Restore a backup¶

To restore a backup, use the pbm restore command supplying the backup name from which you intend to restore. Percona Backup for MongoDB identifies the type of the backup (physical, logical or incremental) and restores the database up to the restore_to_time timestamp (available in pbm list output starting with version 1.4.0).

Considerations¶

Logical Physical Incremental Snapshot-based

While the restore is running, prevent clients from accessing the database. The data will naturally be incomplete while the restore is in progress, and writes the clients make cause the final restored data to differ from the backed-up data.
For versions 2.3.1 and earlier, disable Point-in-time recovery before running pbm restore. This is because Point-in-Time recovery oplog slicing and restore are incompatible operations and cannot be run together.
Backups made with Percona Backup for MongoDB prior to v1.5.0 are incompatible for restore with Percona Backup for MongoDB v1.5.0 and later. This is because processing of system collections Users and Roles has changed: in v1.5.0, Users and Roles are copied to temporary collection during backup and must be present in the backup during restore. In earlier versions of Percona Backup for MongoDB, Users and Roles are copied to a temporary collection during restore. Therefore, restoring from these backups with Percona Backup for MongoDB v1.5.0 isn’t possible.

The recommended approach is to make a fresh backup after upgrading Percona Backup for MongoDB to version 1.5.0.
For versions earlier than 1.x, Percona Backup for MongoDB performs a full all-databases, all collections restore and does not offer an option to restore only a subset of collections in the backup, as MongoDB’s mongodump tool does.
Starting with versions 1.x, Percona Backup for MongoDB replicates mongodump’s behavior to only drop collections in the backup. It does not drop collections that are created new after the time of the backup and before the restore. Run a db.dropDatabase() manually in all non-system databases (these are all databases except “local”, “config” and “admin”) before running pbm restore if you want to guarantee that the post-restore database only includes collections that are in the backup.

Disable point-in-time recovery. A restore and point-in-time recovery oplog slicing are incompatible operations and cannot be run simultaneously.

$ pbm config --set pitr.enabled=false

The Percona Server for MongoDB version for both backup and restore data must be within the same major release.
Make sure all nodes in the cluster are healthy (i.e. either PRIMARY or SECONDARY). Each pbm-agent needs to be able to connect to its local node and run queries in order to perform the restore.
For PBM versions before 2.1.0, physical restores are not supported for deployments with arbiter nodes.

The Percona Server for MongoDB version for both backup and restore data must be within the same major release.
Incremental backups made with PBM before PBM 2.1.0 are incompatible for restore with PBM 2.1.0 and onwards.
Physical restores are not supported for deployments with arbiter nodes.

Supported only for full physical backups
Available only if you run Percona Server for MongoDB in your environment as PBM uses the $backupCursor and $backupCursorExtended aggregation stages .

Before you start¶

Logical Physical Selective Incremental Snapshot-based

Stop the balancer and disable chunks autosplit. To verify that both are disabled, run the following command:
```
> sh.status() 
```
You should see the following output
```
autosplit:
        Currently enabled: no
  balancer:
        Currently enabled: no
        Currently running: no
```
Shut down all mongos nodes to stop clients from accessing the database while restore is in progress. This ensures that the final restored data doesn’t differ from the backed-up data.
Shut down all pmm-agent and other clients that can do the write operations to the database. This is required to ensure data consistency after the restore.
For PBM version 2.3.1 and earlier, manually disable point-in-time recovery if it is enabled. To learn more about point-in-time recovery, see Point-in-time recovery.

Shut down all mongos nodes as the database won’t be available while the restore is in progress.
Shut down all pmm-agent and other clients that can do the write operations to the database. This is required to ensure data consistency after the restore.
Stop the arbiter nodes manually since there’s no pbm-agent on these nodes to do that automatically.

You can restore a specific database or a collection either from a full or a selective backup. Read about known limitations of selective restores.

Before you start, shut down all mongos nodes, pmm-agent processes and clients that can do writes to the database as it won’t be available while the restore is in progress.

Shut down all mongos nodes. If you have set up the automatic restart of the database, disable it.
Stop the arbiter nodes manually since there’s no pbm-agent on these nodes to do that automatically.

Restore a database¶

Logical Physical Selective Incremental Snapshot-based

List the backups to restore from
```
$ pbm list
```
Restore from a desired backup. Replace the <backup_name> with the desired backup in the following command:

$ pbm restore <backup_name>

Note that you can restore a sharded backup only into a sharded environment. It can be your existing cluster or a new one. To learn how to restore a backup into a new environment, see Restoring a backup into a new environment.

Post-restore steps

After a cluster’s restore is complete, do the following:

Start the balancer and all mongos nodes to reload the sharding metadata.
We recommend to make a fresh backup to serve as the new base for future restores.
Enable point-in-time recovery if required.

