Repository navigation
When %matplotlib inline is ON gc.collect() runs on every cell execution #5795
Description
Hi,
I'm not sure if it is the best place to put this issue. I'm working on an IPython notebook with a lot of objects loaded: len(gc.get_objects()) is about 5M. Everything works fine until an inline plot is drawn. Drawing itself is fast, but then there is a delay (5-10s). The delay is caused by gc.collect() executed by matplotlib: https://github.com/matplotlib/matplotlib/blob/master/lib/matplotlib/_pylab_helpers.py#L93. After the first plot the notebook becomes slow for all cells (even without inline plots). When interrupting the kernel (for cells without plots) I can see the same gc.collect() happening inside matplotlib.
As a workaround I'm monkey-patching gc.collect to be no-op, but this is clearly a hack.
It is gc.collect() call in matplotlib code that makes things slow, but why does IPython notebook need to call matplotlib code for cells without charts?
Activity
minrk commented on May 6, 2014
IPython makes matplotlib calls at the end of each cell to see if there are plots to draw. Can you tell what problematic call is being made (full stack up to an IPython call would be ideal) in a cell with no plots?
kmike commented on May 6, 2014
Here is a stacktrace for other cells (stacktrace for plots is the same):
Traceback (most recent call last): File "/Users/kmike/envs/scraping/lib/python2.7/site-packages/IPython/kernel/zmq/ipkernel.py", line 389, in execute_request shell.run_cell(code, store_history=store_history, silent=silent) File "/Users/kmike/envs/scraping/lib/python2.7/site-packages/IPython/core/interactiveshell.py", line 2743, in run_cell self.events.trigger('post_execute') File "/Users/kmike/envs/scraping/lib/python2.7/site-packages/IPython/core/events.py", line 82, in trigger func(*args, **kwargs) File "/Users/kmike/envs/scraping/lib/python2.7/site-packages/IPython/kernel/zmq/pylab/backend_inline.py", line 118, in flush_figures return show(True) File "/Users/kmike/envs/scraping/lib/python2.7/site-packages/IPython/kernel/zmq/pylab/backend_inline.py", line 47, in show matplotlib.pyplot.close('all') File "/Users/kmike/envs/scraping/src/matplotlib/lib/matplotlib/pyplot.py", line 517, in close _pylab_helpers.Gcf.destroy_all() File "/Users/kmike/envs/scraping/src/matplotlib/lib/matplotlib/_pylab_helpers.py", line 93, in destroy_all gc.collect() KeyboardInterruptminrk commented on May 6, 2014
So it looks like the gc call wouldn't happen if we check whether there are any figures to close before calling close('all') (same goes for matplotlib's own destroy_all call).
kmike commented on May 6, 2014
Yes, it should fix it for non-plotting cells. I've fired a related issue to the matplotlib bug tracker.
kmike commented on May 6, 2014
btw, it's just amazing how fast are you responding to tickets here 👍