KubeCon + CloudNativeCon North America 2018

KubeCon Seattle 2018 Understanding CoreDNS in Kubernetes

1. Understanding CoreDNS in Kubernetes Public link to this document
2. Speakers • Cricket Liu ○ EVP Engineering and Chief DNS Architect, Infoblox ○ Co-author of DNS and BIND, 5th Edition ○ Co-author of all of O'Reilly Media’s books on DNS • François Tur ○ Senior Software Manager at Infoblox ○ CoreDNS maintainer, @fturib • John Belamaric ○ Senior Staff Software Engineer, Google ○ CoreDNS maintainer, @johnbelamaric
3. Agenda • Introduction • Cluster DNS Default Configuration • Outside the Defaults • More Resources
4. Introduction • Flexible DNS server written in Go • Plugin-based architecture, easily extended ○ To support different cloud-native stacks, for example • Supports DNS, DNS over TLS (DoT), DNS over gRPC • Started and led by Miek Gieben, author of SkyDNS and SkyDNS2 • Originally a fork of the Caddy HTTP server (“Caddy DNS”)
5. CoreDNS • Native support of service discovery for Kubernetes • Generally available with Kubernetes 1.11 • Now the default in 1.13 • Integration with etcd and cloud vendors (e.g., AWS’s Route 53) • Support for Prometheus metrics • Forwarding to recursive DNS server
6. Why CoreDNS (vs. kube-dns)? ● Easily extensible plugin architecture ● Rich set of (~34) plugins, with new ones being developed all the time ● Simpler, with fewer moving parts (single executable and process) ○ And all written in Go ● Customizable DNS entries in and out of the cluster domain ● Experimental server-side search path to reduce query volume
7. Project Status • Version 1.2.6 (released 11/5/2018) • Incubating project in CNCF • Graduation vote underway • Growing community • 112 contributors (big thanks!) • 16 maintainers • 29+ public adopters • 3000+ stars
8. CoreDNS as Cluster DNS • CoreDNS Kubernetes Resources • Default Corefile • Resolving a Query • Stub Domains • Cache Tuning forward upstream
9. Kubernetes Resources kube-dns -n=kube-system coredns -n=kube-system data: Corefile: .:53 { … } spec: replicas: 2 Template: metadata labels: k8s-app: kube-dns spec: containers: image: k8s.gcr.io/coredns:1.2.6 resources: limits: memory: 170Mi requests: cpu: 100m / memory: 70Mi … livenessProbe: http://8080 dnsPolicy: Default spec: selector: k8s-app: kube-dns clusterIP: x.x.x.10 Ports: … 53 - UDP/TCP 9153 - TCP
10. Default Corefile Enable error logging Serve liveness status on http 8080 Backend to k8s for cluster.local and reverse domains Mimic kube-dns pod records behavior Resolve CNAME targets upstream Continue searching reverse zones .:53 { errors health kubernetes cluster.local in-addr.arpa ip6.arpa { pods insecure upstream fallthrough in-addr.arpa ip6.arpa } prometheus :9153 proxy . /etc/resolv.conf cache 30 Serve prometheus metrics Forward other domains to /etc/resolv.conf ns Cache for up to 30 seconds Reload server if the Corefile change loop reload Shuffle order of returned records loadbalance DNS protocol loop check }
11. Resolving a Query organization.com:53 { errors cache 3600 proxy . } onprem.organization.com:53 { errors cache 300 proxy . } .:53 { errors health kubernetes cluster.local ... { ... } proxy mycompany.com proxy . /etc/resolv.com } www.organization.com mydb.onprem.organization.com www.mycompany.com theservice.default.svc.cluster.local www.google.com ? ? ? ? ? tcp/udp :53 organization.com . onprem.organization.com errors errors errors cache 3600 health cache 300 proxy . prometheus proxy kubernetes cluster.local proxy mycompany.com proxy . /etc/resolv.conf
12. Stub Domains corporate.com :53 { proxy . <DNS-IP-corporate> } local.corporate.com :53 { proxy . <DNS-IP-local-corporate> } .:53 { ... kubernetes cluster.local ... { ... } ... proxy . /etc/resolv.conf ... cache 30 } proxy ● Forward some out-of-cluster domains directly to the right authoritative DNS server ● Handle internal corporate domains
13. Cache Tuning corporate.com { errors log innerservices.corporate.com proxy local.corporate.com <DNS-IP-local> proxy . <DNS-IP-corporate> cache 3600 } .:53 { ... kubernetes cluster.local ... { ... } proxy . /etc/resolv.conf ... cache 30 } log cache proxy ● Allow specific configuration for known zone. cache, logging ... When using plugin, check description and options at https://coredns.io/plugins/
14. Outside the Defaults ● Kubernetes Plugin Options ● Kubernetes-related Plugins ○ Autopath, external, kubernetai, and federation ● Adding Static Records ● Query Rewrites
15. Common Options for Cluster DNS pods [disabled insecure verified] When receiving query for pod : <IP-like>.<ns>.pod.<cluster.local> disabled - unknown domain insecure - reply with the IP provided verified - reply only if IP is a pod IP Watches all pods. Costly in memory & api server load endpoint_pod_names Use the pod name for endpoint address name, if the hostname is empty ttl TTL Change the TTL of kubernetes records fallthrough [ZONES] Second chance option : try with the next plugin in chain for resolving ‘unknown’ domains.
16. All Kubernetes Plugin Options kubernetes [ZONES...] { Connection to k8s API resyncperiod DURATION # API Server resync DURATION period endpoint URL [URL...] # API Server URL tls CERT KEY CACERT # API Server connection TLS config. kubeconfig KUBECONFIG CONTEXT # use kubeconfig context Filtering/Exposing information namespaces NAMESPACE... # exposed ns labels EXPRESSION # selector for exposed services ignore_empty_service # do not expose service which no healthy endpoints noendpoints # don't watch endpoints resource - therefore ignore Headless services upstream [ADDRESS...] # resolve External services and define specific nameserver Tune domain naming pods POD-MODE # pods mode: disabled,insecure,verified endpoint_pod_names # allow using pod names to identify Headless service domain ttl TTL # TTL default: 5s fallthrough [ZONES...] # no record found in the domain, second chance with next plugin DNS workflow transfer to ADDRESS… # allow transfer to secondary DNS server }
17. Plugins Complete list of plugins available at coredns.io Complete list of external plugins Configuration Middleware Backends External metadata loadbalance template kubernetai tls cache host redisc reload rewrite route53 dnssec kubernetes autopath file bind health auto prometheus etcd errors log loop forward proxy ● Only most commonly used are mentioned ● They are shown in order of chaining in the default image (top to bottom, left to right)
18. Other Kubernetes-related Plugins ● Autopath ○ Server-side search path resolution ● K8s External ● Kubernetai ● Federation
19. Autopath - the problem ● Kubernetes has a long DNS search path and ndots value ○ ○ ○ ○ <namespace>.svc.cluster.local svc.cluster.local cluster.local plus the nodes search path ● Enables flexible use of names, but leads to extra queries dnstools# host -v google.com Trying "google.com.default.svc.cluster.local" Trying "google.com.svc.cluster.local" Trying "google.com.cluster.local" Trying "google.com" ;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 62752 ;; flags: qr rd ra; QUERY: 1, ANSWER: 1, AUTHORITY: 4, ADDITIONAL: 4 ...
20. Autopath - the solution ● kubernetes pods verified + autopath ● Since CoreDNS knows the namespace of the source pod IP, it knows the search path ● Execute the search path server-side dnstools# host -v google.com Trying "google.com.default.svc.cluster.local" ;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 38177 ;; flags: qr rd ra; QUERY: 1, ANSWER: 2, AUTHORITY: 0, ADDITIONAL: 0 ;; QUESTION SECTION: ;google.com.default.svc.cluster.local. IN A ;; ANSWER SECTION: google.com.default.svc.cluster.local. 13 IN CNAME google.com. google.com. 13 IN A ...
21. Autopath - results So, why not default?? ● ● Requires pods verified - too much memory, too much API server load Edge case can result in a pod getting the wrong response from the cache ○ Mitigated by enabling only negative cache…?
22. Sneak Peek: Kubernetes External ● ● ● ● Coming in version 1.3.0 Service ExternalIPs and LoadBalancer IPs published in another zone Add to cluster DNS or (preferred) run a separate CoreDNS For example, external zone configured as “apps.example.com”: apiVersion: v1 kind: Service metadata: name: foo namespace: bar spec: type: LoadBalancer ... status: loadBalancer: ingress: - ip: dnstools# host foo.bar.apps.example.com foo.bar.apps.example.com has address
23. Kubernetai . { } errors log kubernetai endpoint } kubernetai endpoint } kubernetai endpoint } ● Single CoreDNS serves Kubernetes service names for multiple clusters cluster.local { assemblage.local { conglomeration.local { ● Remember cluster IPs are not routable ● Only really useful for headless services with routable pod IPs ● ...or for some Istio magic
24. A Couple Interesting Use Cases ● Add Static Data ● Query Rewriting
25. Add Static Records myservices.com { file /etc/coredns/myservice.db myservices.com cache 3600 } .:53 { ... hosts svc.cluster.local { undeployed.default.svc.cluster.local fallthrough } kubernetes cluster.local ... { ... fallthrough svc.cluster.local } auto cluster.local { directory /etc/coredns/cluster.local } proxy . /etc/resolv.conf ... } hosts file auto ● ● ● ● Host non-cluster zones Override IP for specific name Fallback for missing services not yet migrated Populate subdomains other than svc and pod (a little dangerous)
26. Query Rewriting .:53 { errors health rewrite { name regex (.*)\.demo\.com\.$ {1}.default.svc.cluster.local answer name (.*)\.default\.svc\.cluster\.local\.$ {1}.demo.com } } kubernetes cluster.local ... { ... } ... proxy . /etc/resolv.conf ... rewrite ● During migration of services, translate old naming into the Kubernetes DNS scheme. ● Use the same server cert without adding cluster name
27. Resources DNS in Kubernetes Documentation Customizing DNS Service Using CoreDNS for Discovery Service Debugging DNS resolution Kubernetes Blog CoreDNS GA for Kubernetes Cluster DNS CoreDNS Blogs on the Kubernetes Plugin Cluster DNS : CoreDNS versus kube-dns Scaling CoreDNS in a Kubernetes cluster Migration from kube-dns to CoreDNS Deploying CoreDNS with kubeadm How queries are process in CoreDNS CoreDNS for Kubernetes Service discovery Github Resources CoreDNS github Manual deployment of CoreDNS in kubernetes Kubernetes / DNS github
28. Community and Support Thank you! Issues/Questions/Support github: http://github.com/coredns/coredns (also kubernetes/dns) slack: https://slack.cncf.io #coredns security related: security@coredns.io Documentation/Resources http://coredns.io - plugin docs. blogs.
29. Backup Slides
30. Kubernetes DNS Schema ClusterIP Service <service>.<ns>.svc.<zone>. <ttl> IN A <cluster-ip> _<port>._<proto>.<service>.<ns>.svc.<zone>. <ttl> IN SRV <weight> <priority> <port-number> <service>.<ns>.svc.<zone>. <d>.<c>.<b>.<a>.in-addr.arpa. <ttl> IN PTR <service>.<ns>.svc.<zone>. Headless Service <service>.<ns>.svc.<zone>. <ttl> IN A <endpoint-ip> <hostname>.<service>.<ns>.svc.<zone>. <ttl> IN A <endpoint-ip> _<port>._<proto>.<service>.<ns>.svc.<zone>. <ttl> IN SRV <weight> <priority> <port-number> <hostname>.<service>.<ns>.svc.<zone>. <d>.<c>.<b>.<a>.in-addr.arpa. <ttl> IN PTR <hostname>.<service>.<ns>.svc.<zone>. External Service <service>.<ns>.svc.<zone>. <ttl> IN CNAME <extname>. Pod (deprecated) <a>-<b>-<c>-<d>.<ns>.pod.<zone>. <ttl> IN A <a>.<b>.<c>.<d> DNS Version dns-version.<zone>. <ttl> IN TXT <schema-version>
31. CoreDNS - kubernetes plugin API connection options resyncperiod DURATION API Server resync duration (default is 5mn) endpoint URL [URL...] Define the url to connect to the API. If several defined, plugin will use one that is healthy if not define, use the cluster service account available on the pod tls CERT KEY CACERT TLS certificate for the connection to API (requires endpoint) kubeconfig KUBECONFIG CONTEXT authenticates the connection to a remote k8s cluster using a kubeconfig file. (requires endpoint and ignore tls)
32. CoreDNS - kubernetes plugin Filtering k8s domains namespaces NAMESPACE... Only the namespaces indicated will be exposed. All domains in othe domains will be replied “unknown” labels SELECTOR-EXPRESSION Only the objects (pod, service, endpoint) matching the selector will be exposed ignore_empty_service consider “unknwon” any service that have no ready pod instead of returning a valid record with no data noendpoints Do not watch endpoints. Any domains that includes a endpoints will be considered “unknown”. All headless service will be “unknown” upstream [ADDRESS...] If not present, will not resolve the external services. These filters can be combined. To be used in conjunction with “fallthrough” and with another way to resolve the unexposed elements
33. Kubernetes DNS Schema <service>.<ns>.svc.<czone> kubernetes.default.svc.cluster.local kube-dns.kube-system.svc.cluster.local *.default.svc.cluster.local <service>.<ns>.svc.<czone> _<port>._<proto>.<service>.<ns>.svc.<czone> A AAAA PTR If ClusterIP service => the Cluster IP If Headless service => all the sub-domains as <ep>.<service>.... resolved. SRV => all the port / proto supported by the service. A AAAA PTR Only for Headless Services => the IP of the corresponding endpoint A AAAA Change the TTL of kubernetes records _https._tcp.kubernetes.default.svc.cluster.local <ep>.<service>.<ns>.svc.<czone> <ep> is the hostname, if not and if endpoint_pod_names the name of the pod, if not the <ip-like> of the endpoint <ip-like>.<ns>.pod.<czone> 10-96-0-65.kube-system.pod.cluster.local
34. Resolving a Query 1- A www.organization.com 2- A mydb.onprem.organization.com 3- A www.mycompany.com 4- A myservice.default.svc.cluster.local tcp/udp :53 organization.com . onprem.organization.com errors errors errors cache 3600 health cache 300 prometheus proxy . proxy kubernetes cluster.local proxy mycompany.com proxy . /etc/resolv.conf kubernetes ... { …. fallthrough <zone> }
35. Resolving a Query 1- A www.organization.com 2- A mydb.onprem.organization.com 3- A www.mycompany.com 4- A myservice.default.svc.cluster.local tcp/udp :53 organization.com onprem.organization.com errors errors errors cache 3600 health cache 300 proxy . prometheus proxy kubernetes cluster.local . proxy mycompany.com proxy . /etc/resolv.conf kubernetes ... { …. fallthrough <zone> }