- For HA, yes you are right.
- But for some stuff, like EC2 provisioning, we need more than editing puppet confs.
In any case: The main purpose here is to make it so that we can easily build bigtop clusters which are aimed at minimally representing a particaular feature or bug. For example, on monday, maybe i want to test HA. On tuesday, maybe im in HBase mode. .... So lets make these vagrant recipes more reusable, and easier to extend.
Right now we just have one monolithic vagrant provisioner.
ill try to sketch a more concrete idea of what im thinking, but for now, something like this:
So basicallly, to play with a particular hadoop deployment, you just create a new subdirectory, and make some minor customizations and evans and I will try to maintain the base-provisioner script.
In the end some of these sub directories might be great to contribute back to bigtop, some might just be private.