coredns/middleware/kubernetes

kubernetes

kubernetes enables reading zone data from a kubernetes cluster. Record names are constructed as "myservice.mynamespace.coredns.local" where:

• "myservice" is the name of the k8s service (this may include multiple DNS labels, such as "c1.myservice"),
• "mynamespace" is the k8s namespace for the service, and
• "coredns.local" is the zone configured for kubernetes.

The record name format can be changed by specifying a name template in the Corefile.

Syntax

kubernetes [zones...]

• zones zones kubernetes should be authorative for. Overlapping zones are ignored.

kubernetes [zones] {
    endpoint http://localhost:8080
}

• endpoint the kubernetes API endpoint, default to http://localhost:8080

Examples

This is the default kubernetes setup, with everything specified in full:

# Serve on port 53
.:53 {
    # use kubernetes middleware for domain "coredns.local"
    kubernetes coredns.local {
        # Kubernetes data API resync period
        # Example values: 60s, 5m, 1h
        resyncperiod 5m
        # Use url for k8s API endpoint
        endpoint http://localhost:8080
        # Assemble k8s record names with the template
        template {service}.{namespace}.{zone}
        # Only expose the k8s namespace "demo"
        namespaces demo
    }
    # Perform DNS response caching for the coredns.local zone
    # Cache timeout is provided by the integer in seconds
    #cache 180 coredns.local
}

Notes:

• If the namespaces keyword is omitted, all kubernetes namespaces are exposed.
• If the template keyword is omitted, the default template of "{service}.{namespace}.{zone}" is used.
• If the resyncperiod keyword is omitted, the default resync period is 5 minutes.

Basic Setup

Launch Kubernetes

Kubernetes is launched using the commands in the following run_k8s.sh script:

#!/bin/bash

# Based on instructions at: http://kubernetes.io/docs/getting-started-guides/docker/

#K8S_VERSION=$(curl -sS https://storage.googleapis.com/kubernetes-release/release/latest.txt)
K8S_VERSION="v1.2.4"

ARCH="amd64"

export K8S_VERSION
export ARCH

#DNS_ARGUMENTS="--cluster-dns=10.0.0.10 --cluster-domain=cluster.local"
DNS_ARGUMENTS=""

docker run -d \
    --volume=/:/rootfs:ro \
    --volume=/sys:/sys:ro \
    --volume=/var/lib/docker/:/var/lib/docker:rw \
    --volume=/var/lib/kubelet/:/var/lib/kubelet:rw \
    --volume=/var/run:/var/run:rw \
    --net=host \
    --pid=host \
    --privileged \
    gcr.io/google_containers/hyperkube-${ARCH}:${K8S_VERSION} \
    /hyperkube kubelet \
    --containerized \
    --hostname-override=127.0.0.1 \
    --api-servers=http://localhost:8080 \
    --config=/etc/kubernetes/manifests \
    ${DNS_ARGUMENTS} \
    --allow-privileged --v=2

Configure kubectl and test

The kubernetes control client can be downloaded from the generic URL: http://storage.googleapis.com/kubernetes-release/release/${K8S_VERSION}/bin/${GOOS}/${GOARCH}/${K8S_BINARY}

For example, the kubectl client for Linux can be downloaded using the command: curl -sSL "http://storage.googleapis.com/kubernetes-release/release/v1.2.4/bin/linux/amd64/kubectl"

The following setup_kubectl.sh script can be stored in the same directory as kubectl to setup kubectl to communicate with kubernetes running on the localhost:

#!/bin/bash

BASEDIR=`readlink -e $(dirname ${0})`

${BASEDIR}/kubectl config set-cluster test-doc --server=http://localhost:8080
${BASEDIR}/kubectl config set-context test-doc --cluster=test-doc
${BASEDIR}/kubectl config use-context test-doc

alias kubctl="${BASEDIR}/kubectl"

Verify that kubectl is working by querying for the kubernetes namespaces:

$ ./kubectl get namespaces
NAME      STATUS    AGE
default   Active    8d
test      Active    7d

