Upgrade Redpanda in Kubernetes

To benefit from Redpanda’s new features and enhancements, upgrade to the latest version. New features are available after all brokers in the cluster are upgraded and restarted.

Redpanda Self-Managed version numbers follow the convention AB.C.D, where AB is the two digit year, C is the feature release, and D is the patch release. For example, version 22.3.1 indicates the first patch release on the third feature release of the year 2022. Patch releases include bug fixes and minor improvements, with no change to user-facing behavior. New and enhanced features are documented with each feature release. You can find a list of all releases on GitHub.

Limitations

The following limitations ensure a smooth transition between versions and help to maintain the stability of your cluster.

  • Broker upgrades:

    • New features are enabled only after upgrading all brokers in the cluster.

    • You can upgrade only one feature release at a time, for example from 22.2 to 22.3. Skipping feature releases is not supported.

  • Rollbacks: You can roll back to the original version only if at least one broker is still running the original version (not yet upgraded) and the cluster hasn’t yet restarted.

  • Downgrades: Downgrades are possible only between patch releases of the same feature release. For example, you can downgrade from 22.2.2 to 22.2.1. Downgrading to previous feature releases, such as 22.1.x, is not supported.

  • Tiered Storage: If you have Tiered Storage enabled and you’re upgrading to 23.2, object storage uploads are paused until all brokers are upgraded. If the cluster cannot upgrade, roll it back to the original version.

    In a mixed-version state, the cluster could run out of disk space. If you need to force a mixed-version cluster to upload, transfer partition leadership to brokers running the original version.

  • Remote Read Replicas: Upgrade the Remote Read Replica cluster first, ensuring it’s on the same version as the origin cluster or one feature release ahead of the origin cluster. When upgrading to Redpanda 23.2, metadata from object storage is not synchronized until all brokers in the cluster are upgraded. If you need to force a mixed-version cluster to sync read replicas, transfer partition leadership to brokers running the original version.

Prerequisites

Impact of broker restarts

When brokers restart, clients may experience higher latency, nodes may experience CPU spikes when the broker becomes available again, and you may receive alerts about under-replicated partitions. Topics that weren’t using replication (that is, topics that had replication.factor=1) will be unavailable.

Temporary increase in latency on clients (producers and consumers)

When you restart one or more brokers in a cluster, clients (consumers and producers) may experience higher latency due to partition leadership reassignment. Because clients must communicate with the leader of a partition, they may send a request to a broker whose leadership has been transferred, and receive NOT_LEADER_FOR_PARTITION. In this case, clients must request metadata from the cluster to find out the address of the new leader. Clients refresh their metadata periodically, or when the client receives some retryable errors that indicate that the metadata may be stale. For example:

  1. Broker A shuts down.

  2. Client sends a request to broker A, and receives NOT_LEADER_FOR_PARTITION.

  3. Client requests metadata, and learns that the new leader is broker B.

  4. Client sends the request to broker B.

CPU spikes upon broker restart

When a restarted broker becomes available again, you may see your nodes' CPU usage increase temporarily. This temporary increase in CPU usage is due to the cluster rebalancing the partition replicas.

Under-replicated partitions

When a broker is in maintenance mode, Redpanda continues to replicate updates to that broker. When a broker is taken offline during a restart, partitions with replicas on the broker could become out of sync until it is brought back online. Once the broker is available again, data is copied to its under-replicated replicas until all affected partitions are in sync with the partition leader.

Incompatible changes

Patch releases in 22.3.14 and 23.1.2 changed the behavior when remote read is disabled and the requested Raft term falls below the local log’s beginning. In earlier versions, Redpanda returned an offset -1. With the patch, when you request a value older than the lowest offset, Redpanda returns the lowest offset, not -1.

Check your current Redpanda version

Before you perform a rolling upgrade:

  • Find the Redpanda version that you are currently running.

    To find your current version of Redpanda, use rpk redpanda admin brokers list:

    kubectl exec <pod-name> --namespace <namespace> -c redpanda -- \
  rpk redpanda admin brokers list
    bash
    Expected output:

    The Redpanda version for each broker is listed under BROKER-VERSION.

    NODE-ID  BROKER-VERSION
0        v22.2.10
1        v22.2.10
2        v22.2.10

  • Review the Kubernetes compatibility matrix to find out if you need to upgrade the Helm chart or the Redpanda Operator to use your chosen version of Redpanda.

