New execution history logging storage mechanism

[franzlst]

A python shelf is not robust against, see #461.

For not robust systems (or systems with a battery, which can just power off) we need a new storage format, which can handle breakdowns.

A possible solution would be to append json strings (or something similar) to an already exisiting file.

Discussion open.

@SebastianRiedela @doem-an @franzlst @Rbelder

Originally created by @sebastian-brunner ([email protected]) at 2018-03-05 15:22:37+00:00 (moved from RMC internal repository)

[franzlst]

I don't know how the execution logger currently registers a (proper) closing of the GUI.

Maybe it could help to make use of the signal_handler (rafcon/gui/start.py), which is called for various signals (SIGINT, SIGHUP, SIGQUIT, SIGTERM -- not SIGKILL, I think this is not possible).

On the other hand, this signal_handler already stops the execution, so I guess it also gracefully ends the logging?

Originally created by @franzlst ([email protected]) at 2018-03-06 08:25:48+00:00 (moved from RMC internal repository)

[franzlst]

If the shelve cannot handle crashes and looking at the implementation of e.g. https://github.com/dagnelies/pysos, I'd propose a we code up a simple file and line-based log for rafcon.

What is stored in the shelve at the moment is a dictionary of history_item_id (string) -> history_item_info (dict, with possibly pickled data variables). We could store that line-based in a csv-fashion in the format:

history_item_id,pickle_string_of(history_item_info)\n
...

Originally created by SebastianRiedela ([email protected]) at 2018-03-06 09:05:30+00:00 (moved from RMC internal repository)