Restore a snapshot

edit

This guide shows you how to restore a snapshot. Snapshots are a convenient way to store a copy of your data outside of a cluster. You can restore a snapshot to recover indices and data streams after deletion or a hardware failure. You can also use snapshots to transfer data between clusters.

In this guide, you’ll learn how to:

  • Get a list of available snapshots
  • Restore an index or data stream from a snapshot
  • Restore a feature state
  • Restore an entire cluster
  • Monitor the restore operation
  • Cancel an ongoing restore

This guide also provides tips for restoring to another cluster and troubleshooting common restore errors.

Prerequisites

edit
  • To use Kibana’s Snapshot and Restore feature, you must have the following permissions:

  • You can only restore a snapshot to a running cluster with an elected master node. The snapshot’s repository must be registered and available to the cluster.
  • The snapshot and cluster versions must be compatible. See Snapshot compatibility.
  • To restore a snapshot, the cluster’s global metadata must be writable. Ensure there aren’t any cluster blocks that prevent writes. The restore operation ignores index blocks.
  • Before you restore a data stream, ensure the cluster contains a matching index template with data stream enabled. To check, use Kibana’s Index Management feature or the get index template API:

    response = client.indices.get_index_template(
      name: '*',
      filter_path: 'index_templates.name,index_templates.index_template.index_patterns,index_templates.index_template.data_stream'
    )
    puts response
    GET _index_template/*?filter_path=index_templates.name,index_templates.index_template.index_patterns,index_templates.index_template.data_stream

    If no such template exists, you can create one or restore a cluster state that contains one. Without a matching index template, a data stream can’t roll over or create backing indices.

  • If your snapshot contains data from App Search or Workplace Search, ensure you’ve restored the Enterprise Search encryption key before restoring the snapshot.

Considerations

edit

When restoring data from a snapshot, keep the following in mind:

  • If you restore a data stream, you also restore its backing indices.
  • You can only restore an existing index if it’s closed and the index in the snapshot has the same number of primary shards.
  • You can’t restore an existing open index. This includes backing indices for a data stream.
  • The restore operation automatically opens restored indices, including backing indices.
  • You can restore only a specific backing index from a data stream. However, the restore operation doesn’t add the restored backing index to any existing data stream.

Get a list of available snapshots

edit

To view a list of available snapshots in Kibana, go to the main menu and click Stack Management > Snapshot and Restore.

You can also use the get repository API and the get snapshot API to find snapshots that are available to restore. First, use the get repository API to fetch a list of registered snapshot repositories.

response = client.snapshot.get_repository
puts response
GET _snapshot

Then use the get snapshot API to get a list of snapshots in a specific repository. This also returns each snapshot’s contents.

response = client.snapshot.get(
  repository: 'my_repository',
  snapshot: '*',
  verbose: false
)
puts response
GET _snapshot/my_repository/*?verbose=false

Restore an index or data stream

edit

You can restore a snapshot using Kibana’s Snapshot and Restore feature or the restore snapshot API.

By default, a restore request attempts to restore all regular indices and regular data streams in a snapshot. In most cases, you only need to restore a specific index or data stream from a snapshot. However, you can’t restore an existing open index.

If you’re restoring data to a pre-existing cluster, use one of the following methods to avoid conflicts with existing indices and data streams:

Delete and restore

edit

The simplest way to avoid conflicts is to delete an existing index or data stream before restoring it. To prevent the accidental re-creation of the index or data stream, we recommend you temporarily stop all indexing until the restore operation is complete.

If the action.destructive_requires_name cluster setting is false, don’t use the delete index API to target the * or .* wildcard pattern. If you use Elasticsearch’s security features, this will delete system indices required for authentication. Instead, target the *,-.* wildcard pattern to exclude these system indices and other index names that begin with a dot (.).

# Delete an index
DELETE my-index

# Delete a data stream
DELETE _data_stream/logs-my_app-default

In the restore request, explicitly specify any indices and data streams to restore.

