[dsr_]

   You don't have to know about yum, apt or whatever package manager your hosts use thanks to synapse-agent's abstraction layer.

Except, of course, that a sysadmin who doesn't know this is not a sysadmin. And someone who tries installing the same package across different operating systems is going to be in for a shock when it turns out that they have the same name for different versions, or different names for the same version, or completely different ways of configuring it.

This looks neat, but I don't see anything that says they have learned lessons from puppet, bcfg2 and cfengine.

[eschnou]

Indeed, the package example is the easiest one to explain but the least useful due to naming issues accross distro. It works however the same for services, users, etc.

As you mention, the approach is really similar to the abstraction in the Puppet DSL. But instead of using it to describe a recipe, you use it to remotely manage a host.

Obviously you need knowledge of the target system, but at least you can write orchestration scripts that are cross-platform and where the only data changing are package names, path, and content of files.

[dsr_]

Ok, so what did you folks learn from the other projects in this space? Why is this better than Puppet, bcfg2, or cssh?

[eschnou]

Synapse is not a puppet/chef like configuration management tool, it is used for live management & orchestration.

It's in fact much more similar to mcollective and salt, which both innovates from tools like cssh by using a messaging middleware instead of ssh to connect to the target hosts.

[ominous_prime]

Ansible[1] is starting to look really nice. It's kind of a mix between puppet and fabric, handling configuration and orchestration in a single tool. If it didn't mean rewriting all my puppet configs, I'd probably be using it by now (though I've come close a couple times).

1. http://ansible.github.com/