[CI - FAILURE]: / AppData_WithSendArgs.Send_134 (msquictest)

[github-actions]

https://github.com/microsoft/msquic/actions/runs/33526469631

[guhetier]

AI Generated analysis

This failure is not an AppData/WithSendArgs.Send/134 protocol failure: the selected test never executed. In BVT run 33526469631, job 99921014502, global test setup failed while generating the self-signed server certificate:

• The decoded trace reports EVP_PKEY_keygen failed.
• GoogleTest reports the global setup assertion, then marks Send/134 skipped.
• Teardown still runs and calls CxPlatFreeSelfSignedCert with ClientCertParams = 0xbebebebebebebebe, the ASan uninitialized-allocation pattern.
• The resulting invalid free causes the ASan abort/exit 134. The isolated test runner then attributes the missing XML result to the selected test, producing the misleading issue title.

The causal chain is: OpenSSL key generation fails -> QuicTestEnvironment::SetUp() returns before initializing ClientCertParams -> unconditional teardown frees the uninitialized pointer -> ASan aborts -> the runner synthesizes an Application Crashed result for Send/134.

The root cause of the crash is unsafe partial-initialization cleanup in QuicTestEnvironment (src/test/bin/quic_gtest.cpp). The initiating EVP_PKEY_keygen failure remains unexplained because src/platform/selfsign_openssl.c logs only a generic message and does not retain the OpenSSL error stack.

Alternative explanations were eliminated: no connection or stream events occurred, there was no worker-overload signature (0x800704c9 or elevated queue delay), and this Linux/OpenSSL failure does not match the Windows CryptoAPI watchdog pattern. The exact matrix job passed in the preceding nine sampled runs, so this is intermittent rather than a deterministic regression.

Confidence: High for the invalid teardown root cause; inconclusive for the one-off OpenSSL key-generation failure.