Refactoring discussion: Configuration about macrobase HOT 7 CLOSED

See Spark, which uses untyped Key-Value pairs: https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/SparkConf.scala#L30

from macrobase.

IMO KV pairs seem like they would scale much better to more config flags, which is inevitable as the codebase progresses. Not sure how exactly the untyped stuff would work in Java though. And we would need to keep good, up-to-date documentation on all the config options in one file.

from macrobase.

Spark has a good example of typing:

Double getDouble(String param), String getString(String param)

It gets a little annoying to have a bunch of these, like List<String> getStringList(String param)

You can imagine also having:

Double getDoubleOrElse(String param, Double default)

And:

Boolean mandatoryParams(List<String> params).

from macrobase.

I started writing a new config class. Two things:

1. Do we want to rename all MacroBase files to Macrobase?
2. Spark's configuration API forces you to specify a default for each setting. The upside of this is it makes the behavior explicit. The downside is that this makes it awkward to enforce mandatory fields (e.g., targetLowMetrics). Which do we prefer?

751ae21#diff-a68be5bcbbbf7f3d53f49b90310d3e59R11

from macrobase.

I like the idea of defaults for the most part, since most of our parameters do have default values. For those that don't, we can have a default value like null, and then throw an exception if any argument is null.

from macrobase.

I like defaults too. the only thing that we require are the sql query, metrics and attributes. and those seem hard to forget :)
Also since we are moving to automating everything, we should in the end have defaults for everything, no?

from macrobase.

See #77.

from macrobase.