Binary Backups

Binary backups are full backups of Dgraph that are backed up directly to cloud storage such as Amazon S3 or any Minio storage backend. Backups can also be saved to an on-premise network file system shared by all Alpha servers. These backups can be used to restore a new Dgraph cluster to the previous state from the backup. Unlike exports, binary backups are Dgraph-specific and can be used to restore a cluster quickly.

Configure Backup

Backup is only enabled when a valid license file is supplied to a Zero server OR within the thirty (30) day trial period, no exceptions.

Configure Amazon S3 Credentials

To backup to Amazon S3, the Alpha server must have the following AWS credentials set via environment variables:

Starting with v20.07.0 if the system has access to the S3 bucket, you no longer need to explicitly include these environment variables.

In AWS, you can accomplish this by doing the following:

1. Create an IAM Role with an IAM Policy that grants access to the S3 bucket.
2. Depending on whether you want to grant access to an EC2 instance, or to a pod running on EKS, you can do one of these options:
   • Instance Profile can pass the IAM Role to an EC2 Instance
   • IAM Roles for Amazon EC2 to attach the IAM Role to a running EC2 Instance
   • IAM roles for service accounts to associate the IAM Role to a Kubernetes Service Account.

Configure Minio Credentials

To backup to Minio, the Alpha server must have the following Minio credentials set via environment variables:

Create a Backup

To create a backup, make an HTTP POST request to /admin to a Dgraph Alpha HTTP address and port (default, “localhost:8080”). Like with all /admin endpoints, this is only accessible on the same machine as the Alpha server unless whitelisted for admin operations. You can look at BackupInput given below for all the possible options.

input BackupInput {

		"""
		Destination for the backup: e.g. Minio or S3 bucket.
		"""
		destination: String!

		"""
		Access key credential for the destination.
		"""
		accessKey: String

		"""
		Secret key credential for the destination.
		"""
		secretKey: String

		"""
		AWS session token, if required.
		"""
		sessionToken: String

		"""
		Set to true to allow backing up to S3 or Minio bucket that requires no credentials.
		"""
		anonymous: Boolean

		"""
		Force a full backup instead of an incremental backup.
		"""
		forceFull: Boolean
	}

Execute the following mutation on /admin endpoint using any GraphQL compatible client like Insomnia, GraphQL Playground or GraphiQL.

Backup to Amazon S3

mutation {
  backup(input: {destination: "s3://s3.us-west-2.amazonaws.com/<bucketname>"}) {
    response {
      message
      code
    }
    taskId
  }
}

Backup to Minio

mutation {
  backup(input: {destination: "minio://127.0.0.1:9000/<bucketname>"}) {
    response {
      message
      code
    }
    taskId
  }
}

Backup using a MinIO Gateway

Azure Blob Storage

You can use Azure Blob Storage through the MinIO Azure Gateway. You need to configure a storage account and acontainer to organize the blobs.

For MinIO configuration, you will need to retrieve storage accounts keys. The MinIO Azure Gateway will use MINIO_ACCESS_KEY and MINIO_SECRET_KEY to correspond to Azure Storage Account AccountName and AccountKey.

Once you have the AccountName and AccountKey, you can access Azure Blob Storage locally using one of these methods:

• Run MinIO Azure Gateway using Docker

  docker run --publish 9000:9000 --name gateway \
    --env "MINIO_ACCESS_KEY=<AccountName>" \
    --env "MINIO_SECRET_KEY=<AccountKey>" \
    minio/minio gateway azure
  

• Run MinIO Azure Gateway using the MinIO Binary

  export MINIO_ACCESS_KEY="<AccountName>"
  export MINIO_SECRET_KEY="<AccountKey>"
  minio gateway azure
  

Google Cloud Storage

You can use Google Cloud Storage through the MinIO GCS Gateway. You will need to create storage buckets, create a Service Account key for GCS and get a credentials file. See Create a Service Account key for further information.

