[0.64.1] Memory leak in rendering animation.

Fenix · February 6, 2014, 6:21pm

Hi.

I’m using 0.64.1 Synfig (packaged by Debian Testing). I was rendering an animation when I notice that the use of memory increased a lot. I put an eye on and see that the used memory going on without brake. So, I need to restart Synfig every time the use of memory get danger numbers. It is not funny.

I was rendering 1080x70 30fps CairoPNG.

I was looking at bug tracker but I can find nothing about memory leaks… so, could anyone confirme that?

Genete · February 7, 2014, 6:54am

Do you have same issue using normal PNG export (Software renderer)?
Do you use images on your animation?
-G

Yoyobuae · February 7, 2014, 4:44pm

There’s definitely a leak in CairoPNG target:

==2743== 97,088,476 (423,900 direct, 96,664,576 indirect) bytes in 1,413 blocks are definitely lost in loss record 11,295 of 11,296 ==2743== at 0x4023F50: malloc (vg_replace_malloc.c:236) ==2743== by 0x516AA88: _cairo_image_surface_create_for_pixman_image (cairo-image-surface.c:176) ==2743== by 0x516AB77: _cairo_image_surface_create_with_pixman_format (cairo-image-surface.c:345) ==2743== by 0x516AE82: cairo_image_surface_create (cairo-image-surface.c:387) ==2743== by 0x6DED06B: cairo_png_trgt::obtain_surface(_cairo_surface*&) (in /tmp/synfigstudio-0.64.1-20131104.8.i386/lib/synfig/modules/libmod_png.so) ==2743== by 0x83C122A: AsyncTarget_Cairo::obtain_surface(_cairo_surface*&) (in /tmp/synfigstudio-0.64.1-20131104.8.i386/bin/synfigstudio) ==2743== by 0x4294CA3: synfig::Target_Cairo::render(synfig::ProgressCallback*) (in /tmp/synfigstudio-0.64.1-20131104.8.i386/lib/libsynfig.so.0.0.0) ==2743== by 0x83BE940: studio::AsyncRenderer::render_target() (in /tmp/synfigstudio-0.64.1-20131104.8.i386/bin/synfigstudio) ==2743== by 0x4F0DB07: call_thread_entry_slot (slot.h:440) ==2743== by 0x53C9A8E: ??? (in /lib/libglib-2.0.so.0.2800.6) ==2743== by 0x4679954: start_thread (pthread_create.c:300) ==2743== by 0x563A1DD: clone (clone.S:130) ==2743== ==2743== 650,592,000 bytes in 1,255 blocks are possibly lost in loss record 11,296 of 11,296 ==2743== at 0x402328F: calloc (vg_replace_malloc.c:467) ==2743== by 0x5B41559: ??? (in /usr/lib/libpixman-1.so.0.24.0) ==2743== by 0x5B41629: pixman_image_create_bits (in /usr/lib/libpixman-1.so.0.24.0) ==2743== by 0x516AB62: _cairo_image_surface_create_with_pixman_format (cairo-image-surface.c:339) ==2743== by 0x516AE82: cairo_image_surface_create (cairo-image-surface.c:387) ==2743== by 0x6DED06B: cairo_png_trgt::obtain_surface(_cairo_surface*&) (in /tmp/synfigstudio-0.64.1-20131104.8.i386/lib/synfig/modules/libmod_png.so) ==2743== by 0x83C122A: AsyncTarget_Cairo::obtain_surface(_cairo_surface*&) (in /tmp/synfigstudio-0.64.1-20131104.8.i386/bin/synfigstudio) ==2743== by 0x4294CA3: synfig::Target_Cairo::render(synfig::ProgressCallback*) (in /tmp/synfigstudio-0.64.1-20131104.8.i386/lib/libsynfig.so.0.0.0) ==2743== by 0x83BE940: studio::AsyncRenderer::render_target() (in /tmp/synfigstudio-0.64.1-20131104.8.i386/bin/synfigstudio) ==2743== by 0x4F0DB07: call_thread_entry_slot (slot.h:440) ==2743== by 0x53C9A8E: ??? (in /lib/libglib-2.0.so.0.2800.6) ==2743== by 0x4679954: start_thread (pthread_create.c:300) ==2743==

Tested with synfigstudio-0.64.1-20131104.8.i386 using valgrind tool rendering a one minute test animation (just a circle moving) at 24 FPS (1441 frames).

