Raise first error that occurred during coiteration

The reactor will by default log exceptions that occur
in the coiterator, resulting in nothing being raised.
Code using the asyncdns lib except certain errors
to occur when something is wrong so this should keep
the current asyncdns interfacing intact, at the cost of
taking longer to fail

python/nav/asyncdns.py

@@ -78,11 +78,13 @@ def __init__(self):
         )
         self.results = defaultdict(list)
         self._finished = False
+        self._errors = []
 
     def resolve(self, names):
         """Resolves DNS names in parallel"""
-        self._finished = False
         self.results = defaultdict(list)
+        self._finished = False
+        self._errors = []
 
         def lookup_names():
             for name in names:
@@ -96,7 +98,10 @@ def lookup_names():
         coop = task.Cooperator()
         work = lookup_names()
         deferred_list = defer.DeferredList(
-            [coop.coiterate(work) for _ in range(BATCH_SIZE)]
+            [
+                coop.coiterate(work).addErrback(self._save_error)
+                for _ in range(BATCH_SIZE)
+            ]
         )
         deferred_list.addCallback(self._finish)
 
@@ -106,6 +111,10 @@ def lookup_names():
         # iteration to ensure the resolver library closes its UDP sockets
         reactor.iterate()
 
+        # raise first error if any occurred
+        for error in self._errors:
+            raise error
+
         return dict(self.results)
 
     def lookup(self, name):
@@ -128,6 +137,10 @@ def _errback(failure, host):
         """Errback"""
         return host, failure.value
 
+    def _save_error(self, failure):
+        """Errback for coiterator. Saves error so it can be raised later"""
+        self._errors.append(failure.value)
+
     def _finish(self, _):
         self._finished = True