Well... me thinks I've finally gotten over the major hump...

I had been using Mina 2.x (which is in pre-release right now)... I was using it because it has some synchronous request/response handling muck already there which I kinda needed to implement the remote shell proxy muck. And well, I must have spent like a week digging through code, running tests, rebuilding my tree, and really... going crazy because no matter what I did nothing seemed to behave itself as I needed it to.

So I ported and basically re-wrote all that request/response stuff I wanted so I could add debug to it and understand why it was not happy... or try to at least. It really seemed like there was some kind of threading problem or a sync problem but all of the diag I got from logs just made no sense and the results I get from running tests were random at best.

Ugh... so anyways, I wanted to use Mina 2.0 for the request/response stuff... which I rewrote anyways... so a little before midnight on the night before the 20th I decided to move Mina 1.1.2 (latest official release). Updated some apis, hacked around a few other bits... and about 10 or 10 minutes into the 20th I had the tcp transports all ported and... um working!!! for the normally async, req/resp and for the stream i/o all at the same time. It was a nice birthday present... such a good feeling to finally get this crap working after going mad over it over the past few days!!! :-)

And now over the days past I've fixed up the vm:// transport and fixed some bugs in the req/resp handling which were causing things to puke up insanity in a oh so blissful random form.

Um... ya, so I think I've finally mastered... er well, beaten to submission, mina... which I still really like a lot btw... And I've implemented some framework to make using it a little easier for me, probably going to invent some more as I go... but right now... um... it seems to work. Tcp, ssl, vm... all working happy, happy. Message passing along side of stream i/o sitting on top of a few request/ response bits to boot.

Oh ya... and well, you can actually get a remote shell with this stuff now. Its not 100% functional, but you can connect to a remote server (running a rsh-server command), authenticate (internally it will do some rsa key exchange muck too) and then well, you are running commands on the remote system. And the output comes back to you, and you can 'exit' the remote shell and get back to your local shell. Wee.

Its still not done though, all this batting with mina had kept me from finishing up the gshell rsh protocl bits and finishing up the state management and such... but I think I'm past that now, so I really hope to get these loose ends tied up soonish. Its still a bit POC, I'm still learning and adapting to how best to implement this stuff, but I expect the first version to be a "just make it work" and then after that a bit of refactor to clean things up (which really has how most of the gshell implementation has been going... sans the first week of insane code drops when I wrote most of it).

 * * *

Okay, I'll stop babbling now... my point is... the remoting crapo works... and I think that maybe in another week or so there should be something functional enough for me to integration back into Geronimo. Something like a gsh rsh into a running server (over ssl with authentication) and then maybe a few stupid commands to show vm stats, shutdown or something else. I guess since some peeps are keen for scripting muck I can show how you can use the script command to execute a script on a server too ;-)

I'm still working on the 'makecoffee --now' command...

:-P

Oh, and lastly... and then ya... I'll shut up... If anyone is even remotely interested in any of this crap it would be nice to have another mass of grey matter pondering over some of its present and future...

Cheers,

--jason


PS. Sorry, I couldn't resist... I think that the stuff in there now can probably support a telnet as well as ssh transport impl too. Though I've yet to find an ASL friendly java ssh client+server library to build it with...



On Sep 18, 2007, at 3:09 PM, Jason Dillon wrote:

Hiya folks... just a quick update on GShell muck...

I've been madly hacking on the remote shell support for the past week and well, its really close to being functional. I initially was going to implement a POC over ActiveMQ but that was pissing me off soo much that I decided to peek at what Jeff had done with GShell and discovered how incredibly cool Apache Mina is!! So I've based the remote-shell client/server bits off of Mina and its working out fairly well, though I had to go and learn about it (which was actually fun-ish, sorta, okay not really so much :-P ).

Anyways, I've dug through GCache (which really did help me to understand Mina) as well as some other Mina-based projects and I've whipped up a fairly simple system (er I think so), which is extensible supports TCP, SSL and VM (er well almost VM), so you can do something like:

    ./bin/gsh rsh ssl://someserver:9999

or

    ./bin/gsh rsh tcp://someserver:9999

The same is true for the server side, to be able to accept these rsh commands you could:

    ./bin/gsh rsh-server ssl://localhost:9999

or

    ./bin/gsh rsh-server tcp://localhost:9999

etc...

I've implemented a RSA-based client-server handshake (similar to what gcache was doing, actually based on what it was doing) to support a decent authentication mechanism. Might eventually hook up some JAAS crapo later, but... ya later.

