General cluster and database

Inspect objects to diagnose issues

When you deploy a custom resource (CR), you might encounter a variety of issues. To pinpoint an issue, use the following commands to inspect the objects that the CR creates:

kubectl get returns basic information about deployed objects:

$ kubectl get pods -n namespace
$ kubectl get statefulset -n namespace
$ kubectl get pvc -n namespace
$ kubectl get event

kubectl describe returns detailed information about deployed objects:

$ kubectl describe pod pod-name -n namespace
$ kubectl describe statefulset name -n namespace
$ kubectl describe custom-resource-name -n namespace

Verify updates to a custom resource

Because the operator takes time to perform tasks, updates to the custom resource are not effective immediately. Use the kubectl command line tool to verify that changes are applied.

You can use the kubectl wait command to wait for a specified condition. For example, the operator uses the UpgradeInProgress condition to provide an upgrade status. After you begin the image version upgrade, wait until the operator acknowledges the upgrade and sets this condition to True:

$ kubectl wait --for=condition=UpgradeInProgress=True vdb/cluster-name –-timeout=180s

After the upgrade begins, you can wait until the operator leaves upgrade mode and sets this condition to False:

$ kubectl wait --for=condition=UpgradeInProgress=False vdb/cluster-name –-timeout=800s

For more information about kubectl wait, see the kubectl reference documentation.

Pods are running but the database is not ready

When you check the pods in your cluster, the pods are running but the database is not ready:

$ kubectl get pods
NAME                                                    READY   STATUS    RESTARTS   AGE
vertica-crd-sc1-0                                       0/1     Running   0          12m
vertica-crd-sc1-1                                       0/1     Running   1          12m
vertica-crd-sc1-2                                       0/1     Running   0          12m
verticadb-operator-controller-manager-5d9cdc9b8-kw9nv   2/2     Running   0          24m

To find the root cause of the issue, use kubectl logs to check the operator manager. The following example shows that the communal storage bucket does not exist:

$ kubectl logs -l app.kubernetes.io/name=verticadb-operator -c manager -f
2021-08-04T20:03:00.289Z        INFO    controllers.VerticaDB   ExecInPod entry {"verticadb": "default/vertica-crd", "pod": {"namespace": "default", "name": "vertica-crd-sc1-0"}, "command": "bash -c ls -l /opt/vertica/config/admintools.conf && grep '^node\\|^v_\\|^host' /opt/vertica/config/admintools.conf "}
2021-08-04T20:03:00.369Z        INFO    controllers.VerticaDB   ExecInPod stream        {"verticadb": "default/vertica-crd", "pod": {"namespace": "default", "name": "vertica-crd-sc1-0"}, "err": null, "stdout": "-rw-rw-r-- 1 dbadmin verticadba 1243 Aug  4 20:00 /opt/vertica/config/admintools.conf\nhosts = 10.244.1.5,10.244.2.4,10.244.4.6\nnode0001 = 10.244.1.5,/data,/data\nnode0002 = 10.244.2.4,/data,/data\nnode0003 = 10.244.4.6,/data,/data\n", "stderr": ""}
2021-08-04T20:03:00.369Z        INFO    controllers.VerticaDB   ExecInPod entry {"verticadb": "default/vertica-crd", "pod": {"namespace": "default", "name": "vertica-crd-sc1-0"}, "command": "/opt/vertica/bin/admintools -t create_db --skip-fs-checks --hosts=10.244.1.5,10.244.2.4,10.244.4.6 --communal-storage-location=s3://newbucket/db/26100df1-93e5-4e64-b665-533e14abb67c --communal-storage-params=/home/dbadmin/auth_parms.conf --sql=/home/dbadmin/post-db-create.sql --shard-count=12 --depot-path=/depot --database verticadb --force-cleanup-on-failure --noprompt --password ******* "}
2021-08-04T20:03:00.369Z        DEBUG   controller-runtime.manager.events       Normal  {"object": {"kind":"VerticaDB","namespace":"default","name":"vertica-crd","uid":"26100df1-93e5-4e64-b665-533e14abb67c","apiVersion":"vertica.com/v1","resourceVersion":"11591"}, "reason": "CreateDBStart", "message": "Calling 'admintools -t create_db'"}
2021-08-04T20:03:17.051Z        INFO    controllers.VerticaDB   ExecInPod stream        {"verticadb": "default/vertica-crd", "pod": {"namespace": "default", "name": "vertica-crd-sc1-0"}, "err": "command terminated with exit code 1", "stdout": "Default depot size in use\nDistributing changes to cluster.\n\tCreating database verticadb\nBootstrap on host 10.244.1.5 return code 1 stdout '' stderr 'Logged exception in writeBufferToFile: RecvFiles failed in closing file [s3://newbucket/db/26100df1-93e5-4e64-b665-533e14abb67c/verticadb_rw_access_test.txt]: The specified bucket does not exist. Writing test data to file s3://newbucket/db/26100df1-93e5-4e64-b665-533e14abb67c/verticadb_rw_access_test.txt failed.\\nTesting rw access to communal location s3://newbucket/db/26100df1-93e5-4e64-b665-533e14abb67c/ failed\\n'\n\nError: Bootstrap on host 10.244.1.5 return code 1 stdout '' stderr 'Logged exception in writeBufferToFile: RecvFiles failed in closing file [s3://newbucket/db/26100df1-93e5-4e64-b665-533e14abb67c/verticadb_rw_access_test.txt]: The specified bucket does not exist. Writing test data to file s3://newbucket/db/26100df1-93e5-4e64-b665-533e14abb67c/verticadb_rw_access_test.txt failed.\\nTesting rw access to communal location s3://newbucket/db/26100df1-93e5-4e64-b665-533e14abb67c/ failed\\n'\n\n", "stderr": ""}
2021-08-04T20:03:17.051Z        INFO    controllers.VerticaDB   aborting reconcile of VerticaDB {"verticadb": "default/vertica-crd", "result": {"Requeue":true,"RequeueAfter":0}, "err": null}
2021-08-04T20:03:17.051Z        DEBUG   controller-runtime.manager.events       Warning {"object": {"kind":"VerticaDB","namespace":"default","name":"vertica-crd","uid":"26100df1-93e5-4e64-b665-533e14abb67c","apiVersion":"vertica.com/v1","resourceVersion":"11591"}, "reason": "S3BucketDoesNotExist", "message": "The bucket in the S3 path 's3://newbucket/db/26100df1-93e5-4e64-b665-533e14abb67c' does not exist"}

Create an S3 bucket for the cluster:

$ S3_BUCKET=newbucket
$ S3_CLUSTER_IP=$(kubectl get svc | grep minio | head -1 | awk '{print $3}')
$ export AWS_ACCESS_KEY_ID=minio
$ export AWS_SECRET_ACCESS_KEY=minio123
$ aws s3 mb s3://$S3_BUCKET --endpoint-url http://$S3_CLUSTER_IP
make_bucket: newbucket

Use kubectl get pods to verify that the cluster uses the new S3 bucket and the database is ready:

$ kubectl get pods
NAME                                                    READY   STATUS    RESTARTS   AGE
minio-ss-0-0                                            1/1     Running   0          18m
minio-ss-0-1                                            1/1     Running   0          18m
minio-ss-0-2                                            1/1     Running   0          18m
minio-ss-0-3                                            1/1     Running   0          18m
vertica-crd-sc1-0                                       1/1     Running   0          20m
vertica-crd-sc1-1                                       1/1     Running   0          20m
vertica-crd-sc1-2                                       1/1     Running   0          20m
verticadb-operator-controller-manager-5d9cdc9b8-kw9nv   2/2     Running   0          63m

Database is not available

After you create a custom resource instance, the database is not available. The kubectl get custom-resource command does not display information:

$ kubectl get vdb
NAME          AGE   SUBCLUSTERS   INSTALLED   DBADDED   UP
vertica-crd   4s

Use kubectl describe custom-resource to check the events for the pods to identify any issues:

$ kubectl describe vdb
Name:         vertica-crd
Namespace:    default
Labels:       <none>
Annotations:  <none>
API Version:  vertica.com/v1
Kind:         VerticaDB
Metadata:
  ...
  Superuser Password Secret:  su-passwd
Events:
  Type     Reason                           Age                From                Message
  ----     ------                           ----               ----                -------
  Warning  SuperuserPasswordSecretNotFound  5s (x12 over 15s)  verticadb-operator  Secret for superuser password 'su-passwd' was not found

