Hello, dear Mediapipe guys.

I want to inference the hand pose with Mediapipe model and my own model.

I have my own tf-lite models, it can work on the RGB bitmap.

I try to query the RGB bitmap from input frame with data packet.

My code is

private static final String INPUT_VIDEO_STREAM_NAME = "input_video";
processor.addPacketCallback(INPUT_VIDEO_STREAM_NAME, (packet)->{
 

Describe the bug

It's possible to create a stream or table with no value columns using a CREATE TABLE or CREATE STREAM statement. No error is returned.

However, if you try and use the stream or table in a downstream CREATE TABLE AS or CREATE STREAM AS statement you get a error message.

i.e. the table/stream created with no columns isn't useful.

To Reproduce

Currently, you are not able to use middleware per handler in "clean way" (you can still manually wrap the handler function, or have separated router).

It would be nice to be able to add middleware per handler.

As a good reference point, you can use chi router: https://github.com/go-chi/chi

It can be very difficult to piece together a reasonably estimate of a history of events from the current workers logs because none of them have timestamps.

So for that end, I think we should add timestamps to the logs.

This has some cons:

1. We can't just use @printf like we have been until now. We need to either include a timestamp in every @printf call (laborious and error prone) or c

Jobs in 4.3 log start time in ISO8601 format and duration

	Start time: 2020-10-04T20:55:01.701
	Duration: 73,142,855 ms

It would be useful to log the end time in ISO8601 and/or the duration in days/hours/minutes/seconds, as it's not immediately obvious how long is 73 million milliseconds.

All exceptions in namespace JsonMachine\Exception should extend one common exception, say JsonMachineException, so that userland code can catch for only one type and catch anything from this library. Fell free to create pull request and participate :)