Anyways, right now I'm trying to figure out some small problems with the stream message passing and marshaling, which I hope to get resolved really damn soon, cause this is fricken annoying me. And once that is sorted, a few minor clean up things and the rsh/rsh- server bits should be functional enough to actually use.

More to come...

Cheers,

--jason



On Sep 8, 2007, at 12:40 PM, Jason Dillon wrote:

A little bit more insight into what I'm thinking of doing... since some of you can't read minds to well :-P

I'd like to convert all of the assemblies to basically look like what the assemblies/geronimo-jetty6-javaee5-gshell produces.

And then I'd like to start converting the other cli bits to gshell command impls, like: deployer, client and shutdown.

And then (maybe around the same time or before the above), I'd like to adapt the gshell of target jvm bits to load jars from the repository, instead of using the lib/* bits.

A little background for those who haven't looked at assemblies/ geronimo-jetty6-javaee5-gshell and what it produces from a lib/* perspective. Right now I've set up the assembly to produce:

    geronimo-jetty6-javaee5-gshell-2.1-SNAPSHOT/lib
    geronimo-jetty6-javaee5-gshell-2.1-SNAPSHOT/lib/boot
    geronimo-jetty6-javaee5-gshell-2.1-SNAPSHOT/lib/endorsed
    geronimo-jetty6-javaee5-gshell-2.1-SNAPSHOT/lib/gshell

Where the bits in lib/* and lib/endorsed/* are the same as they were before. The bits in lib/boot/* and lib/gshell/* are specific to gshell. And normally a gshell installation would have everything I put into lib/gshell/* into lib/*, but I moved them to a sub dir for now... since the bin/*.jar's load jars from the ../ lib/* dirs.

The lib/boot/* stuff is the very minimal gshell bootstrap classes, which setup up the other happiness... and let you do things like:

java -jar ./geronimo-jetty6-javaee5-gshell-2.1-SNAPSHOT/lib/ boot/gshell-bootstrap.jar

And that will give you a nice shell... or

java -jar ./geronimo-jetty6-javaee5-gshell-2.1-SNAPSHOT/lib/ boot/gshell-bootstrap.jar start-server

That will launch the G server process using all of the right - Djava.ext.dirs and whatever properties that we currently have hacked into platform scripts.

Anyways, so the idea is to move all of the bits which are current in the lib/* into the repository, and then configure the gshell command impl to load put the correct dependency artifacts onto the classpath of the target jvm that is booted up. This will augment the existing kernel bootstrap from repo stuff, putting evertying except what is needed from gshell into the repository...

And really, what I'd like to eventually get to is having the bootstrap from the repository... so that everything except for what is now it lib/boot/* and lib/endorsed/* can live in the repository like happy little communistic jars should be :-P

 * * *

And then there are longer term things for GShell...

Remote administration (via, telnet, ssh, or custom ssl protocol... last is most likely to actually happen soonish)

Process management, which is great for clusters, or staging -> production management. A full suite of command-line tools which can manage the configuration of a server... easily. So, for example, lets say you've got a configuration that is working really well for you... but you want to play with something new...

So you might:

    ./bin/gsh backup-configuration before-mucking
    ./bin/gsh start-server

And then go and change a whole bunch of stuff... and it doesn't work... yikes... so rollback...

    ./bin/gsh backup-configuration hosed-server
    ./bin/gsh restore-configuration before-mucking
    ./bin/gsh start-server

And then maybe you want to play with the "hosed-server" configuration again...

    ./bin/gsh start-server --configuration hosed-server

Of course, all of these could have been run from a single ./bin/ gsh, but just for clarity, you can run them one off too.

Maybe list or mange the configurations

    ./bin/gsh list-configurations
    ./bin/gsh remove-configuration some-unwanted-config
    ./bin/gsh copy-configuration default some-new-config

The sky is the limit really... for what kind of management we can do...

Lets say you wanted to do the above on a remote node?

    ./bin/gsh remote-shell someserver:9443
    Connecting to someserver:9447...
    Connected

    username: system
password: **** (remember this is all jline, so we can mask passwords like one woudl expect)

    someserver:9447 > list-configurations
    someserver:9447 > remove-configuration some-unwanted-config
    someserver:9447 > copy-configuration default some-new-config

So, all of these operations would happen on the node named "someserver" listening on 9443 (over ssl of course). Or how about you want to reboot a server remotely?

    someserver:9447 > restart-server now
    Geronimo server shutting down...
    ....
    Geronimo server shutdown.
    Geronimo server starting...
    ...
    Geronimo server started in ...

