Memory overloading on time series
Hi, The memory usage in yt is filling up the machine for a time series with ~100 outputs. It seems like the data from previous datasets isn't being cleared from memory, despite calling "del pf" and "gc.collect()". I've tried various suggestions made previously in the mailing list to no effect. I'm using the Ramses frontend with the script attached. Any ideas? Thanks, Sam
Hi Sam, I've looked over your script and I'm not entirely sure where the memory usage is not being freed. My one thought is to try this: import yt yt.frontends.ramses.RAMSESDataset._skip_cache=True at the top of your script and seeing if it helps. -Matt On Mon, May 19, 2014 at 9:10 AM, Sam Geen <samgeen@astro.ox.ac.uk> wrote:
Hi,
The memory usage in yt is filling up the machine for a time series with ~100 outputs. It seems like the data from previous datasets isn't being cleared from memory, despite calling "del pf" and "gc.collect()". I've tried various suggestions made previously in the mailing list to no effect. I'm using the Ramses frontend with the script attached. Any ideas?
Thanks,
Sam
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
Hmm, nope, memory usage still slowly creeping up (I also tried reloading yt.mods). I have a couple more ideas to try - one is to run the function in a separate process, the other is to check for existing images and skip that output if they exist, then just re-run every time the memory fills up. Is there any way of profiling what's in the memory with Python? I hope/assume it's not numpy or something leaking. On 20/05/14 13:48, Matthew Turk wrote:
Hi Sam,
I've looked over your script and I'm not entirely sure where the memory usage is not being freed. My one thought is to try this:
import yt yt.frontends.ramses.RAMSESDataset._skip_cache=True
at the top of your script and seeing if it helps.
-Matt
On Mon, May 19, 2014 at 9:10 AM, Sam Geen <samgeen@astro.ox.ac.uk> wrote:
Hi,
The memory usage in yt is filling up the machine for a time series with ~100 outputs. It seems like the data from previous datasets isn't being cleared from memory, despite calling "del pf" and "gc.collect()". I've tried various suggestions made previously in the mailing list to no effect. I'm using the Ramses frontend with the script attached. Any ideas?
Thanks,
Sam
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
Hi Sam, In yt, we have a memory profiler you can use that outputs at regular intervals. with yt.memory_checker(some_interval): do_something_costly() It's primitive, but will show time-based memory increases. On Tue, May 20, 2014 at 8:28 AM, Sam Geen <samgeen@astro.ox.ac.uk> wrote:
Hmm, nope, memory usage still slowly creeping up (I also tried reloading yt.mods). I have a couple more ideas to try - one is to run the function in a separate process, the other is to check for existing images and skip that output if they exist, then just re-run every time the memory fills up. Is there any way of profiling what's in the memory with Python? I hope/assume it's not numpy or something leaking.
On 20/05/14 13:48, Matthew Turk wrote:
Hi Sam,
I've looked over your script and I'm not entirely sure where the memory usage is not being freed. My one thought is to try this:
import yt yt.frontends.ramses.RAMSESDataset._skip_cache=True
at the top of your script and seeing if it helps.
-Matt
On Mon, May 19, 2014 at 9:10 AM, Sam Geen <samgeen@astro.ox.ac.uk> wrote:
Hi,
The memory usage in yt is filling up the machine for a time series with ~100 outputs. It seems like the data from previous datasets isn't being cleared from memory, despite calling "del pf" and "gc.collect()". I've tried various suggestions made previously in the mailing list to no effect. I'm using the Ramses frontend with the script attached. Any ideas?
Thanks,
Sam
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
Hi Sam, You might also try gleaning information about why it's not getting collected with gc.set_debug(...). For instance, 'gc.set_debug(gc.DEBUG_LEAK)'. -tom On 05/20/2014 02:28 PM, Sam Geen wrote:
Hmm, nope, memory usage still slowly creeping up (I also tried reloading yt.mods). I have a couple more ideas to try - one is to run the function in a separate process, the other is to check for existing images and skip that output if they exist, then just re-run every time the memory fills up. Is there any way of profiling what's in the memory with Python? I hope/assume it's not numpy or something leaking.