Once you have a credentials.json, you can access GCS locally using one of these methods:

• Run MinIO GCS Gateway using Docker

  docker run --publish 9000:9000 --name gateway \
    --volume /path/to/credentials.json:/credentials.json \
    --env "GOOGLE_APPLICATION_CREDENTIALS=/credentials.json" \
    --env "MINIO_ACCESS_KEY=minioaccountname" \
    --env "MINIO_SECRET_KEY=minioaccountkey" \
    minio/minio gateway gcs <project-id>
  

• Run MinIO GCS Gateway using the MinIO Binary

  export GOOGLE_APPLICATION_CREDENTIALS=/path/to/credentials.json
  export MINIO_ACCESS_KEY=minioaccesskey
  export MINIO_SECRET_KEY=miniosecretkey
  minio gateway gcs <project-id>
  

Test Using MinIO Browser

MinIO Gateway comes with an embedded web-based object browser. After using one of the aforementioned methods to run the MinIO Gateway, you can test that MinIO Gateway is running, open a web browser, navigate to http://127.0.0.1:9000, and ensure that the object browser is displayed and can access the remote object storage.

Disabling HTTPS for S3 and Minio backups

By default, Dgraph assumes the destination bucket is using HTTPS. If that is not the case, the backup will fail. To send a backup to a bucket using HTTP (insecure), set the query parameter secure=false with the destination endpoint in the destination field:

mutation {
  backup(input: {destination: "minio://127.0.0.1:9000/<bucketname>?secure=false"}) {
    response {
      message
      code
    }
    taskId
  }
}

Overriding Credentials

The accessKey, secretKey, and sessionToken parameters can be used to override the default credentials. Please note that unless HTTPS is used, the credentials will be transmitted in plain text so use these parameters with discretion. The environment variables should be used by default but these options are there to allow for greater flexibility.

The anonymous parameter can be set to “true” to allow backing up to S3 or MinIO bucket that requires no credentials (i.e a public bucket).

Backup to NFS

mutation {
  backup(input: {destination: "/path/to/local/directory"}) {
    response {
      message
      code
    }
    taskId
  }
}

A local filesystem will work only if all the Alpha servers have access to it (e.g all the Alpha servers are running on the same filesystems as a normal process, not a Docker container). However, an NFS is recommended so that backups work seamlessly across multiple machines and/or containers.

Forcing a Full Backup

By default, an incremental backup will be created if there’s another full backup in the specified location. To create a full backup, set the forceFull field to true in the mutation. Each series of backups can be identified by a unique ID and each backup in the series is assigned a monotonically increasing number. The following section contains more details on how to restore a backup series.

mutation {
  backup(input: {destination: "/path/to/local/directory", forceFull: true}) {
    response {
      message
      code
    }
    taskId
  }
}

Listing Backups

The GraphQL admin interface includes the listBackups endpoint that lists the backups in the given location along with the information included in the manifest.json file. An example of a request to list the backups in the /data/backup location is included below:

query backup() {
	listBackups(input: {location: "/data/backup"}) {
		backupId
		backupNum
		encrypted
		groups {
			groupId
			predicates
		}
		path
		since
		type
	}
}

The listBackups input can contain the following fields. Only the location field is required.

input ListBackupsInput {
	"""
	Destination for the backup: e.g. Minio or S3 bucket.
	"""
	location: String!

	"""
	Access key credential for the destination.
	"""
	accessKey: String

	"""
	Secret key credential for the destination.
	"""
	secretKey: String

	"""
	AWS session token, if required.
	"""
	sessionToken: String

	"""
	Whether the destination doesn't require credentials (e.g. S3 public bucket).
	"""
	anonymous: Boolean
}

The output is of [Manifest] type. The fields inside the Manifest type corresponds to the fields in the manifest.json file.

type Manifest {
	"""
	Unique ID for the backup series.
	"""
	backupId: String

	"""
	Number of this backup within the backup series. The full backup always has a value of one.
	"""
	backupNum: Int

	"""
	Whether this backup was encrypted.
	"""
	encrypted: Boolean

	"""
	List of groups and the predicates they store in this backup.
	"""
	groups: [BackupGroup]

	"""
	Path to the manifest file.
	"""
	path: String

	"""
	The timestamp at which this backup was taken. The next incremental backup will
	start from this timestamp.
	"""
	since: Int

	"""
	The type of backup, either full or incremental.
	"""
	type: String
}

type BackupGroup {
	"""
	The ID of the cluster group.
	"""
	groupId: Int

	"""
	List of predicates assigned to the group.
	"""
	predicates: [String]
}

Automating Backups

You can use the provided endpoint to automate backups, however, there are a few things to keep in mind.

