Occasionally interface event is lost

[yue-fred-gao]

This is seen once but it is worth to record the analysis.

1. Below message is seen in the log: syncd vl_api_sw_interface_event_t_handler:632: vpp cannot get interface name for sw index 25, which means interface is not found in interface_name_by_sw_index.
2. When an interface is up, vpp sends the interface-up event to the client (SONiC), which will be stored in a message queue. If it happens that refresh_interfaces_list is called before the event is picked up by vpp event polling thread (SwitchStateBase::vppProcessEvents is the handler), a race condition will happen causing this problem. Here are the details:

• the interface_name_by_sw_index will be cleared first in api_sw_interface_dump
• sending SW_INTERFACE_DUMP and PING API requests to vpp to get interface dump
• calling W to process the responses from the message queue. If the interface up event is already received and currently stored in the queue, vl_api_sw_interface_event_t_handler callback will be invoked when processing the event. Because interface_name_by_sw_index is already cleared, we will end up the error message "vpp cannot get interface name".
• Potential solution: call vpp_sync_for_events first in refresh_interfaces_list. This will process all the events in the queue. Ultimate solution is changing to some async model, which requires some investigation.