This is the result when logging at
trace level, and then parsing the result with
rg "(TEST-UNEX|BrokenPipe|TEST-START|TEST-OK|TEST-SKIP)" live_backing.log
It looks like there is somewhat a correlation between those
BrokenPipe errors and the test failing, but maybe it's a reverse correlation: when the test fails, there is a
BrokenPipe. Also sometimes, there are lots of broken pipe errors, but no test failure (by the end).
Matthew, do you see the same on Windows? Do you know whether it's expected to have those
BrokenPipe issues? From reading the code, this happens in the server when a
call to the client errored out, probably because the client end of the channel has closed it .
Next, I'm going to try to understand what is causing the client to close its end.