We have a distributed environment with load balancing to 9 indexers. At 12am, our logs stop line breaking and start group multiple entries into one event and random log lines (no timestamps) for the next hour. Then it miraculously starts to work again. Never seen such a thing, thoughts?
The problem is that your timestamp is using <date><SingleSpace><SINGLE digit hour=0> for the midnight hour but it is using <DATE><SingleSpace><DOUBLE digit hour> the rest of the time. So you need to make sure that your TIME_FORMAT, MAX_TIMESTAMP_LOOKAHEAD, LINE_BREAKER, etc. can each accommodate this variability. One of these is poorly-written enough to fail on the zero-hour because that format is different than all of the others.