Interpreting objgraph numbers: memory leak?

I have a Python client-server application and its server has been running for about 4 days... Its memory usage (as reported by /proc/PID/status -> VmSize) has grown to about 660Mb, and I was wondering if there could be a memory leak, so I used objgraph to print out its memory usage (first 20 object types):

58524    tuple
33270    dict
15483    function
9976     list
5396     set
2644     weakref
2489     builtin_function_or_method
2482     instancemethod
1898     OrderedSet
1751     _BindParamClause
1680     _generated_label
1485     Comparator
1398     type
1315     InstrumentedAttribute
1267     Custo开发者_C百科mColumn
1165     cell
1146     ScalarAttributeImpl
1146     ColumnProperty
1146     ColumnLoader
1075     wrapper_descriptor

I don't see anything particularly "smelly" there, am I missing something? Maybe this is just because Python doesn't release unused memory to the OS immediately?

(This is a CherryPy+SQLAlchemy application)

To acquire more meaningful data, keep the server running for several days, and pull this summary once per day. The key will be to see which kinds of objects are accumulating from one day to the next; those object types will be the most likely candidates for your memory leak! Let us know what you find.