Also try to log tail of PG logs on CI in a brave hope of catching troublesthere. Currently travis job on 10 rarely fails at test_concurrent_updates, butwe've failed to reproduce the issue locally. Note that test_concurrent_updatesactually doesn't perform anything useful as testgres nodes aren't pickable dueto weird logging implemented in separate thread, but nobody cared to checkapply_async result so this has went unnoticed for a long time.