Documentation

Node-agent Configuration

Glossary

Data Mover Pods: Data Mover Pods are:

  • Pods launched by Velero built-in data mover to run the data transfer during CSI Snapshot Data Movement, i.e., DataUpload pod and DataDownload pod.
  • Pods launched by Velero to run the data transfer during File System backup, i.e., PodVolumeBackup pod and PodVolumeRestore pod.

Overview

The Velero node-agent is a DaemonSet that hosts modules for completing backup and restore operations, including file system backup/restore and CSI snapshot data movement. This document provides comprehensive configuration options for the ConfigMap provisioned by node-agent’s --node-agent-configmap parameter.

Node-agent puts advanced configurations of Data Mover Pods into a ConfigMap that contains JSON configuration. The ConfigMap should be created in the same namespace where Velero is installed, and its name is specified using the --node-agent-configmap parameter.

Creating and Managing the ConfigMap

Notice: The ConfigMap’s life cycle control is out of the scope of Velero. Users need to create and maintain the ConfigMap themselves.

Important: The node-agent server checks configurations at startup time. After editing the ConfigMap, restart the node-agent DaemonSet for changes to take effect. kubectl rollout restart -n <velero-namespace> daemonset/node-agent

To create the ConfigMap:

  1. Save your configuration to a JSON file
  2. Create the ConfigMap:
kubectl create cm <ConfigMap-Name> -n velero --from-file=<json-file-name>

Specify during install

The ConfigMap name can be specified during Velero installation:

velero install --node-agent-configmap=<ConfigMap-Name>

Specify after install

To apply the ConfigMap to the node-agent DaemonSet:

kubectl edit ds node-agent -n velero

Add the ConfigMap reference to the container arguments:

spec:
  template:
    spec:
      containers:
      - args:
        - --node-agent-configmap=<ConfigMap-Name>

Configuration Sections

Load Concurrency (loadConcurrency)

Controls the concurrent number of Data Mover Pods per node to optimize resource usage and performance.

The configurations work for PodVolumeBackup, PodVolumeRestore, DataUpload, and DataDownload pods.

Configuration Options

  • globalConfig: Set default concurrent number applied to all nodes.
  • perNodeConfig: Set different concurrent numbers for specific nodes using label selectors.
  • prepareQueueLength: Set the max number of intermediate backup/restore pods under pending status.

Global Configuration

Sets a default concurrent number applied to all nodes:

{
  "loadConcurrency": {
    "globalConfig": 2
  }
}

Per-node Configuration

Specify different concurrent numbers for specific nodes using label selectors:

{
  "loadConcurrency": {
    "globalConfig": 2,
    "perNodeConfig": [
      {
        "nodeSelector": {
          "matchLabels": {
            "kubernetes.io/hostname": "node1"
          }
        },
        "number": 3
      },
      {
        "nodeSelector": {
          "matchLabels": {
            "beta.kubernetes.io/instance-type": "Standard_B4ms"
          }
        },
        "number": 5
      }
    ]
  }
}
  • Range: Starts from 1 (no concurrency per node), no upper limit
  • Priority: Per-node configuration overrides global configuration
  • Conflicts: If a node matches multiple rules, the smallest number is used
  • Default: 1 if not specified

Use Cases:

  • Increase concurrency on nodes with more resources
  • Reduce concurrency on nodes with limited resources or critical workloads
  • Prevent OOM kills and resource contention

PrepareQueueLength

The prepare queue length controls the maximum number of DataUpload/DataDownload/PodVolumeBackup/PodVolumeRestore CRs under the preparation statuses but are not yet processed by any node, which means the CR corresponding pod is pending state.

If there are thousands of intermediate backup/restore pods, and without this control, they start at the same time, then causing a big burden on the k8s API server.

{
  "loadConcurrency": {
    "prepareQueueLength": 10
  }
}
  • Range: Starts from 1 (for all node-agent pods), no upper limit
  • Scope: This parameter controls all PVB, PVR, DataUpload, and DataDownload pods pending number. It applies to all node-agent pods.
  • Default: No limitation if not specified

Use Cases:

  • Prevent too much workload pods are created, but cannot start.
  • Limit resource consumption from intermediate objects (PVCs, VolumeSnapshots, etc.)
  • Prevent resource exhaustion when backup/restore concurrency is limited
  • Balance between parallelism and resource usage

Affected CR Phases:

  • DataUpload/DataDownload CRs in Accepted or Prepared phases
  • PodVolumeBackup/PodVolumeRestore CRs in preparation phases

Node Selection (loadAffinity)

Constrains which nodes can run Data Mover Pods for CSI Snapshot Data Movement using affinity and anti-affinity rules.

The configurations work for DataUpload, and DataDownload pods.

For detailed information, see Node Selection for Data Movement.

Example:

{
  "loadAffinity": [
    {
      "nodeSelector": {
        "matchLabels": {
          "beta.kubernetes.io/instance-type": "Standard_B4ms"
        },
        "matchExpressions": [
          {
            "key": "kubernetes.io/hostname",
            "values": ["node-1", "node-2", "node-3"],
            "operator": "In"
          },
          {
            "key": "critical-workload",
            "operator": "DoesNotExist"
          }
        ]
      }
    }
  ]
}

Configuration Options

  • nodeSelector: Specify DataUpload and DataDownload pods can run on which nodes.
  • storageClass: Filter DataUpload and DataDownload pods on the PVC’s StorageClass. If not set, its corresponding nodeSelector applies to all DataUpload and DataDownload pods.

Important Limitations:

  • Only the first element without storageClass parameter in the loadAffinity array is used for general node selection
  • Additional elements are only considered if they have a storageClass field
  • To combine multiple conditions, use both matchLabels and matchExpressions in a single element

Use Cases:

  • Prevent data movement on nodes with critical workloads
  • Run data movement only on nodes with sufficient resources
  • Ensure data movement runs only on nodes where storage is accessible
  • Comply with topology constraints

Storage Class Specific Selection

Configure different node selection rules for specific storage classes:

  • For StorageClass fast-ssd, the first match is chosen, which is nodes with label "environment": "production".
  • For StorageClass hdd, the nodes with label "environment": "backup" are chosen.
{
  "loadAffinity": [
    {
      "nodeSelector": {
        "matchLabels": {
          "environment": "production"
        }
      },
      "storageClass": "fast-ssd"
    },
    {
      "nodeSelector": {
        "matchLabels": {
          "environment": "staging"
        }
      },
      "storageClass": "fast-ssd"
    },
    {
      "nodeSelector": {
        "matchLabels": {
          "environment": "backup"
        }
      },
      "storageClass": "hdd"
    }
  ]
}

Pod Resources (podResources)

Configure CPU and memory resources for Data Mover Pods to optimize performance and prevent resource conflict.

The configurations work for PodVolumeBackup, PodVolumeRestore, DataUpload, and DataDownload pods.

{
  "podResources": {
    "cpuRequest": "1000m",
    "cpuLimit": "2000m",
    "memoryRequest": "1Gi",
    "memoryLimit": "4Gi"
  }
}

Use Cases:

  • Limit resource consumption in resource-constrained clusters
  • Guarantee resources for time-critical backup/restore operations
  • Prevent OOM kills during large data transfers
  • Control scheduling priority relative to production workloads

Values: Must be valid Kubernetes Quantity expressions Validation: Request values must not exceed limit values Default: BestEffort QoS if not specified Failure Handling: Invalid values cause the entire podResources section to be ignored

For detailed information, see Data Movement Pod Resource Configuration.

Priority Class (priorityClassName)

Configure the Data Mover Pods’ PriorityClass.

The configurations work for PodVolumeBackup, PodVolumeRestore, DataUpload, and DataDownload pods.

Configuration Options

  • priorityClassName: The name of the PriorityClass to assign to backup/restore pods

Configure pod priority to control scheduling behavior:

High Priority (e.g., system-cluster-critical):

  • ✅ Faster scheduling and less likely to be preempted
  • ❌ May impact production workload performance

Low Priority (e.g., low-priority):

  • ✅ Protects production workloads from resource competition
  • ❌ May delay backup operations or cause preemption
{
  "priorityClassName": "low-priority"
}

Use Cases:

  • Control scheduling priority of backup/restore operations
  • Protect production workloads from resource competition
  • Ensure critical backups are scheduled quickly

Backup PVC Configuration (backupPVC)

Configure intermediate PVCs used during data movement backup operations for optimal performance.

The configurations work for DataUpload pods.

For detailed information, see BackupPVC Configuration for Data Movement Backup.

Configuration Options

  • storageClass: Alternative storage class for backup PVCs (defaults to source PVC’s storage class)
  • readOnly: This is a boolean value. If set to true then ReadOnlyMany will be the only value set to the backupPVC’s access modes. Otherwise ReadWriteOnce value will be used.
  • spcNoRelabeling: This is a boolean value. If set to true, then pod.Spec.SecurityContext.SELinuxOptions.Type will be set to spc_t. From the SELinux point of view, this will be considered a Super Privileged Container which means that selinux enforcement will be disabled and volume relabeling will not occur. This field is ignored if readOnly is false.

Use Cases:

  • Use read-only volumes for faster snapshot-to-volume conversion
  • Use dedicated storage classes optimized for backup operations
  • Reduce replica count for intermediate backup volumes
  • Comply with SELinux requirements in secured environments

Important Notes:

  • Ensure specified storage classes exist and support required access modes
  • In SELinux environments, always set spcNoRelabeling: true when using readOnly: true
  • Failures result in DataUpload CR staying in Accepted phase until timeout (30m default)

Storage Class Mapping

storageClass specifies alternative storage class for backup PVCs (defaults to source PVC’s storage class).

Configure different backup PVC settings per source storage class:

{
  "backupPVC": {
    "fast-storage": {
      "storageClass": "backup-storage-1"
    },
    "slow-storage": {
      "storageClass": "backup-storage-2"
    }
  }
}

ReadOnly and SPC Configuration

Create BackupPVC in ReadOnly mode, which can avoid full data clone during backup process in some storage providers, such as Ceph RBD.

{
  "backupPVC": {
    "source-storage-class": {
      "storageClass": "backup-optimized-class",
      "readOnly": true,
      "spcNoRelabeling": true
    }
  }
}

Restore PVC Configuration (restorePVC)

Configure intermediate PVCs used by Data Mover Pods during CSI Snapshot Data Movement restore.

The configurations work for DataDownload pods.

{
  "restorePVC": {
    "ignoreDelayBinding": true
  }
}

For detailed information, see RestorePVC Configuration for Data Movement Restore.

Configuration Options

  • ignoreDelayBinding: Ignore WaitForFirstConsumer binding mode constraints

Use Cases:

  • Improve restore parallelism by not waiting for pod scheduling
  • Enable volume restore without requiring a pod to be mounted
  • Work around topology constraints when you know the environment setup

Important Notes:

  • Use only when you understand your cluster’s topology constraints
  • May result in volumes provisioned on nodes where workload pods cannot be scheduled
  • Works best with node selection to ensure proper node targeting

Privileged FS Backup and Restore (privilegedFsBackup)

Add privileged permission in PodVolumeBackup and PodVolumeRestore created pod’s SecurityContext, because in some k8s environments, mounting HostPath volume needs privileged permission to work.

The configurations work for PodVolumeBackup, and PodVolumeRestore pods.

