Since upgrading to 12.1-release, we've been having intermittent connectivity issues with FreeBSD guests using the vmx driver (VMXNET3) with our ESXi hosts after vmotion migration. We noticed the vmx driver was ported to iflib in 12.1 (see: https://reviews.freebsd.org/rS343291).
It seems the symptom we're having was reported as well: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=236999. It was reported fixed in 12.1-release-p8: https://www.freebsd.org/security/advisories/FreeBSD-EN-20:16.vmx.asc.
Unfortunately, the issue persists for us on 12.1-release-p13. No errors are present indicating any cause of the issue in the logs, but we see massive packet loss when pinging the guest's gateway. The only immediate remedy seems to be a guest reboot - which does resolve the issue until the next vmotion migration. It's a sporadic issue and seems to be more prevalent on guests with heavy network utilization prior to and during vmotion migration.
Again, we've only seen this symptom since the iflib port of the vmx driver. We've been testing disabling TSO in the mean time - which does look promising. Swapping to the e1000 nic/driver also seems to resolve it.
I'm not sure I have enough info to log a new bug report. Is anyone else seeing this issue still after the fix in 12.1-release-p8?
Any ideas or feedback are greatly appreciated.
It seems the symptom we're having was reported as well: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=236999. It was reported fixed in 12.1-release-p8: https://www.freebsd.org/security/advisories/FreeBSD-EN-20:16.vmx.asc.
Unfortunately, the issue persists for us on 12.1-release-p13. No errors are present indicating any cause of the issue in the logs, but we see massive packet loss when pinging the guest's gateway. The only immediate remedy seems to be a guest reboot - which does resolve the issue until the next vmotion migration. It's a sporadic issue and seems to be more prevalent on guests with heavy network utilization prior to and during vmotion migration.
Again, we've only seen this symptom since the iflib port of the vmx driver. We've been testing disabling TSO in the mean time - which does look promising. Swapping to the e1000 nic/driver also seems to resolve it.
I'm not sure I have enough info to log a new bug report. Is anyone else seeing this issue still after the fix in 12.1-release-p8?
Any ideas or feedback are greatly appreciated.
