Created attachment 1539358 [details] libvirtd log Description of problem: Qemu guest agent lost connection after migration if postcopy phase lasts too long(more than 30s) Version-Release number of selected component (if applicable): libvirt-5.0.0-4.el8.x86_64 qemu-kvm-3.1.0-18.module+el8+2834+fa8bb6e2.x86_64 How reproducible: 100% Steps to Reproduce: 1.Start a vm 2.Migrate vm to another host with postcopy enabled and low postcopy bandwidth, and switch migration to postcopy mode: # date; virsh migrate avocado-vt-vm1 qemu+ssh://10.66.5.148/system --live --verbose --p2p --persistent --postcopy --postcopy-bandwidth 5; date Thu Feb 28 12:41:58 CST 2019 Migration: [100 %] Thu Feb 28 12:42:53 CST 2019 # date; virsh migrate-postcopy avocado-vt-vm1 Thu Feb 28 12:42:01 CST 2019 The postcopy phase lasts 52s as shown above, while QEMU_JOB_WAIT_TIME is 30s. 3.After migration finished, check guest agent on target host: # virsh domtime avocado-vt-vm1 error: Guest agent is not responding: QEMU guest agent is not connected Actual results: As step3 Expected results: Guest agent can work after migration. Additional info:
I couldn't reproduce anymore :( I've tried the latest libvirt (v7.5.0-rc1-3-g7c08141f90) and latest qemu (v6.0.0-1995-g5d2d18ae39). Jin, do you still see this problem?
I can reproduce with libvirt-client-7.4.0-1.module+el8.5.0+11218+83343022.x86_64 and qemu-kvm-6.0.0-21.module+el8.5.0+11555+e0ab0d09.x86_64
I will try with upstream version
I can reproduce with upstream libvirt: libvirt-7.5.0-1. But I think this scenario is common, normally postcopy migration will finish very soon. I will change the severity to low.
I can reproduce with upstream libvirt: libvirt-7.5.0-1. But I think this scenario is not common, normally postcopy migration will finish very soon. I will change the severity to low.
After evaluating this issue, there are no plans to address it further or fix it in an upcoming release. Therefore, it is being closed. If plans change such that this issue will be fixed in an upcoming release, then the bug can be reopened.