• The requests should go to a single Alpha server. The Alpha server that receives the request is responsible for looking up the location and determining from which point the backup should resume.

• Versions of Dgraph starting with v20.07.1, v20.03.5, and v1.2.7 have a way to block multiple backup requests going to the same Alpha server. For previous versions, keep this in mind and avoid sending multiple requests at once. This is for the same reason as the point above.

• You can have multiple backup series in the same location although the feature still works if you set up a unique location for each series.

Export Backups

The export_backup tool lets you convert a binary backup into an exported folder.

If you need to upgrade between two major Dgraph versions that have incompatible changes, you can use the export_backup tool to apply changes (either to the exported .rdf file or to the schema file), and then import back the dataset into the new Dgraph version.

Using exports instead of binary backups

An example of this use-case would be to migrate existing schemas from Dgraph v1.0 to Dgraph v24.0.1. You need to update the schema file from an export so all predicates of type uid are changed to [uid]. Then use the updated schema when loading data into Dgraph v24.0.1.

For example, for the following schema:

name: string .
friend: uid .

becomes

name: string .
friend: [uid] .

Because you have to do a modification to the schema itself, you need an export. You can use the export_backup tool to convert your binary backup into an export folder.

Binary Backups and Exports folders

A Binary Backup directory has the following structure:

backup
├── dgraph.20210102.204757.509
│   └── r9-g1.backup
├── dgraph.20210104.224757.707
│   └── r9-g1.backup
└── manifest.json

An Export directory has the following structure:

dgraph.r9.u0108.1621
├── g01.gql_schema.gz
├── g01.rdf.gz
└── g01.schema.gz

If you want to do the changes cited above, you need to edit the g01.schema.gz file.

Benefits

With the export_backup tool you get the speed benefit from the binary backups, which are faster than regular exports. So if you have a big dataset, you don’t need to wait a long time until an export is completed. Instead, just take a binary backup and convert it to an export only when needed.

How to use it

Ensure that you have created a binary backup. The directory tree of a binary backup usually looks like this:

backup
├── dgraph.20210104.224757.709
│   └── r9-g1.backup
└── manifest.json

Then run the following command:

dgraph export_backup --location "<location-of-your-binary-backup>" --destination "<destination-of-the-export-dir>"

Once completed you will find your export folder (in this case dgraph.r9.u0108.1621). The tree of the directory should look like this:

dgraph.r9.u0108.1621
├── g01.gql_schema.gz
├── g01.rdf.gz
└── g01.schema.gz

Encrypted Backups

Encrypted backups are an Enterprise feature that are available from v20.03.1 and v1.2.3 and allow you to encrypt your backups and restore them. This documentation describes how to implement encryption into your binary backups.

Starting with v20.07.0, we also added support for Encrypted Backups using encryption keys sitting on Hashicorp Vault.

New Encrypted flag in manifest.json

A new Encrypted flag is added to the manifest.json. This flag indicates if the corresponding binary backup is encrypted or not. To be backward compatible, if this flag is absent, it is presumed that the corresponding backup is not encrypted.

For a series of full and incremental backups, per the current design, we don’t allow the mixing of encrypted and unencrypted backups. As a result, all full and incremental backups in a series must either be encrypted fully or not at all. This flag helps with checking this restriction.

AES And Chaining with Gzip

If encryption is turned on an Alpha server, then we use the configured encryption key. The key size (16, 24, 32 bytes) determines AES-128/192/256 cipher chosen. We use the AES CTR mode. Currently, the binary backup is already gzipped. With encryption, we will encrypt the gzipped data.

During backup: the 16 bytes IV is prepended to the Cipher-text data after encryption.

