TestBasicHostAccess fails

[unclejack]

The system tests fail with this error:

FAIL: hostaccess_test.go:9: systemtestSuite.TestBasicHostAccess

hostaccess_test.go:26:
    s.hostAccTest(c)
hostaccess_test.go:85:
    c.Assert(s.pingTestToNonContainer(cList, dest), IsNil)
... value *errors.errorString = &errors.errorString{s:"Ping failed from (container: 58158fc062b5166d0c6ce61e9db07459c742e00315888f6a51cb44caad4ead4b (name: \"epg-a-1\" ip: 13.5.7.3 ipv6:  host: netplugin-node2)) to 192.168.2.10: \"PING 192.168.2.10 (192.168.2.10): 56 data bytes\\r\\n\\r\\n--- 192.168.2.10 ping statistics ---\\r\\n3 packets transmitted, 0 packets received, 100% packet loss\\r\\n\" - Process exited with status 1"} ("Ping failed from (container: 58158fc062b5166d0c6ce61e9db07459c742e00315888f6a51cb44caad4ead4b (name: \"epg-a-1\" ip: 13.5.7.3 ipv6:  host: netplugin-node2)) to 192.168.2.10: \"PING 192.168.2.10 (192.168.2.10): 56 data bytes\\r\\n\\r\\n--- 192.168.2.10 ping statistics ---\\r\\n3 packets transmitted, 0 packets received, 100% packet loss\\r\\n\" - Process exited with status 1")

This failure was observed with Docker 1.11.2. It may or may not happen with a different version of Docker.

[unclejack]

This issue was caused by a change in the Docker install script from get.docker.com. The previous version was allowing us to install the old versions of Docker engine (e.g. before the docker-ce change and the ab.cd versioning scheme).

The CI job which was installing Docker 1.11.2 would now install Docker 17.06 CE. I've fixed this issue temporarily by disabling the 1.11.2 job. This disabled job can be enabled again with Docker 17.06-ce.

@DivyaVavili @gkvijay @rchirakk I'll need your help to investigate and fix this issue.

[rchirakk]

Can you check it locally & see the host has ip forwarding turned on ? in this test, a container is trying to ping master ip and is failing.

[unclejack]

@rchirakk: Sure, I'll try this out and comment here.

[unclejack]

I've looked into this for a bit. IP forwarding is enabled on all of the netplugin VMs. The netfilter firewall and NAT rules don't seem to be the cause either.

CONTIV_DOCKER_VERSION=1.11.2 make stop start can be used to bring up a setup which makes this test fail. I'll investigate this further.

[unclejack]

This issue still requires attention.