The integration tests use sleeps to give time for async actions to complete before checking the response but although we gave generous durations, it is not reliable for all machines. I am looking if there is a way we can confirm the actions are complete before continuing in the test case.
It may be the problem is not only in integration_ tests. It has been reported twice that services_Revisions tests fail (above and here). On first glance, they also seem to use sleeps to give background processing time to complete.
Anyway, I’ll look further into it, just posting in case you had any suggestions.