Score:0

Processes from distributed GPU usage are hanging and can't be killed

dz flag

I was using the PyTorch library's data distributed parallel library in order to use multiple GPU devices when running my machine learning program.

After running, I noticed that three processes were still running and using GPU memory. Funnily enough, the Proecss name column of the nvidia-smi output was empty for these.

When I tried running kill -9 $PID I get -bash: kill: ($PID) - No such process. I've tried to identify them using htop but can't find them either. When I run ps aux | grep 'Z' because I thought they're zombie processes, nothing gets output.

How should I effectively identify and kill these hanging processes?

Edit

I managed to kill these processes via killall5, but this also wiped out all of my Tmux sessions.

I sit in a Tesla and translated this thread with Ai:

mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.