So far we have seen tier1 failures in every single Mach 5 run in jdk9/hs. One test, TestRelease.java, has failed in every single Mach 5 run. In some cases this is the only failure. It fails on different platforms different runs.

The Mach 5 builds I have looked at are the ones labeled as "Fail", I have not looked at the "Incomplete" ones. These are build # 31, 40, 41, 42, 43, 44, 45, 46, 47, 48, 50, 52, 55, 57.

I have sampled five of these and they all contain the same error message:
test result: Error. Agent communication error: java.io.EOFException; check console log for any additional details

Looking at the stdout that is stored in the archive I find:

bash> ACTION: testng -- Error. Agent communication error: java.io.EOFException; check console log for any additional details
bash> REASON: User specified action: run testng jdk.jdeprscan.TestRelease
bash> TIME: 5.765 seconds
bash> messages:
bash> command: testng jdk.jdeprscan.TestRelease
bash> reason: User specified action: run testng jdk.jdeprscan.TestRelease
bash> Mode: agentvm
bash> Additional exports to unnamed modules from @modules: jdk.jdeps/com.sun.tools.jdeprscan
bash> elapsed time (seconds): 5.765

I don't disagree that the reason for the failure is that we ran the test, but I was hoping for a little more detail there.

Also worth noting that even though the test sometimes fails on windows and solaris, there are no files collected from these platforms in the archive.


This test is currently causing a lot of noise up to the point where people risk start ignoring the failure mails. There is also the issue that we are not supposed to integrate jdk9/hs to jdk9/dev unless we have a green Mach 5 build.