Write Counter output even when TypedPipe has no elements (tallyAll)

[connord-stripe]

I noticed that using the new tally API for counters won't create a new diagnostic counter (if I understand the naming correctly) if the associated TypedPipe has 0 elements. I'd prefer if this made the counter and just put the result as 0.

The following repro works for me, assuming you have an input that contains data. I'm looking in our jobtracker (timberlake) to identify info about the counters. If you replace the false below with true, the counter name will indeed appear in our jobtracker.

def execution: Execution[unit] =
...
      val output = TypedPipe.from(WritableSequenceFile[BytesWritable, BytesWritable](input))
        .filter( _ => false )
        .tallyAll("Counts", "testtesttest")
        .writeExecution(WritableSequenceFile[BytesWritable, BytesWritable](nestedOutput))

      Execution
        .sequence(List(output)) // could likely be simplified
        .unit
    }

---

Also a small nit: The current docs (https://twitter.github.io/scalding/api/#com.twitter.scalding.typed.TypedPipe) don't seem to have any reference to the tally functionality, seems like it might need to be regenerated?

[johnynek]

here is where counters are incremented:

https://github.com/twitter/scalding/blob/develop/scalding-core/src/main/scala/com/twitter/scalding/Operations.scala#L694

In the (currently non-existent) cleanup method, if we know all the possible counters that could be updated, we could increment them with 0. Maybe that would work.

The bigger issue is that currently, we allow totally dynamic counters, even though many users make them statically. I think we can pattern match to discover the static cases since we use a case class, I think, to make the TupleConverter.