[tests] fix hybrid_clock-test
I noticed HybridClockTest.AutoTimeSourceNoDedicatedNtpServer was failing
from time to time on AWS instances. It turned out the scenario missed
setting of the fake DNS server, but the comments claimed it did.
This patch addresses the issue.
Change-Id: I1386646d592eb0d127a6a216664c6f241c316c6d
Reviewed-on: http://gerrit.cloudera.org:8080/18928
Reviewed-by: Attila Bukor <abukor@apache.org>
Tested-by: Alexey Serbin <alexey@apache.org>
diff --git a/src/kudu/clock/hybrid_clock-test.cc b/src/kudu/clock/hybrid_clock-test.cc
index c3e8996..80588a7 100644
--- a/src/kudu/clock/hybrid_clock-test.cc
+++ b/src/kudu/clock/hybrid_clock-test.cc
@@ -57,8 +57,10 @@
DECLARE_bool(inject_unsync_time_errors);
DECLARE_string(builtin_ntp_servers);
+DECLARE_string(cloud_curl_dns_servers_for_testing);
DECLARE_string(time_source);
DECLARE_uint32(cloud_metadata_server_request_timeout_ms);
+
METRIC_DECLARE_entity(server);
using kudu::cloud::InstanceDetector;
@@ -514,6 +516,10 @@
// Set very short interval for the instance detection timeout.
FLAGS_cloud_metadata_server_request_timeout_ms = 1;
+ // Configure a bad DNS server to ensure a timeout,
+ // even when run on fast cloud instances.
+ FLAGS_cloud_curl_dns_servers_for_testing = "192.0.2.0";
+
{
// Check for the pre-condition: the auto-detection fails as expected due
// to the super-short timeout setting for the cloud metadata requests and
