Since the TLS tests were added a few people have reported seeing hangs in some of the TLS test cases for migration. Debugging has revealed that in all cases the test was waiting for a STOP event that never arrived.
The problem is that TLS performance is highly dependant on the crypto impl. Some people have been running tests on machines which are highly efficient at running the guest dirtying workload but relatively slow at TLS. This has prevented convergance from being reliably achieved in the configured max downtime. Since this test design has been long standing I suspect the lack of convergance is a likely cause of previous hangs we've seen in various scenarios that resulted in us disabling the test on s390 TCG, ppc TCG and ppc KVM-PR. Thus I have suggested we drop this skip conditions, though I would note that I've not had the ability to actually test the effect that this has. Daniel P. Berrangé (5): tests: wait max 120 seconds for migration test status changes tests: wait for migration completion before looking for STOP event tests: increase migration test converge downtime to 30 seconds tests: use consistent bandwidth/downtime limits in migration tests tests: stop skipping migration test on s390x/ppc64 tests/qtest/migration-helpers.c | 14 ++++++ tests/qtest/migration-test.c | 80 ++++++++++----------------------- 2 files changed, 38 insertions(+), 56 deletions(-) -- 2.36.1