    If your current version of Redpanda is more than one feature release behind the one to which you want to upgrade, you must first upgrade to an intermediate version of Redpanda.

  • Learn what’s changed since your original version.

    To find information about what has changed between Redpanda versions, check the release notes.

Prepare your cluster

Before you upgrade, you must make sure that your cluster is in a healthy state and that your topics are configured to limit outages during the upgrade process.

  1. Check for topics that have a replication factor greater than one.

    If you have topics with a replication factor of 1, and if you have sufficient disk space, temporarily increase the replication factor to limit outages for these topics during the rolling upgrade.

    Increase the replication factor before you upgrade to ensure that Redpanda has time to replicate data to other brokers.

  2. Ensure that the cluster is healthy:

    kubectl exec <pod-name> --namespace <namespace> -c redpanda -- \
  rpk cluster health
    bash

    The draining process won’t start until the cluster is healthy.

    Example output: 
    CLUSTER HEALTH OVERVIEW
=======================
Healthy:                     true (1)
Controller ID:               0
All nodes:                   [0 1 2] (2)
Nodes down:                  [] (3)
Leaderless partitions:       [] (3)
Under-replicated partitions: [] (3)
    1 The cluster is either healthy (true) or unhealthy (false).
    2 The node IDs of all brokers in the cluster.
    3 These fields contain data only when the cluster is unhealthy.

Perform a rolling upgrade

Performing a rolling upgrade allows you to update the version of Redpanda managed by the Redpanda Helm chart without downtime. This process ensures that each broker is sequentially updated and restarted, minimizing the impact on your environment.

You can use two methods to upgrade a Redpanda cluster in Kubernetes. The first method is to upgrade the Helm release to a newer version of the Redpanda Helm chart that uses the desired Redpanda version as a default. The second method is to update the existing Helm release to use a newer Redpanda image. The first method is preferred because upgrading the entire chart ensures that any new parameters required to configure the cluster are defined.

Upgrading a Redpanda cluster in Kubernetes triggers a sequential restart of the Pods managed by the StatefulSet. During each broker’s restart, the following steps occur:

  • The preStop lifecycle hook is executed to place the broker into maintenance mode. This step ensures that the broker stops accepting new connections and finishes processing its current tasks.

  • Kubernetes then terminates the Pod. If the broker does not shut down within the allowed grace period (default 90 seconds), Kubernetes forcefully terminates it using a SIGKILL signal.

  • After the Pod is terminated, the postStart lifecycle hook is executed to take the broker out of maintenance mode, allowing it to rejoin the cluster once restarted.

  1. Review the Kubernetes compatibility matrix and determine the version of the Redpanda Operator that is compatible with the Helm chart version you plan to use. The Redpanda Operator must be able to understand and manage the Helm chart and the Redpanda version you are deploying. If you need to upgrade, see Upgrade the Redpanda Operator.

  2. Check the default Redpanda version of a chart to make sure that it uses the version that you want to upgrade your cluster to.

    helm show chart --version <chart-version> redpanda/redpanda | grep  "appVersion"
    bash

    Replace <chart-version> with the version number of a newer chart.

  3. Upgrade the Redpanda version by either updating the Helm chart version or the Redpanda image.

    redpanda-cluster.yaml
    apiVersion: cluster.redpanda.com/v1alpha1
kind: Redpanda
metadata:
  name: redpanda
spec:
  chartRef:
    chartVersion: <helm-chart-version> 
  clusterSpec:
    image:
      # Optional
      tag: <new-version> 
    statefulset:
      # Optional
      terminationGracePeriodSeconds: <grace-period>
    yaml
    1 The version of the Redpanda Helm chart to deploy.
    2 If you need to upgrade to an intermediate version of Redpanda, use this setting to specify the version of Redpanda to deploy. This version overrides the default one in the Helm chart. Replace <new-version> with a valid version tag.
    3 The statefulset.terminationGracePeriodSeconds setting defines how long Kubernetes will wait for the broker to shut down gracefully before forcefully terminating it. The default value is 90 seconds, which is enough for most clusters, but might require adjustment based on your workload. Modify this setting in your Helm values file if your Redpanda brokers have high loads or hold large amounts of data, as they might need more time to shut down gracefully.

