TODOs and Maybes

TODO NOW

generalize trending code, don't depend on a running graphite.
finish integration between output handling and new expectation system
default to a less noisy log format, ruby format was a neat idea but I don't like it.
Steal ShellOut from Chef for running external commands
message sending needs to IO.select(nil,[writers],nil, timeout) so it doesn't deadlock if all the workers are dead/busy
retries after respawning workers in the above scenario
optimize scheduler
run once mode:
- return code non-zero for failure
- show the results in a pretty format

allow metrics to be defined with less DSLism (i.e., class MyMetric < Metric::Base should do the necessary bookkeeping)
Add primitives for common measurement types
- counters (like for CPU time/percentage) should be easy
- timers should be easy
allow plugins and config declare gem deps, automate installing them. (probably bundler)
Option to print the config in summary form for troubleshooting complex configs
CLI client to submit a task to the queue manually
figure out stable protocol so other programs/libs/langs can submit jobs to the queue
State Machine
- does this go in the client at all?
- if implemented in client, probably need to have monitors return result, main loop will need to track the open sockets...
debug method within monitors, take a block and pp the output when verbosity is on for collection
Finish "story mode"
manpages (ronn)

integrate ohai so you can have auto-monitors (e.g., loop over all available disks)
instantiate metrics from serializable data
live code reloading
use kgio instead of the built-ins.
privilege separation, run some workers as root, others not.

Support story mode by builing primitives for full stack testing where sensible.