In this circumstance, the custom resource uses a Secret named su-passwd to store the Superuser Password Secret, but there is no such Secret available. Create a Secret named su-passwd to store the Secret:

$ kubectl create secret generic su-passwd --from-literal=password=sup3rs3cr3t
secret/su-passwd created

Note

Instead of creating a Kubernetes Secret with kubectl, you can manually base64 encode a string on the command line, and then add the encoded output to a Secrets manifest. For example, pass the string value to the echo command, and pipe the output to the base64 command to encode the value. In the echo command, include the -n option so that it does not append a newline character:

$ echo -n 'secret-value' | base64
c2VjcmV0LXZhbHVl

For detailed steps about creating the Secret manifest and applying it to a namespace, see the Kubernetes documentation.

Use kubectl get custom-resource to verify the issue is resolved:

$ kubectl get vdb
NAME          AGE   SUBCLUSTERS   INSTALLED   DBADDED   UP
vertica-crd   89s   1             0           0         0

Image pull failure

You receive an ImagePullBackOff error when you deploy a Vertica cluster with Helm charts, but you do not pre-pull the Vertica image from the local registry server:

$ kubectl describe pod pod-name-0
...
Events:
  Type     Reason            Age                    From               Message
  ----     ------            ----                   ----               -------
  ...
  Warning  Failed            2m32s                  kubelet            Failed to pull image "k8s-rhel7-01:5000/vertica-k8s:default-1": rpc error: code = Unknown desc = context canceled
  Warning  Failed            2m32s                  kubelet            Error: ErrImagePull
  Normal   BackOff           2m32s                  kubelet            Back-off pulling image "k8s-rhel7-01:5000/vertica-k8s:default-1"
  Warning  Failed            2m32s                  kubelet            Error: ImagePullBackOff
  Normal   Pulling           2m18s (x2 over 4m22s)  kubelet            Pulling image "k8s-rhel7-01:5000/vertica-k8s:default-1"

This occurs because the Vertica image size is too big to pull from the registry while deploying the Vertica cluster. Execute the following command on a Kubernetes host:

$ docker image list | grep vertica-k8s
k8s-rhel7-01:5000/vertica-k8s default-1 2d6f5d3d90d6 9 days ago 1.55GB

To solve this issue, complete one of the following:

Pull the Vertica images on each node before creating the Vertica StatefulSet:

$ NODES=`kubectl get nodes | grep -v NAME | awk '{print $1}'`
$ for node in $NODES; do ssh $node docker pull $DOCKER_REGISTRY:5000/vertica-k8s:$K8S_TAG; done

Use the reduced-size vertica/vertica-k8s:latest image for the Vertica server.

Pending pods due to insufficient CPU

If your host nodes do not have enough resources to fulfill the resource request from a pod, the pod stays in pending status.

Note

As a best practice, do not request the maximum amount of resources available on a host node to leave resources for other processes on the host node.

In the following example, the pod requests 40 CPUs on the host node, and the pod stays in Pending:

$ kubectl describe pod cluster-vertica-defaultsubcluster-0
...
Status:         Pending
...
Containers:
  server:
    Image:       docker.io/library/vertica-k8s:default-1
    Ports:       5433/TCP, 5434/TCP, 22/TCP
    Host Ports:  0/TCP, 0/TCP, 0/TCP
    Command:
      /opt/vertica/bin/docker-entrypoint.sh
      restart-vertica-node
    Limits:
      memory:  200Gi
    Requests:
      cpu: 40
      memory:  200Gi
...
Events:
  Type     Reason            Age    From               Message
  ----     ------            ----   ----               -------
  Warning  FailedScheduling  3h20m  default-scheduler  0/5 nodes are available: 5 Insufficient cpu.