EDIT: Same test with same version using PNG target instead shows no leak.

Fenix · February 7, 2014, 5:47pm

Yes.

Here you are Valgrind report. This is the worst because the amount of memory:

PNG export. Only two frames render.

<error>
  <unique>0x27df</unique>
  <tid>1</tid>
  <kind>Leak_DefinitelyLost</kind>
  <xwhat>
    <text>2,425,896 (15,300 direct, 2,410,596 indirect) bytes in 51 blocks are definitely lost in loss record 10,137 of 10,137</text>
    <leakedbytes>2425896</leakedbytes>
    <leakedblocks>51</leakedblocks>
  </xwhat>
  <stack>
    <frame>
      <ip>0x4028A28</ip>
      <obj>/usr/lib/valgrind/vgpreload_memcheck-x86-linux.so</obj>
      <fn>malloc</fn>
      <dir>/build/valgrind-x3RsCO/valgrind-3.8.1/coregrind/m_replacemalloc</dir>
      <file>vg_replace_malloc.c</file>
      <line>270</line>
    </frame>
    <frame>
      <ip>0x4EF77B5</ip>
      <obj>/usr/lib/i386-linux-gnu/libcairo.so.2.11200.16</obj>
    </frame>
    <frame>
      <ip>0x4EF7BAE</ip>
      <obj>/usr/lib/i386-linux-gnu/libcairo.so.2.11200.16</obj>
    </frame>
    <frame>
      <ip>0x4EF7E7D</ip>
      <obj>/usr/lib/i386-linux-gnu/libcairo.so.2.11200.16</obj>
      <fn>cairo_image_surface_create</fn>
    </frame>
    <frame>
      <ip>0x42AEE69</ip>
      <obj>/usr/lib/i386-linux-gnu/libsynfig.so.0.0.0</obj>
      <fn>synfig::Target_Cairo_Tile::render(synfig::ProgressCallback*)</fn>
    </frame>
    <frame>
      <ip>0x82DDCAA</ip>
      <obj>/usr/bin/synfigstudio</obj>
      <fn>studio::AsyncRenderer::render_target()</fn>
    </frame>
    <frame>
      <ip>0x4DB3DE1</ip>
      <obj>/usr/lib/i386-linux-gnu/libglibmm-2.4.so.1.3.0</obj>
    </frame>
    <frame>
      <ip>0x5381C3D</ip>
      <obj>/lib/i386-linux-gnu/i686/cmov/libc-2.17.so</obj>
      <fn>clone</fn>
      <dir>/build/eglibc-an0j3i/eglibc-2.17/misc/../sysdeps/unix/sysv/linux/i386</dir>
      <file>clone.S</file>
      <line>131</line>
    </frame>
  </stack>
</error>

This is Cairo PNG render. Two frames too.

<error>
  <unique>0x2868</unique>
  <tid>1</tid>
  <kind>Leak_DefinitelyLost</kind>
  <xwhat>
    <text>1,964,152 (14,700 direct, 1,949,452 indirect) bytes in 49 blocks are definitely lost in loss record 10,259 of 10,260</text>
    <leakedbytes>1964152</leakedbytes>
    <leakedblocks>49</leakedblocks>
  </xwhat>
  <stack>
    <frame>
      <ip>0x4028A28</ip>
      <obj>/usr/lib/valgrind/vgpreload_memcheck-x86-linux.so</obj>
      <fn>malloc</fn>
      <dir>/build/valgrind-x3RsCO/valgrind-3.8.1/coregrind/m_replacemalloc</dir>
      <file>vg_replace_malloc.c</file>
      <line>270</line>
    </frame>
    <frame>
      <ip>0x4EF77B5</ip>
      <obj>/usr/lib/i386-linux-gnu/libcairo.so.2.11200.16</obj>
    </frame>
    <frame>
      <ip>0x4EF7BAE</ip>
      <obj>/usr/lib/i386-linux-gnu/libcairo.so.2.11200.16</obj>
    </frame>
    <frame>
      <ip>0x4EF7E7D</ip>
      <obj>/usr/lib/i386-linux-gnu/libcairo.so.2.11200.16</obj>
      <fn>cairo_image_surface_create</fn>
    </frame>
    <frame>
      <ip>0x42AEE69</ip>
      <obj>/usr/lib/i386-linux-gnu/libsynfig.so.0.0.0</obj>
      <fn>synfig::Target_Cairo_Tile::render(synfig::ProgressCallback*)</fn>
    </frame>
    <frame>
      <ip>0x82DDCAA</ip>
      <obj>/usr/bin/synfigstudio</obj>
      <fn>studio::AsyncRenderer::render_target()</fn>
    </frame>
    <frame>
      <ip>0x4DB3DE1</ip>
      <obj>/usr/lib/i386-linux-gnu/libglibmm-2.4.so.1.3.0</obj>
    </frame>
    <frame>
      <ip>0x5381C3D</ip>
      <obj>/lib/i386-linux-gnu/i686/cmov/libc-2.17.so</obj>
      <fn>clone</fn>
      <dir>/build/eglibc-an0j3i/eglibc-2.17/misc/../sysdeps/unix/sysv/linux/i386</dir>
      <file>clone.S</file>
      <line>131</line>
    </frame>
  </stack>
</error>

