in compileapp there is some weird logic: if not os.path.split(model)[0] in paths and c!='appadmin': continue elif compiled: code = read_pyc(model) elif is_gae: code = getcfs(model, model, lambda: compile2(read_file(model), model)) else: code = getcfs(model, model, None)
It was designed to cache compiled models. It was tested long ago when GAE was behaving differently. Could this be the problem? What if you remove the elif is_gae clause? On Feb 19, 5:20 pm, howesc <how...@umich.edu> wrote: > i've learned that i don't know how to figure this out. i tried using the > sys.getsizeof() on globals in various locations in the code and that always > remained constant even while the memory size of the instance changed. > > i did notice that by adding the toolbar to the response on > welcome/default/index that it adds .5-1MB to the instance memory usage per > page refresh. is there some way that views or view environments can be > polluting the global namespace? > > i also noticed a call in gluon/main.py to gc.collect() every 100 > requests.....so is there something we know we are leaving around? i tried > playing with calling it more than once every 100 but got unpredictable > results. > > any ideas on how i can better track this? i don't know much about figuring > out python memory usage in general, let alone in the GAE environment - so > i'm happy to try any tools/tests anyone can recommend. > > thanks! > > christian