[sugar] Memory leaks summary
Marco Pesenti Gritti
mpgritti at gmail.com
Thu Sep 11 21:31:15 EDT 2008
Hello,
here is a summary of what we know so far about the memory leaks
related to network activity (buddies joining/leaving).
1 We was leaking the palette that we create for each buddy. Tomeu
fixed it, should be in the next stable build.
2 We are still leaking the palette icons. It should not be too hard to
track that down. Failing that we can make the palette lazy to work
around the problem for 8.2.0. It's something we should be doing anyway
to save memory and cpu.
3 If I disable the palette completely I still get some leaking. I'm
using a script that continuously create and destroy 10 buddies. Here
is typical relative memory consumption reported by guppy (running it
in a loop of 5 seconds). Note that I completely disabled the sugar
shell model and view before taking this data.
Partition of a set of 1108 objects. Total size = 104160 bytes.
Index Count % Size % Cumulative % Kind (class / dict of class)
0 174 16 23664 23 23664 23 dict (no owner)
1 82 7 15144 15 38808 37 str
2 50 5 10680 10 49488 48 unicode
3 120 11 7920 8 57408 55 dbus.String
4 80 7 7360 7 64768 62 dbus.connection.SignalMatch
5 10 1 5280 5 70048 67 dbus.Dictionary
6 10 1 5200 5 75248 72 dict of 0x94254cc
7 10 1 5200 5 80448 77 dict of dbus.proxies.ProxyObject
8 80 7 4480 4 84928 82 function
9 91 8 4448 4 89376 86 list
<16 more rows. Type e.g. '_.more' to view.>
Does that mean we have a leak in sugar.presence?
The code I'm using:
diff --git a/src/main.py b/src/main.py
index 1ccfbb8..3ab0e52 100644
--- a/src/main.py
+++ b/src/main.py
@@ -79,8 +79,17 @@ def _software_update_cb():
home_box = shell.home_window.get_home_box()
home_box.show_software_updates_alert()
+import guppy
+heapy = guppy.hpy()
+
+def _heapy_cb():
+ print heapy.heap()
+ heapy.setref()
+ return True
+
def main():
gobject.idle_add(_shell_started_cb)
+ gobject.timeout_add(5000, _heapy_cb)
try:
logsmanager.setup()
4) The grid layout is leaking objects when collision solving fails
(i.e. you have a very full mesh view). We are probably not hitting
this case in testing, but still should be fixed.
Marco
More information about the Sugar
mailing list