Backup

Backup is an online tool, meaning it is available when Dgraph Alpha server is running. For encrypted backups, the Dgraph Alpha server must be configured with the --encryption key-file=value. Starting with v20.07.0, the Dgraph Alpha server can alternatively be configured to interface with a Hashicorp Vault server to obtain keys.

Online restore

To restore from a backup to a live cluster, execute a mutation on the /admin endpoint with the following format:

mutation{
  restore(input:{
    location: "/path/to/backup/directory",
    backupId: "id_of_backup_to_restore"
  }){
    message
    code
  }
}

Online restores only require you to send this request. The UID and timestamp leases are updated accordingly. The latest backup to be restored should contain the same number of groups in its manifest.json file as the cluster to which it is being restored.

Online restore can be performed from Amazon S3 / Minio or from a local directory. Below is the documentation for the fields inside RestoreInput that can be passed into the mutation.

input RestoreInput {

		"""
		Destination for the backup: e.g. Minio or S3 bucket.
		"""
		location: String!

		"""
		Backup ID of the backup series to restore. This ID is included in the manifest.json file.
		If missing, it defaults to the latest series.
		"""
		backupId: String

		"""
		Number of the backup within the backup series to be restored. Backups with a greater value
		will be ignored. If the value is zero or is missing, the entire series will be restored.
		"""
		backupNum: Int

		"""
		Path to the key file needed to decrypt the backup. This file should be accessible
		by all Alpha servers in the group. The backup will be written using the encryption key
		with which the cluster was started, which might be different than this key.
		"""
		encryptionKeyFile: String

		"""
		Vault server address where the key is stored. This server must be accessible
		by all Alpha servers in the group. Default "http://localhost:8200".
		"""
		vaultAddr: String

		"""
		Path to the Vault RoleID file.
		"""
		vaultRoleIDFile: String

		"""
		Path to the Vault SecretID file.
		"""
		vaultSecretIDFile: String

		"""
		Vault kv store path where the key lives. Default "secret/data/dgraph".
		"""
		vaultPath: String

		"""
		Vault kv store field whose value is the key. Default "enc_key".
		"""
		vaultField: String

		"""
		Vault kv store field's format. Must be "base64" or "raw". Default "base64".
		"""
		vaultFormat: String

		"""
		Access key credential for the destination.
		"""
		accessKey: String

		"""
		Secret key credential for the destination.
		"""
		secretKey: String

		"""
		AWS session token, if required.
		"""
		sessionToken: String

		"""
		Set to true to allow backing up to S3 or Minio bucket that requires no credentials.
		"""
		anonymous: Boolean

		"""
		All the backups with num >= incrementalFrom will be restored.
		"""
		incrementalFrom: Int

		"""
		If `isPartial` is set to true then the cluster is kept in draining mode after
		restore to ensure that the database is not corrupted by any mutations or tablet moves in
		between two restores.
		"""
		isPartial: Boolean

}

Restore requests returns immediately without waiting for the operation to finish.

Incremental Restore

You can use incremental restore to restore a set of incremental backups on a cluster with a part of the backup already restored. The system does not accept any mutations made between a normal restore and an incremental restore, because the cluster is in the draining mode. When the cluster is in a draining mode only an admin request to bring the cluster back to normal mode is accepted.

Note: Before you start an incremental restore ensure that you set isPartial to true in your normal restore.

To incrementally restore from a backup to a live cluster, execute a mutation on the /admin endpoint with the following format:

mutation{
  restore(input:{
    incrementalFrom:"incremental_backup_from",
    location: "/path/to/backup/directory",
    backupId: "id_of_backup_to_restore"'
  }){
    message
    code
  }
}

Namespace Aware Restore

You can use namespace-aware restore to restore a single namespace from a backup that contains multiple namespaces. The created restore will be available in the default namespace. For example, if you restore namespace 2 using the restoreTenant API, then after the restore operation is completed, the cluster will have only the default namespace, and it will contain the data from namespace 2. Namespace aware restore supports incremental restore.

