How do show_growth deltas behave when memory runs out?

[rschwiebert]

I'm profiling to find a leak in an API endpoint that runs a complex process. I have objgraph.show_growth report at two or three points inside the endpoint's main method.

Hitting the endpoint over and over, I quickly hit a MemoryError, and for a while large positive deltas are reported, but finally the deltas stop being reported at all.

I guess conceivably this could mean one of two things:

1. There is no positive delta to report: memory is saturated with objects involved and no new ones are created.
2. Objects are being created, but show_growth can't function properly in the given memory, so it doesn't report them.

I don't really understand this library well enough to see which one is more likley, but can you comment as to which one you think it is?

Thanks.

[mgedmin]

Ooof. I've never dealt with a MemoryError, so I cannot say for sure. I would speculate that a MemoryError might mean some internal data structures get updated incompletely, which might cause the next show_growth() to report more new objects than there are actual new objects (assuming no more MemoryErrors occur).

You can always double-check by looking into how many objects are tracked by the Python garbage collector, by looking at len(gc.get_objects()) periodically.