The test is failing because the server maintains an internal buffer of size 1024, and it seems that 1103 log lines were logged after the local read occurred, so we fail to find the local read log. We can decrease some of the logging that's happening, and that should help avoid the issue, following the example from SERVER-58166. We can also retry parts of the test that fail if we detect the buffer may have rotated past lines we are looking for. 

An alternative is to make sure the logs we're interested in are not rotated off the buffer by adding a special workaround to never throw away logs matching some specified regex. I think this would be overly complicated to implement.