Adjust Cernan's internal timescale to be 1ms.

[blt]

Presently cernan operates on a 1 second timescale. The native protocol defines time in ms timescales. Cernan's internal model ought to be adjusted to correspond to its native protocol.

The discrepancy is one of gradual change to the system, clarification of goals and conventions.

This will necessarily require a change to the way we interpret bin_width. We'll need to retain the default of 1 second bin_width.

[blt]

I think, absent a hard ask for this, I'm not enthusiastic about the change any longer. 1ms is just as arbitrary as 1s. Closing.

[joedevivo]

It'd be useful to keep receipt-accurate timestamps for loglines

Currently, we're dropping millisecond precision on these events. The text in the log payload still shows the milliseconds, but nobody is searching or sorting based on that information.

[blt]

@joedevivo to confirm, you're pumping logs through Native source into ES sink?

[joedevivo]

Actually, you can see this with the Console sink just as easily. It all comes down to the Native source scrubbing the millisecond data it receives here: https://github.com/postmates/cernan/blob/master/src/source/native.rs#L165