There are only 5% of failing jobs when this shows up, so it's not causing us huge issues in the gate apparently. I'd bet if we look at those failures there are other things failing due to messaging timeouts which are waiting on some response, like bug 1371677.
There are 1450 hits of this in 7 days in the check/gate queues:
message:"model server went away" AND tags:"screen- n-cpu.txt"
http:// logstash. openstack. org/#eyJzZWFyY2 giOiJtZXNzYWdlO lwibW9kZWwgc2Vy dmVyIHdlbnQgYXd heVwiIEFORCB0YW dzOlwic2NyZWVuL W4tY3B1LnR4dFwi IiwiZmllbGRzIjp bXSwib2Zmc2V0Ij owLCJ0aW1lZnJhb WUiOiI2MDQ4MDAi LCJncmFwaG1vZGU iOiJjb3VudCIsIn RpbWUiOnsidXNlc l9pbnRlcnZhbCI6 MH0sInN0YW1wIjo xNDExMjMwMjYxNj I2LCJtb2RlIjoiI iwiYW5hbHl6ZV9m aWVsZCI6IiJ9
There are only 5% of failing jobs when this shows up, so it's not causing us huge issues in the gate apparently. I'd bet if we look at those failures there are other things failing due to messaging timeouts which are waiting on some response, like bug 1371677.