we are using Midonet in version 5.6.0 with Ubuntu16.06 and an OpenStack Pike installation. At this early stage we have two hypervisors and two gateway nodes in addition with integrated VTEPs on Cumulus switches.
Our Midolmans on the gateway nodes get constantly restarted after a flow state update (coming from the other gateway). The gateways are configured to be the flooding proxies (forced with flooding-proxy-weight).
directly bevor an exception we see a new nat mapping for a hardware server connected via VTEP
there is always the same index (234) and capacity (204) in the exception
there are no exceptions if we shut down the servers behind the hardware VTEPs
if we force the flooding proxy to a hypervisor, the exceptions occur in the same moment on both gateways
the connected servers and the setup didn't change during our tests
We got the following debug logs from a midolman on a gateway:
Do you need more information or debug logs? How can we help to narrow down the problem?