Since GShell manages the processes its really easy to perform a full restart of a Server w/o needing magical platform scripting muck. And it will just work the same on each platform too.

Once we have clustering, then we can do the same kinda thing for an entire cluster of nodes...

    someserver:9447 > restart-cluster now
    Shutting down 2 nodes...
    <node1> Geronimo server shutting down...
    <node1>....
    <node2> Geronimo server shutting down...
    <node2>....
    <node1>Geronimo server shutdown.
    <node2>Geronimo server shutdown.
    Starting up 2 nodes...
    <node1>Geronimo server starting...
    <node1>..
    <node2>Geronimo server starting...
    <node2>..
    <node1>Geronimo server started in ...
    <node2>Geronimo server started in ...
    Started up 2 nodes.

And well, if you had some kinda script file which controlled say a logical grouping of nodes you could easily invoke that script (ya even on a remote system) and it will go and do it:

someserver:9447 > script -l groovy local:file://restart- universe.groovy qa-universe

The local: bit of the uri siginals the local URL handler to be used, which will cause the file://restart-universe.groovy to be loaded from the gsh instance where you are actually logged into (and ran the remote-shell gshell command) and will pipe its contents securely to the remote shell running on someserver:9447 and pass it to the script command to execute.

The restart-universe.groovy might look something like this:

<snip>
import universe.Lookup

assert args.size == 1 : 'Missing universe name'

def universe = args[0]

// Look up a list of nodes (for now say they are basically hostname:port)
def nodes = Lookup.lookup(universe)

log.info("Stopping universe ${universe}...")
nodes.each { host ->
        shell.execute("remove-shell $host stop-server")               
}
log.info("Universe ${universe} stopped")

log.info("Starting universe ${universe}...")
nodes.each { host ->
        shell.execute("remove-shell $host start-server")              
}
log.info("Universe ${universe} started")
</snip>

Its kinda crude script, but I think you get the general point...

 * * *

Anyways... I see... well, *HUGE* potential for this stuff...

And really, a lot of what I just described above isn't that far into fantasy, its all relatively easy to implement on top of GShell... as it is now (or really as it was a year+ ago when I wrote it). Its really a matter of do others see the same value... and do others see the vision of using GShell as the core process launcher to allow things like "restart-server", or a "stop-server; copy-configuration default known-good; copy-configuration default testing; start-server", or that uber-fancy remote-shell muck.

So, I'm gonna give y'all a few days to grok (or try to) what I've just spit out... please ask questions or comment, as I like to know I'm not just talking to myself here.

And then maybe later next week, we might vote or come to some other consensus that this is the right direction for Geronimo, and well... then I'll make it become reality.

Aighty, and now I'll shut up :-P

--jason



On Sep 8, 2007, at 11:53 AM, Jason Dillon wrote:

Aighty, well... I've done some long awaited re-factoring, and while its still not _perfect_ its a whole lot better now IMO I think from a framework perspective that its probably mature enough to take on the task of being the server bootloader.

I'm going to continue to refactor the guts of GShell over time, of course... but I think that what is there now is highly usable for a simple platform independent launcher, as well as for porting over the other cli bits we have.

I've done a lot of work in the past week, incase you didn't see the storm of scm messages... pulled out pico, plopped in plexus, pulled out commons-logging and commons-lang, which are suck and boated (in that order). I've gotten the basic framework and supported classes to use GShell down to ~ 1mb (a wee bit under)... though when I started to add the layout.xml abstraction stuff, I had to pull in xstream which bloated her back up to ~1.4m. I may eventually fix that... or not, cause xstream is soooo very handy for xml -> object stuff.

I've fallen in love with annotations... they are *ucking great. They work really well for handling the cli option and argument muck which most every command needs to do. And striping out the insano-sucking commons-cli really simplified command implementations dramatically IMO.

Anyways... I've make a heck of a lot of progress on cleaning up the GShell framework... and more is to come I'm sure... But for now, I think its probably ready for use primetime as the Geronimo Server's bootloader.

I think this provides a some significant value...

1) Platform scripts become consistent and relatively simple, easy to maintain

2) Everyone will now have a consist way of launching the server, even if you like a .sh, .bat, or java -jar, then end process that is launched will be the same for everyone.

3) Opens up the door for some really nice and fancy fancy management muck (like restarting the server from the web console, or cloning a server instance or backing up a server instance...)

4) Lays the ground work for future features, like cluster management, remote administration and scripting...

 * * *

So, I think its time to decide... are we a go or no go for GShell as the core CLI for Geronimo thingys and even more important, are we go or no go for using GShell to boot up the server process?

--jason




Reply via email to