Jason,
I finally got around to looking at this. It's really cool!
This is a great way to address the environment and java opt issues in a
more dynamic way to plugins to potentially exploit.
In addition to some of the items David mentioned it would also be great
to have some more query capabilities both in gsh and the start-server
script. For example; the ability to list all environment variables,
list system or geronimo properties set, list javaopts, list of all
scripts to be run prior to the starting server, etc...
One curious thing I noticed was that it seemed to drag a bit when I
terminated geronimo from the console. It might have nothing to do with
gsh but it seemed to take a long time before the gsh showed the server
finally terminated ... just fyi and fwiw ... may be nothing to do with gsh.
Great work Jason!
Joe
Jason Dillon wrote:
Hiya folks, I finally got around to finishing up my POC of using GShell
to launch the Geronimo Server and I have committed the bits that make it
work to server/trunk. The new module which contains the GShell command
implementation (and support) classes is:
modules/geronimo-commands
And a new assembly which has the GShell bits all in place for folks to
test with is:
assemblies/geronimo-jetty6-javaee5-gshell
The assembly is not hooked up by default, but the code module is. So,
to test it out, you need to do something like:
svn co https://svn.apache.org/repos/asf/geronimo/server/trunk server
cd server
mvn
assemblies/geronimo-jetty6-javaee5-gshell
mvn
Then unzip the assembly:
unzip target/geronimo-jetty6-javaee5-gshell-2.1-SNAPSHOT-bin.zip
And then finally try it out. First lets get the basic GShell
command-line help:
./geronimo-jetty6-javaee5-gshell-2.1-SNAPSHOT/bin/gsh --help
This should spit out something like:
<snip>
____ ____ _ _ _
/ ___/ ___|| |__ ___| | |
| | _\___ \| '_ \ / _ \ | |
| |_| |___) | | | | __/ | |
\____|____/|_| |_|\___|_|_|
GShell -- Geronimo command-line shell
usage: gsh [options] <command> [args]
-n,--non-interactive Run in non-interactive mode
-D,--define <name=value> Define a system property
-V,--version Display GShell version
-c,--commands <string> Read commands from string
-debug,--debug Enable DEBUG logging output
-h,--help Display this help message
-i,--interactive Run in interactive mode
-quiet,--quiet Limit logging output to ERROR
-verbose,--verbose Enable INFO logging output
</snip>
Then lets run the interactive-shell:
./geronimo-jetty6-javaee5-gshell-2.1-SNAPSHOT/bin/gsh
This should leave you at a very plain prompt at the moment:
<snip>
> _
</snip>
At this point you can type something like this to list all of the known
commands:
<snip>
help commands
</snip>
Which should spit back:
<snip>
Available commands (and aliases):
start-server ( start )
set
exit ( quit, bye )
unset
source
help
</snip>
The command that I added is called 'start-server', with an alias of
'start'. This is basically an augmented and enhanced version of what
the geronimo:start-server goal does in the geronimo-maven-plugin. To
see its options run this:
<snip>
start-server --help
</snip>
And it says:
<snip>
start-server -- Starts a new Geronimo server instance
usage: start-server [options]
-A,--javaagent <jar> Use a specific Java Agent, set to
'none' to
disable
-D,--property <name=value> Set a system property
-H,--home <dir> Use a specific Geronimo home directory
-J,--javaopt <option> Set a JVM flag
-b,--background Run the server process in the background.
-h,--help Display this help message
-j,--jvm <dir> Use a specific Java Virtual Machine
for server
process
-l,--logfile <file> Capture console output to file
-m,--module <name> Start up a specific module by name.
-q,--quiet Suppress informative and warning messages
-t,--timeout <seconds> Specify the timeout for the server
process in
seconds
-v,--verbose Enable verbose output; specify
multipule times
to increase verbosity.
</snip>
NOTE: Use -vv for --veryverbose ;-)
And then give it a whirl and try to start the server up from the shell:
<snip>
start-server
</snip>
or if you prefer more output:
<snip>
start-server -v
</snip>
And so on...
This will actually create a newly forked JVM to run the server in, and
will eventually have a robust node manager thingy, but I've not done
that yet ;-)
The platform scripts are now super simple!!! All of the logic is now in
the command implementation. And eventually we can probably have the
geronimo-maven-plugin just invoke the command so that *everything* uses
the exact same method for launching the server process.
* * *
As requested by Jeff, I added support to read in some scripts to augment
the launching of the server... so that plugins can add properties and
such easily. Right now this is the directory which is inspected for
scripts:
etc/rc.d
And currently the scripts need to be named like this:
<command-name>,<custom>.groovy
I've created a default one for you to look at:
etc/rc.d/start-server,default.groovy
Which simply sets the max heap size to 512m:
<snip>
command.javaFlags << '-Xmx512m'
</snip>
When running the start-server command (or its aliases) all of the
etc/rc.d/start-server,*.groovy scripts are run (if any) before the
process is launched to allow the command.properties, command.javaFlags
and other bits to be augmented.
These are Groovy scripts so you can also execute any arbitrary logic to
perform more complex setup muck if needed.
* * *
For now I just dropped all of the jars required for this into lib/gshell
and left lib/* alone, since the command uses the normal invoke of java
-jar server.jar and it requires bits in lib/* to work. Though we may
eventually want to setup the server.jar to use a classpath into
repository/* and then leave the lib/* only for the core launcher and
bits at some point in the near future.
This adds about ~4m to the assembly at the moment, though I've not tried
to reduce this much, but I'm sure it can be done to reduce foot print.
Some may be to simply have the GShell loader pull bits from the
repository and/or shading jars to only include what is really needed.
But I'm going to leave that for later.
Also, we can probably also convert all of the other cli tools into
GShell commands too, which will further simplify the platform scripts
and keep things uniform, and also a wee bit easier to standardize
testing off too ;-)
In the assembly I left most of the scripts as they were, except for
startup* and added gsh*. The gsh* scripts are the main scripty bits,
but they are very simple. And the new startup* scripts are simply
delegates to gsh to invoke the "start-server" command.
* * *
Anyways, enough for now... please take a look, ask questions, comment,
whatever. I hope we can start an easy dialog about how we can make this
basic style of launching and cli muck the standard for 2.1.
Cheers,
--jason
