2013-01-15-53170-win_x86-devprev memoryleak.

Started by Bletchley Park

Bletchley Park Lv 1

I notice that with the current client in puzzle 670 the memory usage is steadily growing.
With only mutates and wiggles the size has grown from 540 Mb to 730 Mb at the time of writing in a matter of a few hours. I cannot explain this with temporary storage needs. Last time the client crashed at 795 Mb.

I am anticipating it will crash again at that point this time. We have seen this particular behaviour before. A few months ago I reported on a memory leak which I think has not been fixed yet.

auntdeen Lv 1

In Main on Mac, it's not just temporary storage.

I opened 670 again tonight to upload an unusual configuration for my team to look at. I did a WA and a mutate from a previously saved pose - that's it, no script, simple, then shared.

The program size in the last client was 351mb. The new client size yesterday was 408mb. When I looked at the size after sharing the protein in 670 - the size had jumped to 421.9mb. I rebooted the computer, and the size is still at 421.9mb.

I had not done tonight's update with the tutorial fix, so that can't account for the growth in size.

I had installed a fresh client in November. My program had grown to over 1.4gb and was very slow. I suspected that the old puzzle files were making it that way, since they are located within the mac program and load every time you open the program. With a fresh install, it was like magic, nice and zippy again. I mentioned this to frood (also on a mac), who did a fresh install a day ago - it worked well for him, too, and he mentioned in global that the program size was now around 300mb.

It would appear then that even my backed up client that is not updated from a few days ago is also growing, at 351mb - but not at the rate of 13mb a day!

Something is very wrong here.

Hanto_FZ140E_4G Lv 1

I cannot deny a small possibility that folder sizes might be growing, but at the moment they are well within what I consider normal in that I am using 3.61 GBs out of 10 GB ( partition size limit ) on 8 folders and 16 clients. I have seen it higher but not by much when everything is working. In the past when this growth was extreme, it would fill up my 10 gig partition within two days. I have not removed or re-installed a folder in a year now, so at the moment see nothing extreme YET. Now that is only for HD storage capacity.

As far as client memory usage during processing between 600 to 800 KBs has always been normal for me and that still seems to be the case on all of my clients. There is probably extra memory usage on mutable puzzles with mutating recipes running, so far none have gone above 800 KBs tho

On 670, I currently have 6 clients on BP's aforementioned devprev. In the several days I have been running 670, I have suffered 4 client quits ( dislike calling them crashes as the client is still there, just quits working ) with no LESS then 4 670 clients continuing to run in all cases. The client quits were quick to recover from and have been running consistently since then, always with mutating recipes on them.

There has NOT been a single glitch on my non mutable puzzles/clients using devprev of concern since the update to the devprev. If there is a memory leak, I suggest it might be in the mutating function of the client only and also be very hard to find.

BTW, the Nvidia 310.70 drivers still working great for foldit Win 7 SP-1 x64

Jean-Bob Lv 1

@Hanto: It's not per se in the mutate function but in the symmetry (AFAIK there's no problem in design puzzles with no symmetry), as for each mutate/shake done the client has to check how it interacts with the other monomers. The cost of such computing grows fast, and if there's the tiniest memleak then…

This feedback is a duplicate, it mentions https://fold.it/portal/node/991306 where Bletchley Park talks about an even older feedback I can't find. So this one may be also a duplicate.
Other duplicate: https://fold.it/portal/node/994347
Creating multiple duplicates doesn't help to solve the issue, please stay focused.

Bletchley Park Lv 1

Thank you for pointing this out to me Jean-Bob, I did not recall where I reported that.
I notice that the second one you mention is assigned to 'anonymous' whereas the first one is assigned to Seth Cooper.

The 'original' mutate causes memory leak mention is in a puzzle comment: