I used ParaView master with VTK master (after your branch was merged).
Thanks ! @shawn.waldon : could you do the same with Vall ?
vall test (again this is ParaView master with VTK master):
Fedora 28, upgraded from previous versions (installed as 24 I think).
Quadro K420
Driver version: 396.24 (installed from this ppa: https://negativo17.org/nvidia-driver/)
Kernel: 4.16.16-300
Xorg: 1.19.6
Qt: 5.9.2
Gnome: 3.28.2
Memory usage from nvidia-smi:
Before:
Global: 99MiB / 1996MiB
/usr/libexec/Xorg 32MiB
/usr/bin/gnome-shell 64MiB
After:
Global: 244MiB / 1996MiB
/usr/libexec/Xorg 125MiB
/usr/bin/gnome-shell 113MiB
edit: I will add that this is definitely the problem. When I went to run the test, I had to reboot vall since the display was locked (nvidia-smi before the reboot confirmed that all graphics memory was used up).
@Michael @Joachim_Pouderoux : since you do not reproduce the bug, could you post your own results here ?
@Joachim_Pouderoux I updated my results above with the ParaView and VTK SHAâs used.
Update: from watching the dashboard as it tests builds and trying nvidia-smi on it:
A typical build of PV master leaks 500+MB of graphics memory. So the branch is a major improvement even if it doesnât fix everything.
+1 for an update VTK branch to pull in the improvement.
The MR is already up but another fix seems to break a single test test on ista. We are looking into it.
paraview/paraview!2511
Tylo seems to fail again, could you check itâs memory ?
shark test (ParaView master with VTK master (on July 1st)):
Xubuntu 18.04
GeForce GTX 660
4K Screen resolution
Driver version: 390.48
Kernel: 4.15.0-20
Xorg: 1.19.6
Qt: 5.9.5
Xfce 4.12 with Compositor ENABLED
Memory usage from nvidia-smi:
Before:
Global: 302MiB / 1999MiB
After:
Global: 582MiB / 1999MiB
LEAK!
PS: I did not compile PV with testing enable to build it faster so I use the following command line:
for i in {1..20}; do bin/paraview --test-script=/path_to_paraview_sources/Applications/ParaView/Testing/XML/UndoRedo2.xml --exit; done
For info I used a minimal ParaView build: TEST=OFF, MPI=OFF, Python=OFF, all_PLUGINS=OFF, Catalyst=OFF
If I turn off the Compositor (xfwm4-tweaks-settings):
NO LEAK!
Attached is a simple test that leaks.
simple.xml (695 Bytes)
Guys, after a rigourous round of tests, the issue is now always reproducible whatever the configuration.
I will keep investigating.
In the meantime, the different fixes that have landed on master recently should limit a little bit the leak.
In any case, if we want to have working buildbot we will need to restart X regularly until this is fixed.
I will manage luigi, but vall (and tylo?) will need to be managed from your side.
Fixed It !
Caused by a Qt Issue and an incorrect null parenting before deletion in pqMultiViewWidget.
See Qt issue and MR.
(I am so happy right now 
)
2 Likes
Congratulations! 
Awesome! Thanks for persisting!
