Memory leak in CEF GPU process about cef_direct3d_offscreen_rendering

[roche-emmanuel]

Hi mihe, actually I noticed that too: I can quickly get my GPU process to use 15GB (or more) by generating a lot of frame per seconds....

... I'm currently investigating this too. if you have any idea on your side, please let me know. From my perspective, I feel this could be due to some kind of problem in the texture release process: if you kept the logging system in place and you set the env. variable NV_CEF_LOG_LEVEL=2 to generate the "cef_nv.log" I think you could also notice that a lot of new texture ids are generated, but they don't get released "fast enough"... I have to figure out what this is exactly.

[roche-emmanuel]

Hello mihe, just in case you would still be running this code on your side, please note that I updated it to fix the issue we discussed here.

And actually, we are now using this implement in production in my company since the CEF behavior is now stable.

Yet, if you haven't done so yet, I would suggest you have a look at the pull request submitted by Isaac Richards on the CEF repo: that implementation is actually simpler than mine here, and thus it is probably the way to go on the long run.

[mihe]

Thanks for the update @roche-emmanuel . I've put this whole thing on the backburner for now, so I'm not currently using your changes. As you suggested I'm keeping a close eye on the PR by Isaac Richards to see if it matures enough that we can use it safely.