Launch a kubernetes service and expose the service

The following commands will create a kubernetes namespace "demo", launch an nginx service in the namespace, and expose the service on port 80:

$ ./kubectl create namespace demo
$ ./kubectl get namespace

$ ./kubectl run mynginx --namespace=demo --image=nginx
$ ./kubectl get deployment --namespace=demo

$ ./kubectl expose deployment mynginx --namespace=demo --port=80
$ ./kubectl get service --namespace=demo

Launch CoreDNS

Build CoreDNS and launch using the configuration file in conf/k8sCorefile. This configuration file sets up CoreDNS to use the zone coredns.local for the kubernetes services.

The command to launch CoreDNS is:

$ ./coredns -conf conf/k8sCoreFile

In a separate terminal a dns query can be issued using dig:

$ dig @localhost mynginx.demo.coredns.local

; <<>> DiG 9.9.4-RedHat-9.9.4-29.el7_2.3 <<>> @localhost mynginx.demo.coredns.local
; (2 servers found)
;; global options: +cmd
;; Got answer:
;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 47614
;; flags: qr aa rd ra; QUERY: 1, ANSWER: 1, AUTHORITY: 0, ADDITIONAL: 1

;; OPT PSEUDOSECTION:
; EDNS: version: 0, flags:; udp: 4096
;; QUESTION SECTION:
;mynginx.demo.coredns.local.    IN  A

;; ANSWER SECTION:
mynginx.demo.coredns.local. 0   IN  A   10.0.0.10

;; Query time: 2 msec
;; SERVER: ::1#53(::1)
;; WHEN: Thu Jun 02 11:07:18 PDT 2016
;; MSG SIZE  rcvd: 71

Implementation Notes/Ideas

Basic Zone Mapping (implemented)

The middleware is configured with a "zone" string. For example: "zone = coredns.local".

The Kubernetes service "myservice" running in "mynamespace" would map to: "myservice.mynamespace.coredns.local".

The middleware should publish an A record for that service and a service record.

Initial implementation just performs the above simple mapping. Subsequent revisions should allow different namespaces to be published under different zones.

For example:

# Serve on port 53
.:53 {
    # use kubernetes middleware for domain "coredns.local"
    kubernetes coredns.local {
        # Use url for k8s API endpoint
        endpoint http://localhost:8080
    }
    # Perform DNS response caching for the coredns.local zone
    # Cache timeout is provided by the integer argument in seconds
    # This works for the kubernetes middleware.)
    #cache 20 coredns.local
    #cache 160 coredns.local
}

Internal IP or External IP?

• Should the Corefile configuration allow control over whether the internal IP or external IP is exposed?
• If the Corefile configuration allows control over internal IP or external IP, then the config should allow users to control the precidence.

For example a service "myservice" running in namespace "mynamespace" with internal IP "10.0.0.100" and external IP "1.2.3.4".

This example could be published as:

Corefile directive	Result
iporder = internal	10.0.0.100
iporder = external	1.2.3.4
iporder = external, internal	10.0.0.100, 1.2.3.4
iporder = internal, external	1.2.3.4, 10.0.0.100
no directive	10.0.0.100, 1.2.3.4

Wildcards

Publishing DNS records for singleton services isn't very interesting. Service names are unique within a k8s namespace therefore multiple services will be commonly run with a structured naming scheme.

For example, running multiple nginx services under the names:

Service name
c1.nginx
c2.nginx

or:

Service name
nginx.c3
nginx.c4

A DNS query with wildcard support for "nginx" in these examples should return the IP addresses for all services with "nginx" in the service name.

TBD:

• How does this relate the the k8s load-balancer configuration?
• Do wildcards search across namespaces? (Yes)
• Initial implementation assumes that a namespace maps to the first DNS label below the zone managed by the kubernetes middleware. This assumption may need to be revised. (Template scheme for record names removes this assumption.)