  4. Apply the Redpanda resource to deploy the Redpanda cluster:

    kubectl apply -f redpanda-cluster.yaml --namespace <namespace>
    bash

Verify the upgrade

After upgrading, verify that your Redpanda cluster is functioning correctly:

  1. Wait for the Pods to be terminated and recreated with the new version of Redpanda.

    kubectl get pod --namespace <namespace> --watch
    bash

    Each Pod in the StatefulSet is terminated one at a time, starting from the one with the highest ordinal.

    Example output 
    NAME                                    READY   STATUS
redpanda-controller-operator            2/2     Running
redpanda-0                              2/2     Running
redpanda-1                              2/2     Running
redpanda-2                              0/2     Init:0/3
redpanda-configuration-88npt            0/1     Completed
redpanda-console-7cf85cf87f-rmtnj       1/1     Running
redpanda-post-upgrade-ljqpr             0/1     Completed

  2. When all of the Pods are ready and have a Running status, verify that the brokers are now running the upgraded version of Redpanda:

    kubectl exec <pod-name> --namespace <namespace> -c redpanda -- \
  rpk redpanda admin brokers list
    bash

Roll back

If something does not go as planned during a rolling upgrade, you can roll back to the original version as long as you have not upgraded all brokers.

The StatefulSet uses the RollingUpdate strategy by default in statefulset.updateStrategy.type, which means all Pods in the StatefulSet are restarted in reverse-ordinal order. For details, see the Kubernetes documentation.

The Redpanda Operator rolls back automatically after three failed attempts to upgrade the cluster.

Troubleshooting

HelmRelease is not ready

If you are using the Redpanda Operator, you may see the following message while waiting for a Redpanda custom resource to be deployed:

NAME       READY   STATUS
redpanda   False   HelmRepository 'redpanda/redpanda-repository' is not ready
redpanda   False   HelmRelease 'redpanda/redpanda' is not ready
bash

While the deployment process can sometimes take a few minutes, a prolonged 'not ready' status may indicate an issue. Follow the steps below to investigate:

  1. Check the status of the HelmRelease:

    kubectl describe helmrelease <redpanda-resource-name> --namespace <namespace>
    bash

  2. Review the Redpanda Operator logs:

    kubectl logs -l app.kubernetes.io/name=operator -c manager --namespace <namespace>
    bash

HelmRelease retries exhausted

The HelmRelease retries exhausted error occurs when the Helm Controller has tried to reconcile the HelmRelease a number of times, but these attempts have failed consistently.

The Helm Controller watches for changes in HelmRelease objects. When changes are detected, it tries to reconcile the state defined in the HelmRelease with the state in the cluster. The process of reconciliation includes installation, upgrade, testing, rollback or uninstallation of Helm releases.

You may see this error due to:

  • Incorrect configuration in the HelmRelease.

  • Issues with the chart, such as a non-existent chart version or the chart repository not being accessible.

  • Missing dependencies or prerequisites required by the chart.

  • Issues with the underlying Kubernetes cluster, such as insufficient resources or connectivity issues.

To debug this error do the following:

  1. Check the status of the HelmRelease:

    kubectl describe helmrelease <cluster-name> --namespace <namespace>
    bash

  2. Review the Redpanda Operator logs:

    kubectl logs -l app.kubernetes.io/name=operator -c manager --namespace <namespace>
    bash

When you find and fix the error, you must use the Flux CLI, fluxctl, to suspend and resume the reconciliation process:

  1. Install Flux CLI.

  2. Suspend the HelmRelease:

    flux suspend helmrelease <cluster-name> --namespace <namespace>
    bash

  3. Resume the HelmRelease:

    flux resume helmrelease <cluster-name> --namespace <namespace>
    bash

Crash loop backoffs

If a broker crashes after startup, or gets stuck in a crash loop, it could produce progressively more stored state that uses additional disk space and takes more time for each restart to process.

To prevent infinite crash loops, the Redpanda Helm chart sets the crash_loop_limit node property to 5. The crash loop limit is the number of consecutive crashes that can happen within one hour of each other. After Redpanda reaches this limit, it will not start until its internal consecutive crash counter is reset to zero. In Kubernetes, the Pod running Redpanda remains in a CrashLoopBackoff state until its internal consecutive crash counter is reset to zero.

To troubleshoot a crash loop backoff:

  1. Check the Redpanda logs from the most recent crashes:

    kubectl logs <pod-name> --namespace <namespace>
    bash
    Kubernetes retains logs only for the current and the previous instance of a container. This limitation makes it difficult to access logs from earlier crashes, which may contain vital clues about the root cause of the issue. Given these log retention limitations, setting up a centralized logging system is crucial. Systems such as Loki or Datadog can capture and store logs from all containers, ensuring you have access to historical data.

  2. Resolve the issue that led to the crash loop backoff.

  3. Reset the crash counter to zero to allow Redpanda to restart. You can do any of the following to reset the counter:

    • Update the redpanda.yaml configuration file. You can make changes to any of the following sections in the Redpanda Helm chart to trigger an update:

      • config.cluster

      • config.node

      • config.tunable

    • Delete the startup_log file in the broker’s data directory.

      kubectl exec <pod-name> --namespace <namespace> -- rm /var/lib/redpanda/data/startup_log
      bash
      It might be challenging to execute this command within a Pod that is in a CrashLoopBackoff state due to the limited time during which the Pod is available before it restarts. Wrapping the command in a loop might work.

    • Wait one hour since the last crash. The crash counter resets after one hour.

To avoid future crash loop backoffs and manage the accumulation of small segments effectively:

  • Monitor the size and number of segments regularly.

  • Optimize your Redpanda configuration for segment management.

  • Consider implementing Tiered Storage to manage data more efficiently.

StatefulSet never rolls out

If the StatefulSet Pods remain in a pending state, they are waiting for resources to become available.

To identify the Pods that are pending, use the following command:

kubectl get pod --namespace <namespace>
bash

The response includes a list of Pods in the StatefulSet and their status.

To view logs for a specific Pod, use the following command.

kubectl logs -f <pod-name> --namespace <namespace>
bash

You can use the output to debug your deployment.

Didn’t match pod anti-affinity rules

If you see this error, your cluster does not have enough nodes to satisfy the anti-affinity rules:

Warning  FailedScheduling  18m  default-scheduler  0/1 nodes are available: 1 node(s) didn't match pod anti-affinity rules. preemption: 0/1 nodes are available: 1 No preemption victims found for incoming pod.

The Helm chart configures default podAntiAffinity rules to make sure that only one Pod running a Redpanda broker is scheduled on each worker node. To learn why, see Number of workers.

To resolve this issue, do one of the following:

  • Create additional worker nodes.

  • Modify the anti-affinity rules (for development purposes only).

    If adding nodes is not an option, you can modify the podAntiAffinity rules in your StatefulSet to be less strict.

    redpanda-cluster.yaml
    apiVersion: cluster.redpanda.com/v1alpha1
kind: Redpanda
metadata:
  name: redpanda
spec:
  chartRef: {}
  clusterSpec:
    statefulset:
      podAntiAffinity:
        type: soft
    yaml
    kubectl apply -f redpanda-cluster.yaml --namespace <namespace>
    bash

Unable to mount volume

If you see volume mounting errors in the Pod events or in the Redpanda logs, ensure that each of your Pods has a volume available in which to store data.

  • If you’re using StorageClasses with dynamic provisioners (default), ensure they exist:

    kubectl get storageclass
    bash

  • If you’re using PersistentVolumes, ensure that you have one PersistentVolume available for each Redpanda broker, and that each one has the storage capacity that’s set in storage.persistentVolume.size:

    kubectl get persistentvolume --namespace <namespace>
    bash

To learn how to configure different storage volumes, see Configure Storage.

Failed to pull image

When deploying the Redpanda Helm chart, you may encounter Docker rate limit issues because the default registry URL is not recognized as a Docker Hub URL. The domain docker.redpanda.com is used for statistical purposes, such as tracking the number of downloads. It mirrors Docker Hub’s content while providing specific analytics for Redpanda.

