python - memory statistic diff in google heap profiler and real memory use

Question

My python stablely use 17GB memory.

10705 hadoop-+  20   0 33.453g 0.016t  93024 S 405.1  6.5   4560:23 python main.py

but google heap profiler tell:

Total: 1354.5 MB
   282.9  20.9%  20.9%    282.9  20.9% std::string::_M_mutate
   180.9  13.4%  34.2%    406.0  30.0% tensorflow::OpKernel::OpKernel
   ....

I'm sure that my top is correctly, so I think I have a error usage with google heap profiler

my profiler code :

export LD_PRELOAD="`pwd`/tf_venv/libcudnn/libtcmalloc.so"
HEAPPROFILE=/tmp/test.prof python main.py $*

my tcmalloc version:4.1.0

question from:https://stackoverflow.com/questions/65867948/memory-statistic-diff-in-google-heap-profiler-and-real-memory-use

Answer 1

Waitting for answers