POST _snapshot/my_repository/my_snapshot_2099.05.06/_restore
{
  "indices": "my-index,logs-my_app-default"
}

Rename on restore

edit

If you want to avoid deleting existing data, you can instead rename the indices and data streams you restore. You typically use this method to compare existing data to historical data from a snapshot. For example, you can use this method to review documents after an accidental update or deletion.

Before you start, ensure the cluster has enough capacity for both the existing and restored data.

The following restore snapshot API request prepends restored- to the name of any restored index or data stream.

POST _snapshot/my_repository/my_snapshot_2099.05.06/_restore
{
  "indices": "my-index,logs-my_app-default",
  "rename_pattern": "(.+)",
  "rename_replacement": "restored-$1"
}

If the rename options produce two or more indices or data streams with the same name, the restore operation fails.

If you rename a data stream, its backing indices are also renamed. For example, if you rename the logs-my_app-default data stream to restored-logs-my_app-default, the backing index .ds-logs-my_app-default-2099.03.09-000005 is renamed to .ds-restored-logs-my_app-default-2099.03.09-000005.

When the restore operation is complete, you can compare the original and restored data. If you no longer need an original index or data stream, you can delete it and use a reindex to rename the restored one.

# Delete the original index
DELETE my-index

# Reindex the restored index to rename it
POST _reindex
{
  "source": {
    "index": "restored-my-index"
  },
  "dest": {
    "index": "my-index"
  }
}

# Delete the original data stream
DELETE _data_stream/logs-my_app-default

# Reindex the restored data stream to rename it
POST _reindex
{
  "source": {
    "index": "restored-logs-my_app-default"
  },
  "dest": {
    "index": "logs-my_app-default",
    "op_type": "create"
  }
}

Restore a feature state

edit

You can restore a feature state to recover system indices, system data streams, and other configuration data for a feature from a snapshot.

If you restore a snapshot’s cluster state, the operation restores all feature states in the snapshot by default. Similarly, if you don’t restore a snapshot’s cluster state, the operation doesn’t restore any feature states by default. You can also choose to restore only specific feature states from a snapshot, regardless of the cluster state.

To view a snapshot’s feature states, use the get snapshot API.

response = client.snapshot.get(
  repository: 'my_repository',
  snapshot: 'my_snapshot_2099.05.06'
)
puts response
GET _snapshot/my_repository/my_snapshot_2099.05.06

The response’s feature_states property contains a list of features in the snapshot as well as each feature’s indices.

To restore a specific feature state from the snapshot, specify the feature_name from the response in the restore snapshot API’s feature_states parameter.

When you restore a feature state, Elasticsearch closes and overwrites the feature’s existing indices.

Restoring the security feature state overwrites system indices used for authentication. If you use Elasticsearch Service, ensure you have access to the Elasticsearch Service Console before restoring the security feature state. If you run Elasticsearch on your own hardware, create a superuser in the file realm to ensure you’ll still be able to access your cluster.

POST _snapshot/my_repository/my_snapshot_2099.05.06/_restore
{
  "feature_states": [ "geoip" ],
  "include_global_state": false,    
  "indices": "-*"                   
}

Exclude the cluster state from the restore operation.

Exclude the other indices and data streams in the snapshot from the restore operation.

Restore an entire cluster

edit

In some cases, you need to restore an entire cluster from a snapshot, including the cluster state and all feature states. These cases should be rare, such as in the event of a catastrophic failure.

Restoring an entire cluster involves deleting important system indices, including those used for authentication. Consider whether you can restore specific indices or data streams instead.

If you’re restoring to a different cluster, see Restore to a different cluster before you start.

  1. If you backed up the cluster’s configuration files, you can restore them to each node. This step is optional and requires a full cluster restart.

    After you shut down a node, copy the backed-up configuration files over to the node’s $ES_PATH_CONF directory. Before restarting the node, ensure elasticsearch.yml contains the appropriate node roles, node name, and other node-specific settings.

    If you choose to perform this step, you must repeat this process on each node in the cluster.

  2. Temporarily stop indexing and turn off the following features:

    • GeoIP database downloader and ILM history store

      response = client.cluster.put_settings(
        body: {
          persistent: {
            "ingest.geoip.downloader.enabled": false,
            "indices.lifecycle.history_index_enabled": false
          }
        }
      )
      puts response
      PUT _cluster/settings
      {
        "persistent": {
          "ingest.geoip.downloader.enabled": false,
          "indices.lifecycle.history_index_enabled": false
        }
      }
    • ILM

      response = client.ilm.stop
      puts response
      POST _ilm/stop
    • Machine Learning

      response = client.ml.set_upgrade_mode(
        enabled: true
      )
      puts response
      POST _ml/set_upgrade_mode?enabled=true
    • Monitoring

      response = client.cluster.put_settings(
        body: {
          persistent: {
            "xpack.monitoring.collection.enabled": false
          }
        }
      )
      puts response
      PUT _cluster/settings
      {
        "persistent": {
          "xpack.monitoring.collection.enabled": false
        }
      }
    • Watcher

      response = client.watcher.stop
      puts response
      POST _watcher/_stop
  3. If you use Elasticsearch security features, log in to a node host, navigate to the Elasticsearch installation directory, and add a user with the superuser role to the file realm using the elasticsearch-users tool.

    For example, the following command creates a user named restore_user.

    ./bin/elasticsearch-users useradd restore_user -p my_password -r superuser

    Use this file realm user to authenticate requests until the restore operation is complete.

  4. Use the cluster update settings API to set action.destructive_requires_name to false. This lets you delete data streams and indices using wildcards.

    response = client.cluster.put_settings(
      body: {
        persistent: {
          "action.destructive_requires_name": false
        }
      }
    )
    puts response
    PUT _cluster/settings
    {
      "persistent": {
        "action.destructive_requires_name": false
      }
    }
  5. Delete all existing data streams on the cluster.

    response = client.indices.delete_data_stream(
      name: '*',
      expand_wildcards: 'all'
    )
    puts response
    DELETE _data_stream/*?expand_wildcards=all
  6. Delete all existing indices on the cluster.

    response = client.indices.delete(
      index: '*',
      expand_wildcards: 'all'
    )
    puts response
    DELETE *?expand_wildcards=all
  7. Restore the entire snapshot, including the cluster state. By default, restoring the cluster state also restores any feature states in the snapshot.

    POST _snapshot/my_repository/my_snapshot_2099.05.06/_restore
    {
      "indices": "*",
      "include_global_state": true
    }
  8. When the restore operation is complete, resume indexing and restart any features you stopped:

    • GeoIP database downloader and ILM history store

      response = client.cluster.put_settings(
        body: {
          persistent: {
            "ingest.geoip.downloader.enabled": true,
            "indices.lifecycle.history_index_enabled": true
          }
        }
      )
      puts response
      PUT _cluster/settings
      {
        "persistent": {
          "ingest.geoip.downloader.enabled": true,
          "indices.lifecycle.history_index_enabled": true
        }
      }
    • ILM

      response = client.ilm.start
      puts response
      POST _ilm/start
    • Machine Learning

      response = client.ml.set_upgrade_mode(
        enabled: false
      )
      puts response
      POST _ml/set_upgrade_mode?enabled=false
    • Monitoring

      response = client.cluster.put_settings(
        body: {
          persistent: {
            "xpack.monitoring.collection.enabled": true
          }
        }
      )
      puts response
      PUT _cluster/settings
      {
        "persistent": {
          "xpack.monitoring.collection.enabled": true
        }
      }
    • Watcher

      response = client.watcher.start
      puts response
      POST _watcher/_start
  9. If wanted, reset the action.destructive_requires_name cluster setting.

    response = client.cluster.put_settings(
      body: {
        persistent: {
          "action.destructive_requires_name": nil
        }
      }
    )
    puts response
    PUT _cluster/settings
    {
      "persistent": {
        "action.destructive_requires_name": null
      }
    }

Monitor a restore

edit

The restore operation uses the shard recovery process to restore an index’s primary shards from a snapshot. While the restore operation recovers primary shards, the cluster will have a yellow health status.

