differential-dataflow: Possible memory leak in dogs^3

This might be just another manifestation of #234, but submitting anyway in case it’s not. Using the same dataflow graph, I insert a single graph edge and two nodes and then remove them at the next epoch in a loop. All is fine with one worker, but with -w 3 I observe that the memory footprint of the program keep growing monotonically. The repro is in the dogs_leak (no pun intended) branch of my DD fork:

https://github.com/ryzhyk/differential-dataflow/tree/dogs_leak

About this issue

  • Original URL
  • State: closed
  • Created 5 years ago
  • Comments: 15 (15 by maintainers)

Commits related to this issue

Most upvoted comments

I’m sorry, I read your notes badly and didn’t observe the multiple workers requirement! I’m definitely seeing growth with that in place. I’ll start pulling some numbers down. The communication mechanisms are different with multiple workers, which could be related. Await further information!