Configuration Options

  • privilegedFsBackup: Boolean value to enable privileged security context for file system backup/restore pods
{
  "privilegedFsBackup": true
}

Use Cases:

  • Enable file system backup in environments requiring privileged container access
  • Support HostPath volume mounting in restricted Kubernetes environments
  • Comply with security policies that restrict container capabilities

Important Notes:

  • In v1.17+, PodVolumeBackup and PodVolumeRestore run as independent pods using HostPath volumes
  • Required when cluster security policies restrict HostPath volume mounting

For detailed information, see Enable file system backup document

Cache PVC Configuration (cachePVC)

Configure intermediate PVCs used for CSI Snapshot Data Movement restore operations to cache the downloaded data.

The configurations work for DataDownload pods.

For detailed information, see Cache PVC Configuration for Data Movement Restore.

Configuration Options

  • thresholdInGB: Minimum backup data size (in GB) to trigger cache PVC creation during restore
  • storageClass: Storage class used to create cache PVCs.

Use Cases:

  • Improve restore performance by caching downloaded data locally
  • Reduce repeated data downloads from object storage
  • Optimize restore operations for large volumes

Important Notes:

  • Cache PVC is only created when restored data size exceeds the threshold
  • Ensure specified storage class exists and has sufficient capacity
  • Cache PVCs are temporary and cleaned up after restore completion
{
  "cachePVC": {
    "thresholdInGB": 1,
    "storageClass": "cache-optimized-storage"
  }
}

Complete Configuration Example

Here’s a comprehensive example showing how all configuration sections work together:

{
  "loadConcurrency": {
    "globalConfig": 2,
    "prepareQueueLength": 15,
    "perNodeConfig": [
      {
        "nodeSelector": {
          "matchLabels": {
            "kubernetes.io/hostname": "node1"
          }
        },
        "number": 3
      }
    ]
  },
  "loadAffinity": [
    {
      "nodeSelector": {
        "matchLabels": {
          "node-type": "backup"
        },
        "matchExpressions": [
          {
            "key": "critical-workload",
            "operator": "DoesNotExist"
          }
        ]
      }
    },
    {
      "nodeSelector": {
        "matchLabels": {
          "environment": "staging"
        }
      },
      "storageClass": "fast-ssd"
    }
  ],
  "podResources": {
    "cpuRequest": "500m",
    "cpuLimit": "1000m",
    "memoryRequest": "1Gi",
    "memoryLimit": "2Gi"
  },
  "priorityClassName": "backup-priority",
  "backupPVC": {
    "fast-storage": {
      "storageClass": "backup-optimized-class",
      "readOnly": true,
      "spcNoRelabeling": true
    },
    "slow-storage": {
      "storageClass": "backup-storage-2"
    }
  },
  "restorePVC": {
    "ignoreDelayBinding": true
  },
  "privilegedFsBackup": true,
  "cachePVC": {
      "thresholdInGB": 1,
      "storageClass": "cache-optimized-storage"
  }
}

This configuration:

  • Allows 2 concurrent operations globally, 3 on worker node1
  • Allows up to 15 operations in preparation phases
  • Runs Data Mover Pods only on backup nodes without critical workloads
  • Uses fast storage nodes for fast-ssd storage class operations
  • Limits pod resources to prevent cluster overload
  • Uses backup-priority PriorityClass for backup operations
  • Optimizes backup PVCs with read-only access and dedicated storage classes
  • Ignores delay binding for faster restores
  • Enable privileged permission for PodVolume pods
  • Enable cache PVC for file system restore
  • The cache threshold is 1GB and use dedicated StorageClass

Troubleshooting

Common Issues

  1. ConfigMap not taking effect: Restart node-agent DaemonSet after changes
  2. Invalid resource values: Check logs for validation errors; entire section ignored on failure
  3. Storage class not found: Ensure specified storage classes exist in the cluster
  4. SELinux issues: Set spcNoRelabeling: true when using readOnly: true
  5. Node selection not working: Verify node labels and check only first loadAffinity element is used

Validation

To verify your configuration is loaded correctly:

kubectl logs -n velero -l app=node-agent | grep -i config

To check current node-agent configuration:

kubectl get cm <ConfigMap-Name> -n velero -o yaml

For detailed information on specific configuration sections:

Getting Started

To help you get started, see the documentation.