To confirm the resources available on the host node. The following command confirms that the host node has only 40 allocatable CPUs:

$ kubectl describe node host-node-1
...
Conditions:
  Type             Status  LastHeartbeatTime                 LastTransitionTime                Reason                       Message
  ----             ------  -----------------                 ------------------                ------                       -------
  MemoryPressure   False   Sat, 20 Mar 2021 22:39:10 -0400   Sat, 20 Mar 2021 13:07:02 -0400   KubeletHasSufficientMemory   kubelet has sufficient memory available
  DiskPressure     False   Sat, 20 Mar 2021 22:39:10 -0400   Sat, 20 Mar 2021 13:07:02 -0400   KubeletHasNoDiskPressure     kubelet has no disk pressure
  PIDPressure      False   Sat, 20 Mar 2021 22:39:10 -0400   Sat, 20 Mar 2021 13:07:02 -0400   KubeletHasSufficientPID      kubelet has sufficient PID available
  Ready            True    Sat, 20 Mar 2021 22:39:10 -0400   Sat, 20 Mar 2021 13:07:12 -0400   KubeletReady                 kubelet is posting ready status
Addresses:
  InternalIP:  172.19.0.5
  Hostname:    eng-g9-191
Capacity:
  cpu:                40
  ephemeral-storage:  285509064Ki
  hugepages-1Gi:      0
  hugepages-2Mi:      0
  memory:             263839236Ki
  pods:               110
Allocatable:
  cpu:                40
  ephemeral-storage:  285509064Ki
  hugepages-1Gi:      0
  hugepages-2Mi:      0
  memory:             263839236Ki
  pods:               110
...
Non-terminated Pods:          (3 in total)
  Namespace                   Name                                   CPU Requests  CPU Limits  Memory Requests  Memory Limits  AGE
  ---------                   ----                                   ------------  ----------  ---------------  -------------  ---
  default                     cluster-vertica-defaultsubcluster-0    38 (95%)      0 (0%)      200Gi (79%)      200Gi (79%)    51m
  kube-system                 kube-flannel-ds-8brv9                  100m (0%)     100m (0%)   50Mi (0%)        50Mi (0%)      9h
  kube-system                 kube-proxy-lgjhp                       0 (0%)        0 (0%)      0 (0%)           0 (0%)         9h
...

To correct this issue, reduce the resource.requests in the subcluster to values lower than the maximum allocatable CPUs. The following example uses a YAML-formatted file named patch.yaml to lower the resource requests for the pod:

$ cat patch.yaml
spec:
  subclusters:
    - name: defaultsubcluster
      resources:
        requests:
          memory: 238Gi
          cpu: "38"
        limits:
          memory: 238Gi
$ kubectl patch vdb cluster-vertica –-type=merge --patch “$(cat patch.yaml)”
verticadb.vertica.com/cluster-vertica patched

Pending pod after node removed

When you remove a host node from your Kubernetes cluster, a Vertica pod might stay in pending status if the pod uses a PersistentVolume (PV) that has a node affinity rule that prevents the pod from running on another node.

To resolve this issue, you must verify that the pods are pending because of an affinity rule, and then use the vdb-gen tool to revive the entire cluster.

First, determine if the pod is pending because of a node affinity rule. This requires details about the pending pod, the PersistentVolumeClaim (PVC) associated with the pod, and the PersistentVolume (PV) associated with the PVC:

Use kubectl describe to return details about the pending pod:

$ kubectl describe pod pod-name
...
Events:
  Type     Reason            Age                From               Message
  ----     ------            ----               ----               -------
  Warning  FailedScheduling  28s (x2 over 48s)  default-scheduler  0/2 nodes are available: 1 node(s) had untolerated taint {node.kubernetes.io/unschedulable: }, 1 node(s) had volume node affinity conflict, 1 node(s) were unschedulable. preemption: 0/2 nodes are available: 2 Preemption is not helpful for scheduling.

The Message column verifies that the pod was not scheduled due a volume node affinity conflict.

Get the name of the PVC associated with the pod:

$ kubectl get pod -o jsonpath='{.spec.volumes[0].persistentVolumeClaim.claimName}{"\n"}' pod-name
local-data-pod-name

Use the PVC to get the PV. PVs are associated with nodes:

$ kubectl get pvc -o jsonpath='{.spec.volumeName}{"\n"}' local-data-pod-name
pvc-1926ae96-574d-4433-99b4-ec9ab0e5e497

Use the PV to get the name of the node that has the affinity rule:

$ kubectl get pv -o jsonpath='{.spec.nodeAffinity.required.nodeSelectorTerms[0].matchExpressions[0].values[0]}{"\n"}' pvc-1926ae96-574d-4433-99b4-ec9ab0e5e497
ip-10-20-30-40.ec2.internal

Verify that the node with the affinity rule is the node that was removed from the Kubernetes cluster.

