Cubed-specific diagnostics to record

[TomNicholas]

Some ideas:

• Projected/measured memory (benchmark memory usage #8)
• Number of tasks run (Benchmark task data #9)
• Intermediate data stored (Benchmark amount of intermediate data stored #10)
• Container seconds (Benchmark task data #9)

All could be added later though. I think duration is what we're interested in most.

Originally posted by @tomwhite in #1 (comment)

[TomNicholas]

Recording the duration is done via a context manager - perhaps the others will need to be handled by passing callbacks to cubed instead.

[TomNicholas]

Can we just use the HistoryCallback every time, then read the results from the csv file that callback writes to?

[tomwhite]

Can we just use the HistoryCallback every time, then read the results from the csv file that callback writes to?

Or get the values directly from the events list stored on the HistoryCallback object.

[TomNicholas]

Or get the values directly from the events list stored on the HistoryCallback object.

In that case what's the best way to get access to the history = HistoryCallback() object from within the wrapping benchmark context manager?

[TomNicholas]

Oh maybe the benchmark context manager should accept an argument (like how in coiled-benchmarks the benchmark_all context manager accepts a client argument).

[tomwhite]

It would also be useful to record number of backup tasks run.