Adjust memory consumption¶

Starting with version 1.3.2, Percona Backup for MongoDB config includes the restore options to adjust the memory consumption by the pbm-agent in environments with tight memory bounds. This allows preventing out of memory errors during the restore operation.

restore:
  batchSize: 500
  numInsertionWorkers: 10

The default values were adjusted to fit the setups with the memory allocation of 1GB and less for the agent.

Note

The lower the values, the less memory is allocated for the restore. However, the performance decreases too.

Restore from a logical backup made on previous major version of Percona Server for MongoDB¶

In some cases you may need to restore from a backup made on previous major version of Percona Server for MongoDB. To make this happen, Feature Compatibility Version (FCV) values in both backup and the destination environment must match.

Starting with version 2.1.0, Percona Backup for MongoDB stores the FCV value in the backup metadata. If it doesn’t match the FCV value on the destination environment, you see the warning in the pbm status output so that you can manually adjust it before the restore.

2023-04-10T10:48:54Z 302.80KB <logical> [ERROR: backup FCV "6.0" is incompatible with the running mongo FCV "5.0"] [2023-04-10T10:49:14Z]
2023-04-10T08:40:10Z 172.25KB <logical> [ERROR: backup mongo version "6.0.5-4" is incompatible with the running mongo version "5.0.15-13"] [2023-04-10T08:40:28Z]

The following example illustrates the restore from a backup made on Percona Server for MongoDB 4.4 on Percona Server for MongoDB 5.0.

Check the FCV value for the backup

$ pbm status

Sample output:

Snapshots:
2023-04-10T10:51:28Z 530.73KB <logical> [ERROR: backup FCV "4.4" is incompatible with the running mongo FCV "5.0"] [2023-04-10T10:51:44Z]

Set the Feature Compatibility Version value to 4.4

> db.adminCommand( { setFeatureCompatibilityVersion: "4.4" } )

Restore the database
```
$ pbm restore 2023-04-10T10:51:28Z
```

Set the Feature Compatibility Version value to 5.0

> db.adminCommand( { setFeatureCompatibilityVersion: "5.0" } )

List the backups
```
$ pbm list
```
Make a restore
```
$ pbm restore <backup_name>
```

During the physical restore, pbm-agent processes stop the mongod nodes, clean up the data directory and copy the data from the storage onto every node. During this process, the database is restarted several times.

You can track the restore progress using the pbm describe-restore command. Don’t run any other commands since they may interrupt the restore flow and cause the issues with the database.

Post-restore steps

After the restore is complete, do the following:

Restart all mongod nodes.
Note

You may see the following message in the mongod logs after the cluster restart:
```
"s":"I",  "c":"CONTROL",  "id":20712,   "ctx":"LogicalSessionCacheReap","msg":"Sessions collection is not set up; waiting until next sessions reap interval","attr":{"error":"NamespaceNotFound: config.system.sessions does not exist"}}}}
```
This is expected behavior of periodic checks upon the database start. During the restore, the config.system.sessions collection is dropped but Percona Server for MongoDB recreates it eventually. It is a normal procedure. No action is required from your end.
Restart all pbm-agents
Run the following command to resync the backup list with the storage:
```
$ pbm config --force-resync
```
Start the balancer and start mongos nodes.
We recommend to make a fresh backup to serve as the new base for future restores.
Enable point-in-time recovery if required.

Define a `mongod` binary location¶

Version added: 2.0.4

During physical restores, Percona Backup for MongoDB performs several restarts of the database. By default, it uses the location of the mongod binaries from the $PATH variable to access the database. If you have defined the custom path to the mongod binaries, make Percona Backup for MongoDB aware of it by specifying this path in the configuration file:

restore:
    mongodLocation: /path/to/mongod

If you have different paths to mongod binaries on every node of your cluster / replica set, use the mongodLocationMap option to specify your custom paths for each node.

restore:
    mongodLocationMap:
       "node01:27017": /path/to/mongod
       "node03:27017": /another/path/to/mongod

Parallel data download¶

Version added: 2.1.0

Percona Backup for MongoDB downloads data chunks from the S3 storage concurrently during physical restore. Read more about benchmarking results in the Speeding up MongoDB restores in PBM blog post by Andrew Pogrebnoi.

Here’s how it works:

During the physical restore, Percona Backup for MongoDB starts the workers. The number of workers equals to the number of CPU cores by default. Each worker has a memory buffer allocated for it. The buffer is split into spans for the size of the data chunk. The worker acquires the span to download a data chunk and stores it into the buffer. When the buffer is full, the worker waits for the free span to continue the download.

