Every Kubernetes cluster has a built-in DNS add-on (Kube-DNS or CoreDNS) to provide domain name resolution for workloads in the cluster. When handling a high concurrency of DNS queries, Kube-DNS/CoreDNS may encounter a performance bottleneck, that is, it may fail occasionally to fulfill DNS queries. There are cases when Kubernetes workloads initiate unnecessary DNS queries. This makes DNS overloaded if there are many concurrent DNS queries. Tuning DNS configuration for workloads will reduce the risks of DNS query failures to some extent.
For more information about DNS, see CoreDNS.
nameserver 10.247.x.x search default.svc.cluster.local svc.cluster.local cluster.local options ndots:5
The value ndots:5 means that if a domain name has fewer than 5 dots (.), DNS queries will be attempted by combining the domain name with each domain in the search list in turn. If no match is found after all the domains in the search list are tried, the domain name is then used for DNS query. If the domain name has 5 or more than 5 dots, it will be tried first for DNS query. In case that the domain name cannot be resolved, DNS queries will be attempted by combining the domain name with each domain in the search list in turn.
For example, the domain name www.***.com has only two dots (smaller than the value of ndots), and therefore the sequence of DNS queries is as follows: www.***.com.default.svc.cluster.local, www.***.com.svc.cluster.local, www.***.com.cluster.local, and www.***.com. This means that at least seven DNS queries will be initiated before the domain name is resolved into an IP address. It is clear that when many unnecessary DNS queries will be initiated to access an external domain name. There is room for improvement in workload's DNS configuration.
For more information about configuration options in the resolver configuration file used by Linux operating systems, visit http://man7.org/linux/man-pages/man5/resolv.conf.5.html.
Kubernetes provides DNS-related configuration options for applications. The use of application's DNS configuration can effectively reduce unnecessary DNS queries in certain scenarios and improve service concurrency. The following procedure uses an Nginx application as an example to describe how to add DNS configurations for a workload on the console.
apiVersion: apps/v1 kind: Deployment metadata: name: nginx namespace: default spec: replicas: 1 selector: matchLabels: app: nginx template: metadata: labels: app: nginx spec: containers: - name: container-1 image: nginx:latest imagePullPolicy: IfNotPresent imagePullSecrets: - name: default-secret dnsPolicy: None dnsConfig: options: - name: ndots value: '5' - name: timeout value: '3' nameservers: - 10.2.3.4 searches: - my.dns.search.suffix
Parameter |
Description |
---|---|
ClusterFirst (default value) |
Custom DNS configuration added to the default DNS configuration. By default, the application connects to CoreDNS (CoreDNS of the CCE cluster connects to the DNS on the cloud by default). The custom dnsConfig will be added to the default DNS parameters. Containers can resolve both the cluster-internal domain names registered by a Service and the external domain names exposed to public networks. The search list (search option) and ndots: 5 are present in the DNS configuration file. Therefore, when accessing an external domain name and a long cluster-internal domain name (for example, kubernetes.default.svc.cluster.local), the search list will usually be traversed first, resulting in at least six invalid DNS queries. The issue of invalid DNS queries disappears only when a short cluster-internal domain name (for example, kubernetes) is being accessed. |
ClusterFirstWithHostNet |
By default, the applications configured with the host network are interconnected with the DNS configuration of the node where the pod is located. The DNS configuration is specified in the DNS file that the kubelet --resolv-conf parameter points to. In this case, the CCE cluster uses the DNS on the cloud. If workloads need to use Kube-DNS/CoreDNS of the cluster, set dnsPolicy to ClusterFirstWithHostNet and container's DNS configuration file is the same as ClusterFirst, in which invalid DNS queries still exist.
... spec: containers: - image: nginx:latest imagePullPolicy: IfNotPresent name: container-1 restartPolicy: Always hostNetwork: true dnsPolicy: ClusterFirstWithHostNet |
Default |
The DNS configuration of the node where the pod is located is inherited, and the custom DNS configuration is added to the inherited configuration. Container's DNS configuration file is the DNS configuration file that the kubelet's --resolv-conf flag points to. In this case, a cloud DNS is used for CCE clusters. Both search and options fields are left unspecified. This configuration can only resolve the external domain names registered with the Internet, and not cluster-internal domain names. This configuration is free from the issue of invalid DNS queries. |
None |
The default DNS configuration is replaced by the custom DNS configuration, and only the custom DNS configuration is used. If dnsPolicy is set to None, the dnsConfig field must be specified because all DNS settings are supposed to be provided using the dnsConfig field. |
If the dnsPolicy field is not specified, the default value is ClusterFirst instead of Default.