After all primary shards are recovered, the replication process creates and distributes replicas across eligible data nodes. When replication is complete, the cluster health status typically becomes green.

Once you start a restore in Kibana, you’re navigated to the Restore Status page. You can use this page to track the current state for each shard in the snapshot.

You can also monitor snapshot recover using Elasticsearch APIs. To monitor the cluster health status, use the cluster health API.

$response = $client->cluster()->health();
resp = client.cluster.health()
print(resp)
response = client.cluster.health
puts response
res, err := es.Cluster.Health()
fmt.Println(res, err)
const response = await client.cluster.health()
console.log(response)
GET _cluster/health

To get detailed information about ongoing shard recoveries, use the index recovery API.

GET my-index/_recovery

To view any unassigned shards, use the cat shards API.

response = client.cat.shards(
  v: true,
  h: 'index,shard,prirep,state,node,unassigned.reason',
  s: 'state'
)
puts response
GET _cat/shards?v=true&h=index,shard,prirep,state,node,unassigned.reason&s=state

Unassigned shards have a state of UNASSIGNED. The prirep value is p for primary shards and r for replicas. The unassigned.reason describes why the shard remains unassigned.

To get a more in-depth explanation of an unassigned shard’s allocation status, use the cluster allocation explanation API.

GET _cluster/allocation/explain
{
  "index": "my-index",
  "shard": 0,
  "primary": false,
  "current_node": "my-node"
}

Cancel a restore

edit

You can delete an index or data stream to cancel its ongoing restore. This also deletes any existing data in the cluster for the index or data stream. Deleting an index or data stream doesn’t affect the snapshot or its data.

# Delete an index
DELETE my-index

# Delete a data stream
DELETE _data_stream/logs-my_app-default

Restore to a different cluster

edit

Elasticsearch Service can help you restore snapshots from other deployments. See Work with snapshots.

Snapshots aren’t tied to a particular cluster or a cluster name. You can create a snapshot in one cluster and restore it in another compatible cluster. Any data stream or index you restore from a snapshot must also be compatible with the current cluster’s version. The topology of the clusters doesn’t need to match.

To restore a snapshot, its repository must be registered and available to the new cluster. If the original cluster still has write access to the repository, register the repository as read-only. This prevents multiple clusters from writing to the repository at the same time and corrupting the repository’s contents. It also prevents Elasticsearch from caching the repository’s contents, which means that changes made by other clusters will become visible straight away.

Before you start a restore operation, ensure the new cluster has enough capacity for any data streams or indices you want to restore. If the new cluster has a smaller capacity, you can:

  • Add nodes or upgrade your hardware to increase capacity.
  • Restore fewer indices and data streams.
  • Reduce the number of replicas for restored indices.

    For example, the following restore snapshot API request uses the index_settings option to set index.number_of_replicas to 1.

    POST _snapshot/my_repository/my_snapshot_2099.05.06/_restore
    {
      "indices": "my-index,logs-my_app-default",
      "index_settings": {
        "index.number_of_replicas": 1
      }
    }

If indices or backing indices in the original cluster were assigned to particular nodes using shard allocation filtering, the same rules will be enforced in the new cluster. If the new cluster does not contain nodes with appropriate attributes that a restored index can be allocated on, the index will not be successfully restored unless these index allocation settings are changed during the restore operation.

The restore operation also checks that restored persistent settings are compatible with the current cluster to avoid accidentally restoring incompatible settings. If you need to restore a snapshot with incompatible persistent settings, try restoring it without the global cluster state.

Troubleshoot restore errors

edit

Here’s how to resolve common errors returned by restore requests.

Cannot restore index [<index>] because an open index with same name already exists in the cluster

edit

You can’t restore an open index that already exists. To resolve this error, try one of the methods in Restore an index or data stream.

Cannot restore index [<index>] with [x] shards from a snapshot of index [<snapshot-index>] with [y] shards

edit

You can only restore an existing index if it’s closed and the index in the snapshot has the same number of primary shards. This error indicates the index in the snapshot has a different number of primary shards.

To resolve this error, try one of the methods in Restore an index or data stream.