From d04cf6b2ae12596a481c37b8d25f652f099ae5cd Mon Sep 17 00:00:00 2001 From: Steve Kuznetsov Date: Tue, 13 Jun 2017 09:57:00 -0700 Subject: Disable negative caching, set cache TTL to 1s When running headless services as pods on OpenShift, no ClusterIP is assigned. In these cases, peer pods are relying on DNS to locate the service endpoints. When a pod is deleted or another trigger causes the endpoint to change, the OpenShift DNS is updated immediately. However, dnsmasq has a default TTL of 30s, so the wrong response is returned on name resolution. Removing negative caching and turning the TTL to a very short 1s should resolve the issue. Signed-off-by: Steve Kuznetsov --- roles/openshift_node_dnsmasq/templates/origin-dns.conf.j2 | 2 ++ 1 file changed, 2 insertions(+) (limited to 'roles/openshift_node_dnsmasq') diff --git a/roles/openshift_node_dnsmasq/templates/origin-dns.conf.j2 b/roles/openshift_node_dnsmasq/templates/origin-dns.conf.j2 index f397cbbf1..8bae9aaac 100644 --- a/roles/openshift_node_dnsmasq/templates/origin-dns.conf.j2 +++ b/roles/openshift_node_dnsmasq/templates/origin-dns.conf.j2 @@ -1,3 +1,5 @@ no-resolv domain-needed server=/{{ openshift.common.dns_domain }}/{{ openshift.common.kube_svc_ip }} +no-negcache +max-cache-ttl=1 -- cgit v1.2.1