Deal with the fact that the AWS API is eventually consistent.

Sometimes we get an instance id back from the creation call but then
get an exception saying the ID was not found when we do a get_all_instances.

util/vpc-tools/abbey.py

@@ -595,7 +595,16 @@ def launch_and_configure(ec2_args):
         "Waiting for instance {} to reach running status:".format(instance_id)),
     status_start = time.time()
     for _ in xrange(EC2_RUN_TIMEOUT):
-        res = ec2.get_all_instances(instance_ids=[instance_id])
+        try:
+            res = ec2.get_all_instances(instance_ids=[instance_id])
+        except EC2ResponseError as e:
+            if e.code == "InvalidInstanceID.NotFound":
+                print("Instance not found({}), will try again.".format(
+                    instance_id))
+                time.sleep(1)
+                continue
+            else:
+                raise(e)
         if res[0].instances[0].state == 'running':
             status_delta = time.time() - status_start
             run_summary.append(('EC2 Launch', status_delta))