Logging

As they run, the worker threads each log events to their own circular buffer. When KillThreads is called each thread saves their log to a file, these are named

• worker_thread_log_0.txt
• worker_thread_log-1.txt
• worker_thread_log-2.txt
• etc...

You can use the LogVisualizer.py to produce a plot of the scheduling:

The above diagram shows one top level task kicking off 3 child tasks which each kick of three child tasks of their own, which do ~2ms of work.

LogVisualizer.py accepts the following arguments:

• --txt_glob - glob expression to find log.txt files, defaults to "../worker_thread_*.txt"
• --dump_json - convert the logs into a chronological list of events by fiber and dump as json to stdout
• --help - show help
• --out - output png path - optional, if not passed, the plot is an interactive window
• --width - width in inches, only relevant with --out, is optional with a default
• --height - height in inches, only relevant with --out, is optional with a default
• --dpi - dots per inch, only relevant with --out, is optional with a default