Bo

Our model currently is 1. Keep DAG and light computation (steps) locally, even with multiprocessing. 2. Send specialized tasks to remote hosts as processes. 3. Send heavy computations to task...

https://unix.stackexchange.com/questions/103699/kill-process-spawned-by-ssh-when-ssh-dies

In stead of resorting to killing workers, we should implement a heartbeat mechanism. That is to say, the worker should kill itself if it cannot detect master for a while,...

So basically I am suggesting that 1. If `localhost` is defined, the `paths` will be made available as runtime variables to be used in scripts. 2. When the task is...

I am wondering if we should make the paths more explicit. That is to say, instead of making ``` {'home': '/Users/user'} ``` part of the dictionary and use ``` input:...

With #1219, `CONFIG` is now pickled to remote host so that `CONFIG` is the same across local and remote hosts. The idea of this ticket is that we can make...

So I seem to incline to ``` HOST['home'] ``` since users can write ``` HOST.get('scratch', '/some/other/way') ``` if they are not sure if `scratch` is defined on certain remote host.

The only problem is that ``` CONFIG['host']['home'] ``` is 10 character longer than ``` HOST['home'] ```

Good suggestion but will not work because things defined in `global` will be `frozen` (execute only once). Right now, I * load config files * execute global section * save...

We can change `CONFIG` but there is no guarantee that user derived variables will be changed accordingly. I mean, if we define `CONFIG['host']` as a dynamic dictionary for local and...