With Cairo render there is a possible lost with a big amount:

<error>
  <unique>0x2869</unique>
  <tid>1</tid>
  <kind>Leak_PossiblyLost</kind>
  <xwhat>
    <text>7,372,800 bytes in 2 blocks are possibly lost in loss record 10,260 of 10,260</text>
    <leakedbytes>7372800</leakedbytes>
    <leakedblocks>2</leakedblocks>
  </xwhat>
  <stack>
    <frame>
      <ip>0x402A9B5</ip>
      <obj>/usr/lib/valgrind/vgpreload_memcheck-x86-linux.so</obj>
      <fn>calloc</fn>
      <dir>/build/valgrind-x3RsCO/valgrind-3.8.1/coregrind/m_replacemalloc</dir>
      <file>vg_replace_malloc.c</file>
      <line>593</line>
    </frame>
    <frame>
      <ip>0x5BFFFE7</ip>
      <obj>/usr/lib/i386-linux-gnu/libpixman-1.so.0.32.4</obj>
    </frame>
    <frame>
      <ip>0x5C00085</ip>
      <obj>/usr/lib/i386-linux-gnu/libpixman-1.so.0.32.4</obj>
    </frame>
    <frame>
      <ip>0x4EF7B9C</ip>
      <obj>/usr/lib/i386-linux-gnu/libcairo.so.2.11200.16</obj>
    </frame>
    <frame>
      <ip>0x4EF7E7D</ip>
      <obj>/usr/lib/i386-linux-gnu/libcairo.so.2.11200.16</obj>
      <fn>cairo_image_surface_create</fn>
    </frame>
    <frame>
      <ip>0x82DFCA8</ip>
      <obj>/usr/bin/synfigstudio</obj>
      <fn>AsyncTarget_Cairo::obtain_surface(_cairo_surface*&amp;)</fn>
    </frame>
    <frame>
      <ip>0x82DDCAA</ip>
      <obj>/usr/bin/synfigstudio</obj>
      <fn>studio::AsyncRenderer::render_target()</fn>
    </frame>
    <frame>
      <ip>0x4DB3DE1</ip>
      <obj>/usr/lib/i386-linux-gnu/libglibmm-2.4.so.1.3.0</obj>
    </frame>
    <frame>
      <ip>0x5381C3D</ip>
      <obj>/lib/i386-linux-gnu/i686/cmov/libc-2.17.so</obj>
      <fn>clone</fn>
      <dir>/build/eglibc-an0j3i/eglibc-2.17/misc/../sysdeps/unix/sysv/linux/i386</dir>
      <file>clone.S</file>
      <line>131</line>
    </frame>
  </stack>
</error>

All the test was with Cairo activate in navigator and workspace. I can made a new testing desactivating all Cairo… but I think the problem is in rendering.

Yes. But only one. Mainly are vector objects made inside Synfig.

Fenix · February 7, 2014, 6:18pm

In my test there is leak in PNG export, too.

Is really strange.

Yoyobuae · February 7, 2014, 7:07pm

Well, the PNG target DID show other leaks but not 750MB in size like the test with CairoPNG. Also I had cairo render disabled in workarea and navigator.

Attached valgrind output for both tests.
valgrind-3935-PNG.zip (248 KB)
valgrind-2743-CairoPNG.zip (247 KB)