OutOfMemory Error during routing

[LimesKey]

• What version of freerouting are you using?
  v1.90.0 (latest)

• On what platform did you run it? Windows, Linux, MacOS? 32-bit or 64-bit?
  Windows & Arch Linux

• What EDA do you use? KiCad, KiCad with freerouting plugin, EAGLE, EasyEDA, other? What is its exact version?
  KiCad with freerouting plugin & Standalone

• What types of files are you working with? Can you attach the project files (.pro, .kicad_pcb, etc) or the exported DSN file? (Please note that GitHub doesn't allow DSN file uploads, so you need to ZIP it before uploading.)
  myboard.zip

• Did you use any command line arguments?
  No

• Can you post a screenshot of the problem?
  
  

• Can you include the latest, related log entries from the .log file?
  I'm not sure how to by using the KiCad plugin

• Please list the detailed steps to reproduce the problem
  Freeroute the provided file on a 16GB RAM system

• What would you expect freerouting to do in this situation?
  Freeroute without errors

• Is fixing this issue critical for your work?
  Yes

I get "java.lang.OutofMemoryError: java heap space failed reallocation of scalar replaced objects" and "Java.lang.OutofMemoryError: Heap Space". I'm KiCad with the plugin on a laptop with more than 16GB of available RAM, I've also tried running the file on another computer, same result. I've tried increasing my Windows Page File system but no luck.

[LimesKey]

I tried running the program in Github Codespaces on a 64GB RAM machine but launching the program headless did not work and gave a 'could not find X11 display' error.

[Djzhan0207]

I run this on my windows and mem break too. According to my recent study, when FR can't extend a route, the mem will increase immediately, it relates to its queue data structure.For now, it can't be fixed, you can try another auto-router such as electra.

[andrasfuchs]

I can confirm that this is an issue, thank you @LimesKey for the DSN file.

Here is the profiler's report during my recent session that consumed 8GB of RAM in a matter of minutes:

[andrasfuchs]

I implemented a HashMap-based caching mechanism for some key planar classes to significantly reduce the new object creation count, but it made the algorithm less CPU-efficient because of the thread-safe hash lookups.

I will discard these changes, because the memory need didn't decrease significantly in the end.

I think there must be a separate routing issue as well related to your board, that needs further investigation.

[LimesKey]

Is the bug/issue consider a memory leak or just it working but using up more RAM than intended? I can try it in a Github Codespace with 64GB of RAM if I can solve the headless issue.

I tried Electra last night, and while it was super fast and worked, it conflicted with my graphic on my front copper layer. This is turning into less of a freerouting issues/question, but do you think I would require a third layer because of the complexity of the board? And does anyone know how to prevent the copper graphic from conflciting with the auto routing traces?

[andrasfuchs]

I think the algorithm gets into to a loop incorrectly that uses up as much memory as it can over time, so in that sense it's a memory leak. It's not dangerous or harmful in any way, it's just that it would probably never finish.

I suspect that it is connected to the fact that your board edge is more complex than other boards, but I might be wrong about that. If you have time you can try to test it with a simple rectangular edge.

I can't help you with Electra unfortunately.

[L1uTongweiNewAccount]

In fact, I also met this in some of my projects when my RAM is 32GiB and it was NOT full.

Maybe this is infinty recursion or loop made a stack overflow?

I don't know how it works, but I know some silly bug can make this. I found a repetition in my new issue #299.

Although it will not lead the program to crash, but i think it can rellect some issues.

[L1uTongweiNewAccount]

@andrasfuchs i think this issue is just the bigger issue of #299, and i think if we have more time, we will get the same crash in it.

So to reserch #299, maybe we can fix the bug of memory leak, slow and... anyway!