[corytheboyd]

I hadn't even considered collecting traces/spans in this way yet, and have taken the approach of "stuff outputting logs in JSON format to stderr/local file". I usually end up writing a (temporary, structured) log message with the relevant span tags, but wouldn't it be much better to run the actual trace/span code and be able to verify it locally without the ad-hoc log message?

The prototype I built is a web application that creates websocket connections, and if those connections receive messages that are JSON, log lines are added. Columns are built dynamically as log messages arrive, and then you can pick which columns to render in the table. If you're curious here's the code, including a screenshot: https://github.com/corytheboyd-smartsheet/json-log-explorer

With websockets, it's very easy to use websocketd (http://websocketd.com), which will watch input files for new lines, and write them verbatim as websocket messages to listeners (the web app).

To make the idea real, would want to figure out how to not require the user to run websocketd out of band, but watching good ol' files is dead simple, and very easy to add to most code (add a new log sink, use existing log file, etc.)