the ProgressBarStartupMonitor has).
I initially thought we could use the -quiet option, but that results
in no progress being output and it would be nice to be able to look at
the geronimo.out file to see what is happening rather than having to
look through possibly heaps of messages in the log4j log files.
It would be also be helpful if the output redirected to the
geronimo.out file also has the summary of listening ports and started
application modules & web applications.
[4]. Proposed new Geronimo startup options:
-interactive (default)
Specify this when output is sent to an interactive
terminal/console. During startup (if -quiet is not specified) the
progress message for a configuration will be updated approx every
500ms (using carriage returns to move the cursor on the display to the
beginning of the current line to enable the progress message to be
updated. Mutually exclusive with the -noninteractive parameter.
-noninteractive
Specify this when output is being redirected to a file or
printer. During startup, a new message (each message on a new line)
will be issued during different stages startup. Mutually exclusive
with the -interactive parameter.
The above option could also be stored in case in the future we want to
enhance shutdown processing to show some progress messages.
The startup.sh script would pass -noninteractive if the process is
started in the background.
[5]. New method on StartupMonitor interface
A new method setInteractive(boolean b) could be added to the
StartupMonitor interface and invoked by the Daemon class before the
systemStarted(kernel) method is called.
Comments?
Thanks,
John
Dave Colasurdo wrote:
Have attached the patches for both unix (.sh) and windows (.bat)
environments to GERONIMO-1166. Please test them out..
Thanks
-Dave-
Dave Colasurdo wrote:
I've opened a JIRA for this issue and created a patch for the
windows platform. Still investigating the unix environment...
http://issues.apache.org/jira/browse/GERONIMO-1166
John Sisson wrote:
Hi Dave,
I don't think I had any objections to making the startup scripts
follow Tomcat as much as possible. See the following discussions
on scripts, I think there were a number of issues discussed that we
need to cover:
http://www.mail-archive.com/[email protected]/msg05926.html
http://www.mail-archive.com/[email protected]/msg05851.html
http://www.mail-archive.com/[email protected]/msg06483.html
Regards,
John
Dave Colasurdo wrote:
Jeff Genender wrote:
Dave Colasurdo wrote:
The shutdown scripts are a step forward in usability over
manually killing the java process via CTL-C. While quite
simple, CTL-C does not seem very user friendly and should not be
the default mechanism.
I really don't believe there is a default mechanism, IMHO. I
think we are offering multiple ways to do the same thing. The
CTRL-C would be heavily used by developers. The shutdown script
could be used by people using a daemon or backgrounding the
server (which is easily done on both Windows and *nix systems) or
a remote server. The console would/maybe be used by
mouse-clicking administrators.
I would surely hope that in a prod environment one is not running
the server in a terminal window ;-)
However, it does seem strange that a user needs to open a new
window to shutdown the server. Seems like the initial startup
command should return the command prompt back to the user so
that shutdown can be issued from the same window. One way to
accomplish this is to have the startup script launch a new
window that controls the java process (and output the startup
messages) while the initial prompt is returned to the user.
This would allow the shutdown to be issued from the initial
window.
For a developer (and me being selfish), running in a terminal
window is not strange and it seems to be the norm from a command
line perspective, rather than the exception.
IMHO, ss a developer, sending the server into the background is
not appealing. I think if one wants control over their terminal,
they could issue a startup.sh& (notice the ampersand) to
background the process. Quite possibly we could also add another
script called startup_background.sh (or bat) that could so this
as well. We could also create daemon scripts for the different
platforms. Wasn't there a JIRA issue for an NT Service for
Windows? We could add init.d scripts for Unix too.
I agree the current behavior is appropriate for a developer. I
was thinking more about end users. Similar to your suggestion,
should we consider adding an option to the startup.sh|bat script
to put the process in background? Actually, I'm wondering if the
default behavior (startup.sh|bat w/o any options) should be geared
toward end users and would run the process in background. And
specifying the option (-foreground) would allow the process to be
run in the current window for developers.
Of course, windows service and init.d are also useful. I think
both proposals are worth pursuing
Will look to see if there are current JIRAs open on these..
Also, if we ever support sharing one binary installation that
can start multiple instances of geronimo (each with it's own
unique configuration) then we will also likely need this
behavior.
-Dave-
