Memtools

This has a couple of utilities/examples of finding memory leaks in python.

Most important is MemGraph, which uses graphviz to show which classes refer to the most memory.

Memgraph

MemGraph requires graphviz and pympler.

Copy memgraph.py to your project, then:

    from pympler import muppy
    from memgraph import MemGraph

    g = MemGraph(muppy.get_objects())
    g.render()  # maybe with view=True?

By default, only classes that are responsible for >5% of memory usage are represented. This applies to both nodes and edges.

Files are output to ./name-%Y%m%d-%H%M%S.gv.

Output

Here's a simple, unbalanced case:

Each node and edge show the number of instances (objs), the total referred to memory (including self), and memory just used by class instances (self).

All objs is a special case and isn't linked. For it, objs and self represent all objects passed in to MemGraph. ref includes memory used in generating the graph. ref is not used anywhere - min_ref_percent is only based on objects passed in to MemGraph.

---

Here is more complex output from test_classify:

This output is actually identical to the first example, but with ReferencesMem using less memory.