TlsTest flakiness

[zhangkun83]

Found on Travis linux when testing for #1508

io.grpc.testing.integration.TlsTest > noClientAuthFailure[OPENSSL] FAILED
    java.lang.AssertionError: expected:<UNAVAILABLE> but was:<UNKNOWN>
        at org.junit.Assert.fail(Assert.java:88)
        at org.junit.Assert.failNotEquals(Assert.java:743)
        at org.junit.Assert.assertEquals(Assert.java:118)
        at org.junit.Assert.assertEquals(Assert.java:144)
        at io.grpc.testing.integration.TlsTest.noClientAuthFailure(TlsTest.java:235)

[ejona86]

And I just saw this on #1511:

io.grpc.testing.integration.TlsTest > serverRejectsUntrustedClientCert[OPENSSL] FAILED
    java.lang.AssertionError: expected:<UNAVAILABLE> but was:<UNKNOWN>
        at org.junit.Assert.fail(Assert.java:88)
        at org.junit.Assert.failNotEquals(Assert.java:743)
        at org.junit.Assert.assertEquals(Assert.java:118)
        at org.junit.Assert.assertEquals(Assert.java:144)
        at io.grpc.testing.integration.TlsTest.serverRejectsUntrustedClientCert(TlsTest.java:195)
io.grpc.testing.integration.TlsTest > noClientAuthFailure[OPENSSL] FAILED
    java.lang.AssertionError: expected:<UNAVAILABLE> but was:<UNKNOWN>
        at org.junit.Assert.fail(Assert.java:88)
        at org.junit.Assert.failNotEquals(Assert.java:743)
        at org.junit.Assert.assertEquals(Assert.java:118)
        at org.junit.Assert.assertEquals(Assert.java:144)
        at io.grpc.testing.integration.TlsTest.noClientAuthFailure(TlsTest.java:235)

[ejona86]

From #1505

io.grpc.testing.integration.TlsTest > clientRejectsUntrustedServerCert[OPENSSL] FAILED
    java.lang.AssertionError: expected:<UNAVAILABLE> but was:<UNKNOWN>
        at org.junit.Assert.fail(Assert.java:88)
        at org.junit.Assert.failNotEquals(Assert.java:743)
        at org.junit.Assert.assertEquals(Assert.java:118)
        at org.junit.Assert.assertEquals(Assert.java:144)
        at io.grpc.testing.integration.TlsTest.clientRejectsUntrustedServerCert(TlsTest.java:278)

[ejona86]

Saw noClientAuthFailure in #1509

[carl-mastrangelo]

@ejona86 I can't imagine how #1505 might have caused that. Another close by commit perhaps?

[ejona86]

I'm guessing this change has exposed a pre-existing flake. Before we would randomly get a useless UNAVAILABLE Status without any description of what went wrong, but now it is showing up as UNKNOWN.

We could revert that one part of that CL and probably unflake ourselves. I'm investigating.