TODO

• SkyDNS compatibility/equivalency:
  • Kubernetes packaging and execution
    • Automate packaging to allow executing in Kubernetes. That is, add Docker container build as target in Makefile. Also include anything else needed to simplify launch as the k8s DNS service. Note: Dockerfile already exists in coredns repo to build the docker image. This work item should identify how to pass configuration and run as a SkyDNS replacement.
    • Identify any kubernetes changes necessary to use coredns as k8s DNS server. That is, how do we consume the "--cluster-dns=" and "--cluster-domain=" arguments.
    • Work out how to pass CoreDNS configuration via kubectl command line and yaml service definition file.
    • Ensure that resolver in each kubernetes container is configured to use coredns instance.
    • Update kubernetes middleware documentation to describe running CoreDNS as a SkyDNS replacement. (Include descriptions of different ways to pass CoreFile to coredns command.)
    • Remove dependency on healthz for health checking in kubernetes-rc.yaml file.
    • Expose load-balancer IP addresses.
    • Calculate SRV priority based on number of instances running. (See SkyDNS README.md)
  • Functional work
    • (done. '?' not supported yet) Implement wildcard-based lookup. Minimally support *, consider ? as well.
    • (done) Note from Miek on PR 181: "SkyDNS also supports the word any.
    • Implement SkyDNS-style synthetic zones such as "svc" to group k8s objects. (This should be optional behavior.) Also look at "pod" synthetic zones.
    • Implement test cases for SkyDNS equivalent functionality.
  • SkyDNS functionality, as listed in SkyDNS README: https://github.com/kubernetes/kubernetes/blob/release-1.2/cluster/addons/dns/README.md
    • Expose pods and srv objects.
    • A records in form of pod-ip-address.my-namespace.cluster.local. For example, a pod with ip 1.2.3.4 in the namespace default with a dns name of cluster.local would have an entry: 1-2-3-4.default.pod.cluster.local.
    • SRV records in form of _my-port-name._my-port-protocol.my-namespace.svc.cluster.local CNAME records for both regular services and headless services. See SkyDNS README.
    • A Records and hostname Based on Pod Annotations (k8s beta 1.2 feature). See SkyDNS README.
    • Note: the embedded IP and embedded port record names are weird. I would need to know the IP/port in order to create the query to lookup the name. Presumably these are intended for wildcard queries.
  • Performance
    • Improve lookup to reduce size of query result obtained from k8s API. (namespace-based?, other ideas?)
• Additional features:

  • Reverse IN-ADDR entries for services. (Is there any value in supporting reverse lookup records?) (need tests, functionality should work based on @aledbf's code.)

  • How to support label specification in Corefile to allow use of labels to indicate zone? (Is this even useful?) For example, the following configuration exposes all services labeled for the "staging" environment and tenant "customerB" in the zone "customerB.stage.local":

      kubernetes customerB.stage.local {
      	# Use url for k8s API endpoint
      	endpoint http://localhost:8080
      	label "environment" : "staging", "tenant" : "customerB"
      }
    

    Note: label specification/selection is a killer feature for segmenting test vs staging vs prod environments.

  • Implement IP selection and ordering (internal/external). Related to wildcards and SkyDNS use of CNAMES.

  • Flatten service and namespace names to valid DNS characters. (service names and namespace names in k8s may use uppercase and non-DNS characters. Implement flattening to lower case and mapping of non-DNS characters to DNS characters in a standard way.)

  • Expose arbitrary kubernetes repository data as TXT records?

• DNS Correctness
  • Do we need to generate synthetic zone records for namespaces?
  • Do we need to generate synthetic zone records for the skydns synthetic zones?
• Test cases
  • Test with CoreDNS caching. CoreDNS caching for DNS response is working using the cache directive. Tested working using 20s cache timeout and A-record queries. Automate testing with cache in place.
  • Automate CoreDNS performance tests. Initially for zone files, and for pre-loaded k8s API cache. With and without CoreDNS response caching.
  • Try to get rid of kubernetes launch scripts by moving operations into .travis.yml file.