You can fine-tune the parallel download depending on your hardware resources and database load. Edit the PBM configuration file and specify the following settings:

restore:
   numDownloadWorkers: <int>
   maxDownloadBufferMb: <int>
   downloadChunkMb: 32

numDownloadWorkers - the number of workers to download data from the storage. By default, it equals to the number of CPU cores
maxDownloadBufferMb - the maximum size of memory buffer to store the downloaded data chunks for decompression and ordering. It is calculated as numDownloadWorkers * downloadChunkMb * 16
downloadChunkMb is the size of the data chunk to download (by default, 32 MB)

List the backups
```
$ pbm list
```

Run the pbm restore command in the format:

$ pbm restore <backup_name> --ns <database.collection>

You can specify several namespaces as a comma-separated list for the --ns flag: <db1.col1>, <db2.*>.

During the restore, Percona Backup for MongoDB retrieves the file for the specified database / collection and restores it.

Restore with users and roles¶

To restore a custom database with users and roles from a full backup, add the --with-users-and-roles flag to the pbm restore command:

$ pbm restore <backup_name> --ns <database.*> --with-users-and-roles

Restore a collection under a different name¶

You can restore a specific collection under a different name alongside the current collection. This is useful when you troubleshoot database issues and need to compare the data in both collections to identify the root of the issue.

Note that in version 2.8.0 you can restore a single collection and this collection must be unsharded.

To restore a collection, pass the collection name from the backup for the --ns-from flag and the new name for the --ns-to flag:

$ pbm restore <backup_name> --ns-from <database.collection> --ns-to <database.collection_new>

The new collection has the same data and indexes as the source collection. You must provide a unique name for the collection you restore, otherwise the restore fails.

You can restore a collection under a new name up to the specified time. Instead of the backup name, specify the timestamp, the source collection name and the new name as follows:

$ pbm restore --time=<timestamp> --ns-from <database.collection> --ns-to <database.collection_new>

Restore flow from an incremental backup is the same as the restore from a full physical backup: specify the backup name for the pbm restore command:

$ pbm restore 2022-11-25T14:13:43Z

Percona Backup for MongoDB recognizes the backup type, finds the base incremental backup, restores the data from it and then restores the modified data from applicable incremental backups.

After the restore is complete, do the following:

Restart all mongod nodes and pbm-agents.
Note

You may see the following message in the mongod logs after the cluster restart:
```
"s":"I",  "c":"CONTROL",  "id":20712,   "ctx":"LogicalSessionCacheReap","msg":"Sessions collection is not set up; waiting until next sessions reap interval","attr":{"error":"NamespaceNotFound: config.system.sessions does not exist"}}}}
```
This is expected behavior of periodic checks upon the database start. During the restore, the config.system.sessions collection is dropped but Percona Server for MongoDB recreates it eventually. It is a normal procedure. No action is required from your end.
Resync the backup list from the storage.
Start the balancer and the mongos node.
As the general recommendation, make a new base backup to renew the starting point for subsequent incremental backups.

See snapshot-based backups.

Restoring a backup into a new environment¶

To restore a backup from one environment to another, ensure the following:

Percona Backup for MongoDB configuration in the new environment must point to the same remote storage that is defined for the original environment, including the authentication credentials if it is an object store. Once you run pbm list and see the backups made from the original environment, then you can run the pbm restore command.
Don’t run pbm backup from the new environment while Percona Backup for MongoDB configuration is pointing to the remote storage location of the original environment.

Restoring into a cluster / replica set with a different name¶

Starting with version 1.8.0, you can restore logical backups into a new environment that has the same or more number of shards and these shards have different replica set names. Starting with version 2.2.0, you can restore environments that have custom shard names.

Starting with version 2.2.0, you can restore physical and incremental physical backups into a new environment with a different replica set names. Note that the number of shards must be the same as in the environment where the you made the backup.

To restore data to the environment with different replica set names, configure the name mapping between the source and target environments. You can either set the PBM_REPLSET_REMAPPING environment variable for pbm CLI or use the --replset-remapping flag for PBM commands. The mapping format is <rsTarget>=<rsSource>.

Important

Configure replica set name mapping for all shards in your cluster. Otherwise, Percona Backup for MongoDB attempts to restore the unspecified shard to the target shard with the same name. If there is no shard with such name or it is already mapped to another source shard, the restore fails.

Configure the replica set name mapping:

Using the environment variable for pbm CLI in your shell Using the command line

$ export PBM_REPLSET_REMAPPING="rsX=rsA,rsY=rsB"

$ pbm restore <timestamp> --replset-remapping="rsX=rsA,rsY=rsB"

The --replset-remapping flag is available for the following commands: pbm restore, pbm list, pbm status, pbm oplog-replay.

Note

Follow the post-restore steps on the new environment after the restore is complete.

This ability to restore data to clusters with different replica set names and the number of shards extends the set of environments compatible for the restore.

Next steps¶

Point-in-time recovery

Useful links¶

View restore progress

Get expert help¶

If you need assistance, visit the community forum for comprehensive and free database knowledge, or contact our Percona Database Experts for professional support and services.

Community Forum Get a Percona Expert Join Percona Squad

Last update: November 19, 2024
Created: November 19, 2024