kubernetes: Config files + setup script for secure multiregion clusters

cloud/kubernetes/README.md

@@ -71,6 +71,13 @@ deployment until node-local storage has been more rigorously hardened.

Secure mode currently works by requesting node/client certificates from the kubernetes controller at pod initialization time.

### Geographically distributed clusters

The configuration files and instructions in this directory are limited to
deployments that exist entirely within a single Kubernetes cluster. If you'd
like to deploy CockroachDB across multiple geographically distributed Kubernetes
clusters, see the [multiregion subdirectory](multiregion).

## Creating your kubernetes cluster

### Locally on minikube

cloud/kubernetes/multiregion/README.md

@@ -0,0 +1,86 @@
# Running CockroachDB across multiple Kubernetes clusters

The script and configuration files in this directory enable deploying
CockroachDB across multiple Kubernetes clusters that are spread across different
geographic regions. It deploys a CockroachDB
[StatefulSet](https://kubernetes.io/docs/concepts/workloads/controllers/statefulset/)
into each separate cluster, and links them together using DNS.

To use the configuration provided here, check out this repository (or otherwise
download a copy of this directory), fill in the constants at the top of
[setup.py](setup.py) with the relevant information about your Kubernetes
clusters, optionally make any desired modifications to
[cockroachdb-statefulset-secure.yaml](cockroachdb-statefulset-secure.yaml) as
explained in [our Kubernetes performance tuning
guide](https://www.cockroachlabs.com/docs/stable/kubernetes-performance.html),
then finally run [setup.py](setup.py).

You should see a lot of output as it does its thing, hopefully ending after
printing out `job "cluster-init-secure" created`. This implies that everything
was created successfully, and you should soon see the CockroachDB cluster
initialized with 3 pods in the "READY" state in each Kubernetes cluster. At this
point you can manage the StatefulSet in each cluster independently if you so
desire, scaling up the number of replicas, changing their resource requests, or
making other modifications as you please.

If anything goes wrong along the way, please let us know via any of the [normal
troubleshooting
channels](https://www.cockroachlabs.com/docs/stable/support-resources.html).
While we believe this creates a highly available, maintainable multi-region
deployment, it is still pushing the boundaries of how Kubernetes is typically
used, so feedback and issue reports are very appreciated.

## Limitations

### Pod-to-pod connectivity

The deployment outlined in this directory relies on pod IP addresses being
routable even across Kubernetes clusters and regions. This achieves optimal
performance, particularly when compared to alternative solutions that route all packets between clusters through load balancers, but means that it won't work in certain environments.

This requirement is satisfied by clusters deployed in cloud environments such as Google Kubernetes Engine, and
can also be satisfied by on-prem environments depending on the [Kubernetes networking setup](https://kubernetes.io/docs/concepts/cluster-administration/networking/) used. If you want to test whether your cluster will work, you can run this basic network test:

```shell
$ kubectl run network-test --image=alpine --restart=Never -- sleep 999999
pod "network-test" created
$ kubectl describe pod network-test | grep IP
IP:           THAT-PODS-IP-ADDRESS
$ kubectl config use-context YOUR-OTHER-CLUSTERS-CONTEXT-HERE
$ kubectl run -it network-test --image=alpine --restart=Never -- ping THAT-PODS-IP-ADDRESS
If you don't see a command prompt, try pressing enter.
64 bytes from 10.12.14.10: seq=1 ttl=62 time=0.570 ms
64 bytes from 10.12.14.10: seq=2 ttl=62 time=0.449 ms
64 bytes from 10.12.14.10: seq=3 ttl=62 time=0.635 ms
64 bytes from 10.12.14.10: seq=4 ttl=62 time=0.722 ms
64 bytes from 10.12.14.10: seq=5 ttl=62 time=0.504 ms
...
```

If the pods can directly connect, you should see successful ping output like the
above. If they can't, you won't see any successful ping responses. Make sure to
delete the `network-test` pod in each cluster when you're done!

### Exposing DNS servers to the Internet

As currently configured, the way that the DNS servers from each Kubernetes
cluster are hooked together is by exposing them via a load balanced IP address
that's visible to the public Internet. This is because [Google Cloud Platform's Internal Load Balancers do not currently support clients in one region using a load balancer in another region](https://cloud.google.com/compute/docs/load-balancing/internal/#deploying_internal_load_balancing_with_clients_across_vpn_or_interconnect). 

