aarch64 mac jdk11/18 : jdk_nio consistently failed on test-macstadium-macos11-arm64-2

[sophia-guo]

aarch64 mac jdk11 : jdk_nio consistently failed on test-macstadium-macos11-arm64-2

• test suite/name : jdk_nio
• a link into recent Test_ job on Test_openjdk11_hs_extended.openjdk_aarch64_mac_testList_0
• Hyperlink to re-run in Grinder: rerunLink
• Is there an existing issue elsewhere covering this?
• Which machine(s) does it work on? test-macstadium-macos11-arm64-1
• Which machine(s) does it fail on?test-macstadium-macos11-arm64-2

Any other details:
https://trss.adoptium.net/deepHistory?testId=62d7d4f9250c3c428ca02918
adoptium/aqa-tests#3868 (comment)

[sophia-guo]

@sxa @Haroon-Khel

[sxa]

@sophia-guo I guess that since we have a machine it passes on this is not critical for the release?
Also can you add in a grinder re-run link into the description template please?

[smlambert]

Added link above, and here Rerun Failed target in Grinder on same machine

[sophia-guo]

@sxa not critical. Just a note might hit the same issue for other release version.

[sxa]

Agreed. Although I think it's best that we take that penalty for this cycle and re-run as Grinders if required given the limited number of systems we have just now.

[sxa]

Referenced machine is no longer present in the CI and has been replaced with orka machines - re-running on those:

• https://ci.adoptium.net/job/Grinder/11756/ (master) 16:06:29 TEST RESULT: Failed. Execution failed: 'main' threw exception: java.net.SocketException: Resource busy from BasicMulticastTests - RE-GRIND LINK - running that with 1000 iterations
• https://ci.adoptium.net/job/Grinder/11757/ (Original 0.9.3 AQA branch) 16:05:41 Error: Unexpected exception occurred! java.lang.IllegalArgumentException: 7.3.1+1
• https://ci.adoptium.net/job/Grinder/11758/ (master again) ✅ PASSED
• https://ci.adoptium.net/job/Grinder/11759/ (master again, 200 iterations to run over the weekend) Passed 6/400 (Presumably 400 from running in two modes) mostly java.net.SocketException: Exec format error

Noting that it's not clear from this issue what the original failure was.

[sxa]

Also noting that https://ci.adoptium.net/job/Test_openjdk11_hs_extended.openjdk_aarch64_mac/ seems to be passing jdk_nio consistently.

[sxa]

Some more tests:

• 50 runs of jdk_nio on aarch64: https://ci.adoptium.net/job/Grinder/11799/
• 200 runs of the specific test on aarch64: https://ci.adoptium.net/job/Grinder/11800/ ❌ All failed (Exec format error - java -version was ok. Incorrect test material build somewhere?)
• 200 runs of the specific test on x64 (macos10.14 system): https://ci.adoptium.net/job/Grinder/11801 ✅ All passed
• 200 runs of the specific test on x64 (queued on macos10.14): https://ci.adoptium.net/job/Grinder/11802 ✅ All passed
• 200 runs forced to a pre-existing orka node: https://ci.adoptium.net/job/Grinder/11833/

[sxa]

Upstream issue showing a similar error with other tests: https://bugs.openjdk.org/browse/JDK-8144003
As per that bug it has been raised with Apple - Issue ID FB15368430

Based on that issue, here are other affected tests:

• java/nio/channels/DatagramChannel/AdaptorMulticasting.java
• java/nio/channels/DatagramChannel/Promiscuous.java

Grinder with all three tests:

• aarch64 jdk11 (No AdapterMulticasting as it's not in 11) - Failed 1/200: Promiscous had Exec format error. BasicMulticast failed but for another reason (2/1/197)
• aarch64 jdk21 Both failed 1/200 but not with Exec format error (1/1/198)
• x64 jdk21 Stuck - aborted (2/1/135)
• x64 jdk21 AdapterMulticasting passed all runs. Other two failed 1/200 but not with Exec format error (11/1/188)
• aarch64 jdk17 Both failed 1/200 but not with Exec format error (0/1/199)
• aarch64 jdk24 All three tests failed 195/200 times. (5/195/0)

Numbers in brackets are from curl -s https://ci.adoptium.net/job/Grinder/11840/consoleText | egrep '_PASSED|_FAILED' | grep -v echo | cut -d' ' -f2 | uniq -c and shows when it's passing and failing in the run, so e.g. 11/1/188 means that the first 11 passed, one failed, then 188 passed. Apart frmo teh jdk24 one, most seem to have had a single failure towards the start of the run.

[sxa]

Noting that due to other problems - most recently adoptium/temurin-build#4058 - we haven't had a jdk11u/mac/aarch64 build for a while and so there is a lack of reliable recent history of this execution (I couldn't easily track down anything in TRSS for it)

[sxa]

Based on grep -ir 'exec format error' Test*mac* on the jenkins server, this has been seen in:

• Test_openjdk17_hs_extended.openjdk_aarch64_mac_testList_1#28
• Test_openjdk21_hs_extended.openjdk_aarch64_mac_testList_1#25
• Test_openjdk23_hs_extended.openjdk_x86-64_mac_testList_3#12
• Test_openjdk23_hs_extended.openjdk_x86-64_mac_testList_4#7
• Test_openjdk23_hs_extended.openjdk_x86-64_mac_testList_5#2

All of them have java.net.SocketException: Resource busy (setsockopt failed) for BasicMulticast Test and java.net.SocketException: Exec format error (setsockopt failed) in the AdapterMulticasting one. The exception is for the last two (note - only 2/3 of the jdk23 ones) where BasicMulticast passed ok.