On 20/05/14 13:48, Matthew Turk wrote:
Hi Sam,
I've looked over your script and I'm not entirely sure where the memory usage is not being freed. My one thought is to try this:
import yt yt.frontends.ramses.RAMSESDataset._skip_cache=True
at the top of your script and seeing if it helps.
-Matt
On Mon, May 19, 2014 at 9:10 AM, Sam Geen <samgeen@astro.ox.ac.uk> wrote:
Hi,
The memory usage in yt is filling up the machine for a time series with ~100 outputs. It seems like the data from previous datasets isn't being cleared from memory, despite calling "del pf" and "gc.collect()". I've tried various suggestions made previously in the mailing list to no effect. I'm using the Ramses frontend with the script attached. Any ideas?
Thanks,
Sam
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
Thanks guys, will try those suggestions! On 20/05/14 14:40, tom fogal wrote:
Hi Sam,
You might also try gleaning information about why it's not getting collected with gc.set_debug(...). For instance, 'gc.set_debug(gc.DEBUG_LEAK)'.
-tom
On 05/20/2014 02:28 PM, Sam Geen wrote:
Hmm, nope, memory usage still slowly creeping up (I also tried reloading yt.mods). I have a couple more ideas to try - one is to run the function in a separate process, the other is to check for existing images and skip that output if they exist, then just re-run every time the memory fills up. Is there any way of profiling what's in the memory with Python? I hope/assume it's not numpy or something leaking.
On 20/05/14 13:48, Matthew Turk wrote:
Hi Sam,
I've looked over your script and I'm not entirely sure where the memory usage is not being freed. My one thought is to try this:
import yt yt.frontends.ramses.RAMSESDataset._skip_cache=True
at the top of your script and seeing if it helps.
-Matt
On Mon, May 19, 2014 at 9:10 AM, Sam Geen <samgeen@astro.ox.ac.uk> wrote:
Hi,
The memory usage in yt is filling up the machine for a time series with ~100 outputs. It seems like the data from previous datasets isn't being cleared from memory, despite calling "del pf" and "gc.collect()". I've tried various suggestions made previously in the mailing list to no effect. I'm using the Ramses frontend with the script attached. Any ideas?
Thanks,
Sam
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
OK, so in the end I cheated a little and used the multiprocessing.Process object to spawn separate instances of python for each function, and isolated yt to those processes; I've attached the code if anyone's interested. I had a quick look at the gc debug output but it's quite hard to figure out what I'm looking at at first glance. On 20/05/14 14:42, Sam Geen wrote:
Thanks guys, will try those suggestions!
On 20/05/14 14:40, tom fogal wrote:
Hi Sam,
You might also try gleaning information about why it's not getting collected with gc.set_debug(...). For instance, 'gc.set_debug(gc.DEBUG_LEAK)'.
-tom
On 05/20/2014 02:28 PM, Sam Geen wrote:
Hmm, nope, memory usage still slowly creeping up (I also tried reloading yt.mods). I have a couple more ideas to try - one is to run the function in a separate process, the other is to check for existing images and skip that output if they exist, then just re-run every time the memory fills up. Is there any way of profiling what's in the memory with Python? I hope/assume it's not numpy or something leaking.
On 20/05/14 13:48, Matthew Turk wrote:
Hi Sam,
I've looked over your script and I'm not entirely sure where the memory usage is not being freed. My one thought is to try this:
import yt yt.frontends.ramses.RAMSESDataset._skip_cache=True
at the top of your script and seeing if it helps.
-Matt
On Mon, May 19, 2014 at 9:10 AM, Sam Geen <samgeen@astro.ox.ac.uk> wrote:
Hi,
The memory usage in yt is filling up the machine for a time series with ~100 outputs. It seems like the data from previous datasets isn't being cleared from memory, despite calling "del pf" and "gc.collect()". I've tried various suggestions made previously in the mailing list to no effect. I'm using the Ramses frontend with the script attached. Any ideas?
Thanks,
Sam
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
participants (3)
-
Matthew Turk -
Sam Geen -
tom fogal