Failed to pull image "docker.redpanda.com/redpandadata/redpanda:v<version>": rpc error: code = Unknown desc = failed to pull and unpack image "docker.redpanda.com/redpandadata/redpanda:v<version>": failed to copy: httpReadSeeker: failed open: unexpected status code 429 Too Many Requests - Server message: toomanyrequests: You have reached your pull rate limit. You may increase the limit by authenticating and upgrading: https://www.docker.com/increase-rate-limit

To fix this error, do one of the following:

  • Replace the image.repository value in the Helm chart with docker.io/redpandadata/redpanda. Switching to Docker Hub avoids the rate limit issues associated with docker.redpanda.com.

    redpanda-cluster.yaml
    apiVersion: cluster.redpanda.com/v1alpha1
kind: Redpanda
metadata:
  name: redpanda
spec:
  chartRef: {}
  clusterSpec:
    image:
      repository: docker.io/redpandadata/redpanda
    yaml
    kubectl apply -f redpanda-cluster.yaml --namespace <namespace>
    bash

  • Authenticate to Docker Hub by logging in with your Docker Hub credentials. The docker.redpanda.com site acts as a reflector for Docker Hub. As a result, when you log in with your Docker Hub credentials, you will bypass the rate limit issues.

Dig not defined

This error means that you are using an unsupported version of Helm:

Error: parse error at (redpanda/templates/statefulset.yaml:203): function "dig" not defined

To fix this error, ensure that you are using the minimum required version: 3.10.0.

helm version
bash

Repository name already exists

If you see this error, remove the redpanda chart repository, then try installing it again.

helm repo remove redpanda
helm repo add redpanda https://charts.redpanda.com
helm repo update
bash

Fatal error during checker "Data directory is writable" execution

This error appears when Redpanda does not have write access to your configured storage volume under storage in the Helm chart.

Error: fatal error during checker "Data directory is writable" execution: open /var/lib/redpanda/data/test_file: permission denied

To fix this error, set statefulset.initContainers.setDataDirOwnership.enabled to true so that the initContainer can set the correct permissions on the data directories.

Cannot patch "redpanda" with kind StatefulSet

This error appears when you run helm upgrade with the --values flag but do not include all your previous overrides.

Error: UPGRADE FAILED: cannot patch "redpanda" with kind StatefulSet: StatefulSet.apps "redpanda" is invalid: spec: Forbidden: updates to statefulset spec for fields other than 'replicas', 'template', 'updateStrategy', 'persistentVolumeClaimRetentionPolicy' and 'minReadySeconds' are forbidden

To fix this error, do one of the following:

  • Include all the value overrides from the previous installation or upgrade using either the --set or the --values flags.

  • Use the --reuse-values flag.

    Do not use the --reuse-values flag to upgrade from one version of the Helm chart to another. This flag stops Helm from using any new values in the upgraded chart.

Cannot patch "redpanda-console" with kind Deployment

This error appears if you try to upgrade your deployment and you already have console.enabled set to true.

Error: UPGRADE FAILED: cannot patch "redpanda-console" with kind Deployment: Deployment.apps "redpanda-console" is invalid: spec.selector: Invalid value: v1.LabelSelector{MatchLabels:map[string]string{"app.kubernetes.io/instance":"redpanda", "app.kubernetes.io/name":"console"}, MatchExpressions:[]v1.LabelSelectorRequirement(nil)}: field is immutable

To fix this error, set console.enabled to false so that Helm doesn’t try to deploy Redpanda Console again.

Helm is in a pending-rollback state

An interrupted Helm upgrade process can leave your Helm release in a pending-rollback state. This state prevents further actions like upgrades, rollbacks, or deletions through standard Helm commands. To fix this:

  1. Identify the Helm release that’s in a pending-rollback state:

    helm list --namespace <namespace> --all
    bash

    Look for releases with a status of pending-rollback. These are the ones that need intervention.

  2. Verify the Secret’s status to avoid affecting the wrong resource:

    kubectl --namespace <namespace> get secret --show-labels
    bash

    Identify the Secret associated with your Helm release by its pending-rollback status in the labels.

    Ensure you have correctly identified the Secret to avoid unintended consequences. Deleting the wrong Secret could impact other deployments or services.

  3. Delete the Secret to clear the pending-rollback state:

    kubectl --namespace <namespace> delete secret -l status=pending-rollback
    bash

After clearing the pending-rollback state:

  • Retry the upgrade: Restart the upgrade process. You should investigate the initial failure to avoid getting into the pending-rollback state again.

  • Perform a rollback: If you need to roll back to a previous release, use helm rollback <release-name> <revision> to revert to a specific, stable release version.

Suggested reading

Set up a real-time dashboard to monitor your cluster health, see Monitor Redpanda.