Endpoint becomes unresponsive after several hours

[hickeng]

The following is seen in journalctl output:

Mar 20 08:32:11 Linux systemd[1]: vic-init.service: Main process exited, code=exited, status=2/INVALIDARGUMENT
Mar 20 08:32:11 Linux systemd[1]: vic-init.service: Unit entered failed state.
Mar 20 08:32:11 Linux systemd[1]: vic-init.service: Failed with result 'exit-code'.

Current speculation is that this is related to DHCP lease failure on the management network, but no clarity on why this would result in vic-init exit.

Speculative fix is to use --asymmetric-routes but that's pure speculation due to martian packets being logged.

Bug2084707 tracks this and contains the original details and logs.

[hickeng]

@dbarkelew Opened this issue for tracking. If you manage a recreate please let me know.

[hickeng]

Added to the release as we've had several customers run into this. We need a least a root cause and documentation covering it for 1.4 if not a code fix.

[gigawhitlocks]

Moving back to ToDo to pick up a higher priority issue, sorry for all the pipeline spam in the history of this ticket.

[hickeng]

Next step:
Modify the dhcp client to support asymmetric routes. rp_filter may be configured correctly for asymmetry but I suspect that given the dhcp client is listening on a specific interface we will not be getting the benefit of that.

This could also address bug2139912 if it the asymmetric routing configuration was propagated into the containerVMs (maybe associated with a specific network?).

[zjs]

I think this is worth a release note, as it was a customer-reported issue.

[stuclem]

Added as resolved issue in https://github.com/vmware/vic/releases/tag/v1.4.3