Last time I had the problem, I formed a hypothesis, which may be unfounded: Possibly it is some event that my test provoked (a server crash or similar) that triggers an update from within the Chrome browser itself and not the code in the update folder (as part of error handling). Does this make sense. did your test fail in ways that may have triggered error handling in the browser?
I do not currently have the problems you mention, so it is not some 'global issue'. I had to downgrade to 43 on all my servers within a the span of 2-3 days a week ago, but it has been OK since. (And I am about to move to 45 now).