Don't trace DB errors when we're retrying

If we know we're retrying on DB failures, let's be smarter and not trace
an exception at ERROR level while looping, since this clutters up the
logs and causes confusion when trying to track down failures.

DBDeadlock errors show up quite a bit in q-svc and n-api logs in gate
runs but when you check logstash for those, they are primarily in jobs
that are successful, so the trace is on the first try in the loop and
then we pass on the second try - but the trace itself is confusing while
debugging gate failures.

So let's just be smarter and log at debug level between retries, and if
we hit an unexpected exception, log that error (as before), and raise it
up to the caller.

Closes-Bug: #1523990

Change-Id: I15b4a9b5c7ec9bfede9ec9989de02c1da46eac81

oslo_db/api.py

@@ -139,6 +139,11 @@ class wrap_db_retry(object):
                     with excutils.save_and_reraise_exception() as ectxt:
                         if remaining > 0:
                             ectxt.reraise = not self._is_exception_expected(e)
+                            if ectxt.reraise:
+                                # We got an unexpected exception so stop
+                                # retrying, log it and raise it up to the
+                                # caller.
+                                LOG.exception(_LE('DB error.'))
                         else:
                             LOG.exception(_LE('DB exceeded retry limit.'))
                             # if it's a RetryRequest, we need to unpack it
@@ -166,7 +171,7 @@ class wrap_db_retry(object):
             # and not an error condition in case retries are
             # not exceeded
             if not isinstance(exc, exception.RetryRequest):
-                LOG.exception(_LE('DB error.'))
+                LOG.debug('DB error: %s', exc)
             return True
         return self.exception_checker(exc)