None of the services in your Kubernetes cluster will be made accessible, but
their names could leak out to a motivated attacker. If this is unacceptable,
please let us know and we can demonstrate other options. [Your voice could also
help convince Google to allow clients from one region to use an Internal Load
Balancer in another](https://issuetracker.google.com/issues/111021512),
eliminating the problem.

## Cleaning up

To remove all the resources created in your clusters by [setup.py](setup.py),
copy the parameters you provided at the top of [setup.py](setup.py) to the top
of [teardown.py](teardown.py) and run [teardown.py](teardown.py).

## More information

For more information on running CockroachDB in Kubernetes, please see the [README
in the parent directory](../README.md).

cloud/kubernetes/multiregion/client-secure.yaml

@@ -0,0 +1,27 @@
apiVersion: v1
kind: Pod
metadata:
  name: cockroachdb-client-secure
  labels:
    app: cockroachdb-client
spec:
  serviceAccountName: cockroachdb
  containers:
  - name: cockroachdb-client
    image: cockroachdb/cockroach:v2.0.5
    imagePullPolicy: IfNotPresent
    volumeMounts:
    - name: client-certs
      mountPath: /cockroach-certs
    # Keep a pod open indefinitely so kubectl exec can be used to get a shell to it
    # and run cockroach client commands, such as cockroach sql, cockroach node status, etc.
    command:
    - sleep
    - "2147483648" # 2^31
  # This pod isn't doing anything important, so don't bother waiting to terminate it.
  terminationGracePeriodSeconds: 0
  volumes:
  - name: client-certs
    secret:
      secretName: cockroachdb.client.root
      defaultMode: 256

cloud/kubernetes/multiregion/cluster-init-secure.yaml

@@ -0,0 +1,28 @@
apiVersion: batch/v1
kind: Job
metadata:
  name: cluster-init-secure
  labels:
    app: cockroachdb
spec:
  template:
    spec:
      serviceAccountName: cockroachdb
      containers:
      - name: cluster-init
        image: cockroachdb/cockroach:v2.0.5
        imagePullPolicy: IfNotPresent
        volumeMounts:
        - name: client-certs
          mountPath: /cockroach-certs
        command:
          - "/cockroach/cockroach"
          - "init"
          - "--certs-dir=/cockroach-certs"
          - "--host=cockroachdb-0.cockroachdb"
      restartPolicy: OnFailure
      volumes:
      - name: client-certs
        secret:
          secretName: cockroachdb.client.root
          defaultMode: 256

cloud/kubernetes/multiregion/cockroachdb-statefulset-secure.yaml

@@ -0,0 +1,224 @@
apiVersion: v1
kind: ServiceAccount
metadata:
  name: cockroachdb
  labels:
    app: cockroachdb
---
apiVersion: rbac.authorization.k8s.io/v1beta1
kind: Role
metadata:
  name: cockroachdb
  labels:
    app: cockroachdb
rules:
- apiGroups:
  - ""
  resources:
  - secrets
  verbs:
  - create
  - get
---
apiVersion: rbac.authorization.k8s.io/v1beta1
kind: ClusterRole
metadata:
  name: cockroachdb
  labels:
    app: cockroachdb
rules:
- apiGroups:
  - certificates.k8s.io
  resources:
  - certificatesigningrequests
  verbs:
  - create
  - get
  - watch
---
apiVersion: rbac.authorization.k8s.io/v1beta1
kind: RoleBinding
metadata:
  name: cockroachdb
  labels:
    app: cockroachdb
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: Role
  name: cockroachdb
subjects:
- kind: ServiceAccount
  name: cockroachdb
  namespace: default
---
apiVersion: rbac.authorization.k8s.io/v1beta1
kind: ClusterRoleBinding
metadata:
  name: cockroachdb
  labels:
    app: cockroachdb
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: cockroachdb
subjects:
- kind: ServiceAccount
  name: cockroachdb
  namespace: default
---
apiVersion: v1
kind: Service
metadata:
  # This service is meant to be used by clients of the database. It exposes a ClusterIP that will
  # automatically load balance connections to the different database pods.
  name: cockroachdb-public
  labels:
    app: cockroachdb
spec:
  ports:
  # The main port, served by gRPC, serves Postgres-flavor SQL, internode
  # traffic and the cli.
  - port: 26257
    targetPort: 26257
    name: grpc
  # The secondary port serves the UI as well as health and debug endpoints.
  - port: 8080
    targetPort: 8080
    name: http
  selector:
    app: cockroachdb
---
apiVersion: v1
kind: Service
metadata:
  # This service only exists to create DNS entries for each pod in the stateful
  # set such that they can resolve each other's IP addresses. It does not
  # create a load-balanced ClusterIP and should not be used directly by clients
  # in most circumstances.
  name: cockroachdb
  labels:
    app: cockroachdb
  annotations:
    # Use this annotation in addition to the actual publishNotReadyAddresses
    # field below because the annotation will stop being respected soon but the
    # field is broken in some versions of Kubernetes:
    # https://github.com/kubernetes/kubernetes/issues/58662
    service.alpha.kubernetes.io/tolerate-unready-endpoints: "true"
    # Enable automatic monitoring of all instances when Prometheus is running in the cluster.
    prometheus.io/scrape: "true"
    prometheus.io/path: "_status/vars"
    prometheus.io/port: "8080"
spec:
  ports:
  - port: 26257
    targetPort: 26257
    name: grpc
  - port: 8080
    targetPort: 8080
    name: http
  # We want all pods in the StatefulSet to have their addresses published for
  # the sake of the other CockroachDB pods even before they're ready, since they
  # have to be able to talk to each other in order to become ready.
  publishNotReadyAddresses: true
  clusterIP: None
  selector:
    app: cockroachdb
---
apiVersion: policy/v1beta1
kind: PodDisruptionBudget
metadata:
  name: cockroachdb-budget
  labels:
    app: cockroachdb
spec:
  selector:
    matchLabels:
      app: cockroachdb
  maxUnavailable: 1
---
apiVersion: apps/v1beta1
kind: StatefulSet
metadata:
  name: cockroachdb
spec:
  serviceName: "cockroachdb"
  replicas: 3
  template:
    metadata:
      labels:
        app: cockroachdb
    spec:
      serviceAccountName: cockroachdb
      affinity:
        podAntiAffinity:
          preferredDuringSchedulingIgnoredDuringExecution:
          - weight: 100
            podAffinityTerm:
              labelSelector:
                matchExpressions:
                - key: app
                  operator: In
                  values:
                  - cockroachdb
              topologyKey: kubernetes.io/hostname
      containers:
      - name: cockroachdb
        image: cockroachdb/cockroach:v2.0.5
        imagePullPolicy: IfNotPresent
        ports:
        - containerPort: 26257
          name: grpc
        - containerPort: 8080
          name: http
        livenessProbe:
          httpGet:
            path: "/health"
            port: http
            scheme: HTTPS
          initialDelaySeconds: 30
          periodSeconds: 5
        readinessProbe:
          httpGet:
            path: "/health?ready=1"
            port: http
            scheme: HTTPS
          initialDelaySeconds: 10
          periodSeconds: 5
          failureThreshold: 2
        volumeMounts:
        - name: datadir
          mountPath: /cockroach/cockroach-data
        - name: certs
          mountPath: /cockroach/cockroach-certs
        env:
        - name: COCKROACH_CHANNEL
          value: kubernetes-secure
        command:
          - "/bin/bash"
          - "-ecx"
          # The use of qualified `hostname -f` is crucial:
          # Other nodes aren't able to look up the unqualified hostname.
          - "exec /cockroach/cockroach start --logtostderr --certs-dir /cockroach/cockroach-certs --advertise-host $(hostname -f) --http-host 0.0.0.0 --join JOINLIST --locality LOCALITYLIST --cache 25% --max-sql-memory 25%"
      # No pre-stop hook is required, a SIGTERM plus some time is all that's
      # needed for graceful shutdown of a node.
      terminationGracePeriodSeconds: 60
      volumes:
      - name: datadir
        persistentVolumeClaim:
          claimName: datadir
      - name: certs
        secret:
          secretName: cockroachdb.node
          defaultMode: 256
  podManagementPolicy: Parallel
  updateStrategy:
    type: RollingUpdate
  volumeClaimTemplates:
  - metadata:
      name: datadir
    spec:
      accessModes:
        - "ReadWriteOnce"
      resources:
        requests:
          storage: 100Gi