glenn opdycke-hansen wrote:
Steve, I have been looking at using Ant for application deployment to multiple servers. I see there is a known problem with ant 1.6.5 where a sshexec request will close stdout. This can be patched, but this problem is pretty serious since multiple sshexec requests would be needed.
Do you know if it has been fixed in ant1.7? If not, now is the time to get a patch in.
I have been looking at the documentation for SmartFrog. Can you point out a doc that describes the issues of deployment?
I would have a look at the overview presentation we gave at OSCon: http://www.hpl.hp.com/research/smartfrog/presentations/Taming_deployment_with_SmartFrog.pdf
This would include distribution of build files, execution of deployment scripts, configuration, and logging/notification?
Probably the best paper on how to use ant for large scale deployment is the "Advanced deployment" chapter in Java Development with Ant, in which we covered how to use property files to control uploading of build files to different targets, then remotely executing them. I probably used ftp+telnet though, not scp/ssh.
Smartfrog is not a scripting tool; you are really trying to describe the static system state; which things should be running in a correctly functioning system. It just so happens that to deploy some thing you need to run scripts or do other actions in a sequence, so there are some workflow components you deploy to run stuff in a sequence, handle failures etc. A lot of the examples (and real stuff) has tended to drop down to unix shell scripts to do stuff like install apache, mysql, etc. One of the things I did after joining the project was try and add more platform independence, with constructs like Ant's file type, and a better java runner, for java execution (that is a bit unfair, as smartfrog has always been really good at deploying stuff in new processes, but I needed separate process execution to escape from the security restrictions of signed-code-only).
Logging and notification. hmmm. Logging is done, though logging in a distributed system is part of the ongoing problem of "how do you monitor distributed systems properly". It helps to have all the clocks in sync, otherwise it becomes hard to determine the 'absolute' order of things happening. Realistically, there is no absolute order of events in a distributed system, we are all radio astronomers looking at distant events in the past, trying to work out why a server disapperaed for no apparent reason.
-steve --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
