tweak GC

[quinox]

last time I debugged I didn't find any leaks, but it seems the Python GC doesn't run often enough or something. Maybe we can add a 15-minute timer to force the collection.

Reference material:

 http://guppy-pe.sourceforge.net/  http://nedbatchelder.com/blog/200809/a_server_memory_leak.html  http://code.google.com/p/pympler/

[quinox]

I've added a Bash script to the SVN ( here) to track memory usage.

[quinox]

It seems it's more fundamental that CG. At startup:

12:07:32        24773      7.00      0.00  446264  71464   1.77  nicotine
12:07:47        24773      3.87      0.00  446264  71296   1.76  nicotine
12:08:02        24773      1.60      0.07  446264  61296   1.52  nicotine
12:08:17        24773      3.47      1.07  446264  60700   1.50  nicotine
12:08:32        24773      4.80      0.80  446264  60904   1.51  nicotine
12:08:47        24773      0.07      0.00  446264  60904   1.51  nicotine
12:09:02        24773      0.27      0.00  446264  60904   1.51  nicotine
12:09:17        24773      2.00      0.00  446264  60904   1.51  nicotine
12:09:32        24773      0.47      0.00  446264  60904   1.51  nicotine

then it stays around 1.51% for a long time. Then right when I open up filelists (either remote or local) it explodes:

13:56:47        24773    210.07      0.00  554964 117040   2.90  nicotine
13:57:02        24773    420.93      0.00  554964 117040   2.90  nicotine
13:57:17        24773    514.80      0.00  573640 135612   3.35  nicotine
13:57:32        24773     25.73      0.00  575092 137112   3.39  nicotine
13:57:47        24773      0.00      0.00  575092 137112   3.39  nicotine
13:58:02        24773      0.00      0.00  575092 137112   3.39  nicotine
13:58:17        24773      0.00      0.00  575092 137112   3.39  nicotine
13:58:32        24773    106.87      0.00  589724 143488   3.55  nicotine
13:58:47        24773    102.80      0.00  595868 149632   3.70  nicotine
13:59:02        24773      0.07      0.00  587672 149628   3.70  nicotine
13:59:17        24773      0.00      0.00  587672 149628   3.70  nicotine

And it never goes down, even after forcing the CG it stays at 3.7%

[quinox]

After opening and closing 61 filelists:

14:10:48        28725      0.47      0.00  976508 636324  15.74  nicotine

Heapy reports the diff. in memory:

Partition of a set of 767 objects. Total size = 117832 bytes.
 Index  Count   %     Size   % Cumulative  % Kind (class / dict of class)
     0    234  31    33992  29     33992  29 str
     1      7   1    23464  20     57456  49 dict of pynicotine.transfers.Transfer
     2    140  18    10760   9     68216  58 tuple
     3    100  13     9600   8     77816  66 time.struct_time
     4     14   2     3920   3     81736  69 dict of 0x7f89befee8c0
     5     29   4     3304   3     85040  72 list
     6      3   0     3144   3     88184  75 dict of pynicotine.pynicotine.PeerConnection
     7      3   0     3144   3     91328  78 dict of threading._Timer
     8     10   1     2800   2     94128  80 dict of pynicotine.pynicotine.UserAddr
     9      9   1     2520   2     96648  82 dict (no owner)

So it seems we're not leaking any python objects. The memory usage doesn't go up evenly with the amount of filelists we open either.

[OffHand]

I wonder if this change fucks up the OS X build. I'll test that tonight.

[irredgecarcix]

logically