While verifying the BuildTestEmbedded cancel/resume flake fix in

Reported (UTC): 2026-07-15 22:08

While verifying the BuildTestEmbedded cancel/resume flake fix in /code/workspace/BuildTestEmbedded-cancelflake, running 'scripts/test.bash --test tests/e2eCancelWhileRestartRecoveredRunIsAdmissionQueued.kts' printed 'Unexpected error in TCP communication: buildtest.embedded.BuildCancelledException' followed by a full java.lang.ClassNotFoundException: buildtest.embedded.BuildCancelledException stack trace rooted at ObjectInputStream.resolveClass and community.kotlin.kompile.testrunner.jvm.TcpCommunicationHandler.handleCommunication(JvmUnitTestLauncher.kt:333), yet the launcher then reported 'ALL TESTS PASSED (1/1 selected tests completed successfully)'. Impact: a child-JVM throwable that is absent from the parent launcher classpath cannot be deserialized, and the transport failure is not reflected in the test result, which can discard diagnostically useful failures or produce a false-green outcome. No workaround was applied because this is outside the cancel-resume matcher task; the sibling test's public assertions completed successfully. Durable fix: transport failures should use classpath-independent failure frames (or a safe fallback representation) and must fail the selected test when communication cannot be decoded, with an end-to-end regression using a test-owned exception type.

Production verification — 2026-07-17

Status: STILL EXISTS. Current production remains degraded rather than retired: buildtest.kotlin.build returned HTTP 200 but took 8.89 seconds, while the production host showed load average 60.59, CPU pressure near 90%, and the BuildTest JVM at about 1.8 GiB RSS.

This record was retained because its underlying mechanism remains observable or its durable fix is still open; historical incident details above remain useful reproduction evidence.