Next, you must revive the entire cluster to get all pods running again. When you revive the cluster, you create new PVCs that restore the association between each pod and a PV to satisfy the node affinity rule.

While you have nodes running in the cluster, you can use the vdb-gen tool to generate a manifest and revive the database:

Download the vdb-gen tool from the vertica-kubernetes GitHub repository:

$ wget https://github.com/vertica/vertica-kubernetes/releases/latest/download/vdb-gen

Copy the tool into a pod that has a running Vertica process:
```
$ kubectl cp vdb-gen pod-name:/tmp/vdb-gen
```
The vdb-gen tool requires the database name, so retrieve it with the following command:
```
$ kubectl get vdb -o jsonpath='{.spec.dbName}{"\n"}' v
database-name
```

Run the vdb-gen tool with the database name. The following command runs the tool and pipes the output to a file named revive.yaml:

$ kubectl exec -i pod-name -- bash -c "chmod +x /tmp/vdb-gen && /tmp/vdb-gen --ignore-cluster-lease --name v localhost database-name | tee /tmp/revive.yaml"

Copy revive.yaml to your local machine so that you can use it after you remove the cluster:
```
$ kubectl cp pod-name:/tmp/revive.yaml revive.yaml
```
Save the current VerticaDB Custom Resource (CR). For example, the following command saves a CR named vertdb to a file named orig.yaml:
```
$ kubectl get vdb vertdb -o yaml > orig.yaml
```
Update revive.yaml with parts of orig.yaml that vdb-gen did not capture. For example, custom resource limits.

Delete the existing Vertica cluster:

$ kubectl delete vdb vertdb
verticadb.vertica.com "vertdb" deleted

Confirm that all PVCs that are associated with the deleted cluster were removed:

Retrieve the PVC names. A PVC name uses the dbname-subcluster-podindex format:

$ kubectl get pvc
NAME                     STATUS   VOLUME                                     CAPACITY ACCESS MODES   STORAGECLASS   AGE
local-data-vertdb-sc-0   Bound    pvc-e9834c18-bf60-4a4b-a686-ba8f7b601230   1Gi      RWO            local-path     34m
local-data-vertdb-sc-1   Bound    pvc-1926ae96-574d-4433-99b4-ec9ab0e5e497   1Gi      RWO            local-path     34m
local-data-vertdb-sc-2   Bound    pvc-4541f7c9-3afc-47f0-8d04-67fac370ee88   1Gi      RWO            local-path     34m

Delete the PVCs:

$ kubectl delete pvc local-data-vertdb-sc-0 local-data-vertdb-sc-1 local-data-vertdb-sc-2
persistentvolumeclaim "local-data-vertdb-sc-0" deleted
persistentvolumeclaim "local-data-vertdb-sc-1" deleted
persistentvolumeclaim "local-data-vertdb-sc-2" deleted

Revive the database with revive.yaml:

$ kubectl apply -f revive.yaml
verticadb.vertica.com/vertdb created

After the revive completes, all Vertica pods are running, and PVCs are recreated on new nodes. Wait for the operator to start the database.

Deploying to Istio

Vertica does not officially support Istio because the Istio sidecar port requirement conflicts with the port that Vertica requires for internal node communication. However, you can deploy Vertica on Kubernetes to Istio with changes to the Istio InboundInterceptionMode setting. Vertica provides access to this setting with annotations on the VerticaDB CR.

REDIRECT mode

REDIRECT mode is the default InboundInterceptionMode setting, and it requires that you disable network address translation (NAT) on port 5434, the port that the pods use for internal communication. Disable NAT on this port with the excludeInboundPorts annotation:

apiVersion: vertica.com/v1
kind: VerticaDB
metadata:
  name: vdb
spec:
  annotations:
    traffic.sidecar.istio.io/excludeInboundPorts: "5434"