while using using IPython to process some scientific data with numpy, scipy, matplotlib etc... I sometime experience a lot of latency when I try to move the cursor backwards to change a parameter in a previously written line. Actually the latency is due to considerable amount of disk write access (~1 MB/s) triggered when moving the cursor on a already written text.
It is difficult to say under what conditions it does happen. At first every thing is fine, I run my processing routine and start to plot results. And then at some point, maybe after 5-6 plots , it happens. At that point I still have 1G of free RAM and the swap is not used.
So iotop gives a 1 MB/s write disk access of ipython but I don't know how to find what file it's writing to. I tried to use lsof for that purpose, and the only difference it gives me between the moment I'm moving the cursor and the following is:
ipython 5947 rth 17u REG 8,3 37430 2388918 /home/rth/.local/share/recently-used.xbel.ETR1TV
But It seems to be just some kde bookmarks of recently used files, so I'm not sure if it is connected. And the file is 37k only...
I would be grateful for an advice as to what I should look for to provide more complete debug information,
As far as I know, the only thing that can happen when you move the cursor is readline stuff, so it might not be something we're doing.
We're aiming to release 0.11 soon, so if you've got time, could you try it with the current development head?
It seems that after I changed a little my processing script and if I import it instead of using %edit each time (I'm not saying that it's correlated), I can't reproduce the issue at all (with 0.10.2 so I haven't tried 0.11) . Anyway as you said , it's probably nor related to IPython. Thanks and sorry to take your time.