To restore from a backup to a live cluster, execute a mutation on the /admin endpoint with the following format:

mutation {
  restoreTenant(
    input: {
      restoreInput: {
        incrementalFrom: "incremental_backup_from"
        location: "/path/to/backup/directory"
        backupId: "id_of_backup_to_restore"
      }
      fromNamespace: namespaceToBeRestored
    }
  ) {
    message
    code
  }
}

Documentation of restoreTenant inputs

input RestoreTenantInput {
	"""
	restoreInput contains fields that are required for the restore operation,
	i.e., location, backupId, and backupNum
	"""
	restoreInput: RestoreInput

	"""
	fromNamespace is the namespace of the tenant that needs to be restored into namespace 0 of the new cluster.
	"""
	fromNamespace: Int!
}

Offline restore (DEPRECATED)

The restore utility is now a standalone tool. A new flag, --encryption key-file=value, is now part of the restore utility, so you can use it to decrypt the backup. The file specified using this flag must contain the same key that was used for encryption during backup. Alternatively, starting with v20.07.0, the vault superflag can be used to restore a backup.

You can use the dgraph restore command to restore the postings directory from a previously-created backup to a directory in the local filesystem. This command restores a backup to a new Dgraph cluster, so it is not designed to restore a backup to a Dgraph cluster that is currently live. During a restore operation, a new Dgraph Zero server might run to fully restore the backup state.

You can use the --location (-l) flag to specify a source URI with Dgraph backup objects. This URI supports all the schemes used for backup.

You can use the --postings (-p) flag to set the directory where restored posting directories are saved. This directory contains a posting directory for each group in the restored backup.

You can use the --zero (-z) flag to specify a Dgraph Zero server address to update the start timestamp and UID lease using the restored version. If no Dgraph Zero server address is passed, the command will complain unless you set the value of the --force_zero flag to false. If do not pass a zero value to this command, you need to manually update the timestamp and UID lease using the Dgraph Zero server’s HTTP ‘assign’ endpoint. The updated values should be those that are printed near the end of the command’s output.

You use the --backup_id optional flag to specify the ID of the backup series to restore. A backup series consists of a full backup and all of the incremental backups built on top of it. Each time a new full backup is created, a new backup series with a different ID is started. The backup series ID is stored in each manifest.json file stored in each backup folder.

You use the --encryption key-file=value flag in cases where you took the backup in an encrypted cluster. The string for this flag must point to the location of the same key file used to run the cluster.

You use the --vault superflag to specify the Hashicorp Vault server address (addr), role id (role-id-file), secret id (secret-id-file) and the field that contains the encryption key (enc-field) that was used to encrypt the backup.

The restore feature creates a cluster with as many groups as the original cluster had at the time of the last backup. For each group, dgraph restore creates a posting directory (p<N>) that corresponds to the backup group ID. For example, a backup for Dgraph Alpha group 2 would have the name .../r32-g2.backup and would be loaded to posting directory p2.

After running the restore command, the directories inside the postings directory need to be manually copied over to the machines/containers running the Dgraph Alpha servers before running the dgraph alpha command. For example, in a database cluster with two Dgraph Alpha groups and one replica each, p1 needs to be moved to the location of the first Dgraph Alpha node and p2 needs to be moved to the location of the second Dgraph Alpha node.

By default, Dgraph will look for a posting directory with the name p, so make sure to rename the directories after moving them. You can also use the -p option of the dgraph alpha command to specify a different path from the default.

Restore from Amazon S3

dgraph restore --postings "/var/db/dgraph" --location "s3://s3.<region>.amazonaws.com/<bucketname>"

Restore from MinIO

dgraph restore --postings "/var/db/dgraph" --location "minio://127.0.0.1:9000/<bucketname>"

Restore from Local Directory or NFS

dgraph restore --postings "/var/db/dgraph" --location "/var/backups/dgraph"

Restore and Update Timestamp

Specify the Zero server address and port for the new cluster with --zero/-z to update the timestamp.

dgraph restore --postings "/var/db/dgraph" --location "/var/backups/dgraph" --zero "localhost:5080"