Parameter |
Description |
---|---|
options |
An optional list of objects where each object may have a name property (required) and a value property (optional). The contents in this property will be merged to the options generated from the specified DNS policy in dnsPolicy. Duplicate entries are removed. |
nameservers |
A list of IP addresses that will be used as DNS servers. If workload's dnsPolicy is set to None, the list must contain at least one IP address, otherwise this property is optional. The servers listed will be combined to the nameservers generated from the specified DNS policy in dnsPolicy with duplicate addresses removed. |
searches |
A list of DNS search domains for hostname lookup in the pod. This property is optional. When specified, the provided list will be merged into the search domain names generated from the chosen DNS policy in dnsPolicy. Duplicate domain names are removed. Kubernetes allows for at most 6 search domains. |
The following example describes how to configure DNS for workloads.
Scenario
Kubernetes in-cluster Kube-DNS/CoreDNS applies to resolving only cluster-internal domain names or cluster-internal domain names + external domain names. This is the default DNS for workloads.
Example:
apiVersion: v1 kind: Pod metadata: namespace: default name: dns-example spec: containers: - name: test image: nginx:alpine dnsPolicy: ClusterFirst imagePullSecrets: - name: default-secret
Container's DNS configuration file:
nameserver 10.247.3.10 search default.svc.cluster.local svc.cluster.local cluster.local options ndots:5
Scenario
A DNS cannot resolve cluster-internal domain names and therefore applies to the scenario where workloads access only external domain names registered with the Internet.
Example:
apiVersion: v1 kind: Pod metadata: namespace: default name: dns-example spec: containers: - name: test image: nginx:alpine dnsPolicy: Default # The DNS configuration file that the kubelet --resolv-conf parameter points to is used. In this case, the CCE cluster uses the DNS on the cloud. imagePullSecrets: - name: default-secret
Container's DNS configuration file:
nameserver 100.125.x.x
Scenario
By default, a DNS is used for workloads running with hostNetwork. If workloads need to use Kube-DNS/CoreDNS, set dnsPolicy to ClusterFirstWithHostNet.
Example:
apiVersion: v1 kind: Pod metadata: name: nginx spec: hostNetwork: true dnsPolicy: ClusterFirstWithHostNet containers: - name: nginx image: nginx:alpine ports: - containerPort: 80 imagePullSecrets: - name: default-secret
Container's DNS configuration file:
nameserver 10.247.3.10 search default.svc.cluster.local svc.cluster.local cluster.local options ndots:5
Scenario
You can flexibly customize the DNS configuration file for applications. Using dnsPolicy and dnsConfig together can address almost all scenarios, including the scenarios in which an on-premises DNS will be used, multiple DNSs will be cascaded, and DNS configuration options will be modified.
Example 1: Using Your On-Premises DNS
Set dnsPolicy to None so application's DNS configuration file is generated based on dnsConfig.
apiVersion: v1 kind: Pod metadata: namespace: default name: dns-example spec: containers: - name: test image: nginx:alpine dnsPolicy: "None" dnsConfig: nameservers: - 10.2.3.4 # IP address of your on-premises DNS searches: - ns1.svc.cluster.local - my.dns.search.suffix options: - name: ndots value: "2" - name: timeout value: "3" imagePullSecrets: - name: default-secret
Container's DNS configuration file:
nameserver 10.2.3.4 search ns1.svc.cluster.local my.dns.search.suffix options timeout:3 ndots:2
Example 2: Modifying the ndots Option in the DNS Configuration File to Reduce Invalid DNS Queries
Set dnsPolicy to a value other than None so the DNS parameters configured in dnsConfig are added to the DNS configuration file generated based on dnsPolicy.
apiVersion: v1 kind: Pod metadata: namespace: default name: dns-example spec: containers: - name: test image: nginx:alpine dnsPolicy: "ClusterFirst" dnsConfig: options: - name: ndots value: "2" # The ndots:5 option in the DNS configuration file generated based on the ClusterFirst policy is changed to ndots:2. imagePullSecrets: - name: default-secret
Container's DNS configuration file:
nameserver 10.247.3.10 search default.svc.cluster.local svc.cluster.local cluster.local options ndots:2
Example 3: Using Multiple DNSs in Serial Sequence
apiVersion: v1 kind: Pod metadata: namespace: default name: dns-example spec: containers: - name: test image: nginx:alpine dnsPolicy: ClusterFirst # Added DNS configuration. The cluster connects to CoreDNS by default. dnsConfig: nameservers: - 10.2.3.4 # IP address of your on-premises DNS imagePullSecrets: - name: default-secret
Container's DNS configuration file:
nameserver 10.247.3.10 10.2.3.4 search default.svc.cluster.local svc.cluster.local cluster.local options ndots:5