it seems multiply calls to
cerebro.plot(savefile=True) method result in subtile memory leak.
Here comparative test charts: take a look at
BTgymServer process, running backtarder loop, violet line.
With plotting feature
on here is evidence of memory usage increase (40 runs with plot() after every one).
Second run has no plot calls, memory usage is flat line.
I guess it can be connected with the known matplotlib memory issues and the fact internal
.savefig() methods do not explicitly close generated matplotlib figures. [or use multiple pyplot import statement]