Tracking for 4.8.z Description of problem: Opened as a follow-up to https://bugzilla.redhat.com/show_bug.cgi?id=1996660 Observed an additional slower goroutine leak. How reproduceible: Very reproduceible, must be run for an extended time to observe increase. Steps to Reproduce: Allow for any stabilization between each step. While observing goroutines metric on virt-handler, 1. Create and Start a batch of 100-300 VMs 2. Delete all VMs in batch 3. Repeat Expected Results: Goroutines do not increase between VM deployments after initial deployment Actual Results: Over the course of approximately 48 hours, goroutine count increased between 14-28 per virt-handler.
Because the goroutine leak is in virt-handler, it may be possible to reproduce it in a smaller cluster. Sarah, do you remember approximately how many VMs were running on each node? Can you share the VM definition that you used?
Created attachment 1829079 [details] test VM yaml Hi Andrej, I deployed 100-300 VMs across three worker nodes using the attached VM definition, so should have been somewhere around 30-100 VMs per node. Process should be repeated continuously over the test period, but allowing time for the control plane to stabilize between state transitions (start/delete/start/delete/...). On the three worker node cluster, allowed approximately 10-15 minutes for stabilization between states.
*** This bug has been marked as a duplicate of bug 2004299 ***