Creating a Backup
Choosing the backup type #
As described in Backup types, you choose to run a full Backup (the default) or an incremental Backup.
Full backups will use the backup engine chosen in the tablet's configuration. Incremental backups will always copy MySQL's binary logs, irrespective of the configured backup engine.
Using xtrabackup #
The default backup implementation is builtin
, however we strongly recommend using the xtrabackup
engine as it is more robust and allows for non-blocking backups. Restores will always be done with whichever engine was used to create the backup.
Prerequisites #
A compatible version of xtrabackup and xbstream, if needed, must be present in your $PATH
prior to running the Backup[Shard]
command.
Supported Versions #
Configuration #
To use xtrabackup
with Vtbackup, VTTablet or Vtctld, the following flags must be set.
Required flags:
--backup_engine_implementation=xtrabackup
--xtrabackup_user string
- The user that xtrabackup will use to connect to the database server. This user must have the necessary privileges.
- This user will need to be authorized to connect to mysql locally without a password using auth_socket.
Additionally required for MySQL 8.0:
--xtrabackup_stream_mode=xbstream
Common Errors and Resolutions #
No xtrabackup User passed to vttablet:
E0310 08:15:45.336083 197442 main.go:72] remote error: rpc error: code = Unknown desc = TabletManager.Backup on zone1-0000000102 error: xtrabackupUser must be specified.: xtrabackupUser must be specified
Fix: Set the vtctld and vttablet flag --xtrabackup_user
xtrabackup binary not found in $PATH:
E0310 08:22:22.260044 200147 main.go:72] remote error: rpc error: code = Unknown desc = TabletManager.Backup on zone1-0000000102 error: unable to start backup: exec: "xtrabackup": executable file not found in $PATH: unable to start backup: exec: "xtrabackup": executable file not found in $PATH
Fixes:
* Ensure the xtrabackup binary is in the $PATH for the $USER running vttablet
* Alternatively, set --xtrabackup_root_path on vttablet provide path to xtrabackup/xbstream binaries via vtctld and vttablet flags
Tar format no longer supported in 8.0:
I0310 12:34:47.900363 211809 backup.go:163] I0310 20:34:47.900004 xtrabackupengine.go:310] xtrabackup stderr: Invalid --stream argument: tar
Streaming in tar format is no longer supported in 8.0; use xbstream instead
Fix: Set the --xtrabackup_stream_mode
flag to to xbstream on vttablets and vtctlds
Unsupported mysql server version:
I0310 12:49:32.279729 215835 backup.go:163] I0310 20:49:32.279435 xtrabackupengine.go:310] xtrabackup stderr: Error: Unsupported server version 8.0.23-0ubuntu0.20.04.1.
I0310 12:49:32.279773 215835 backup.go:163] I0310 20:49:32.279485 xtrabackupengine.go:310] xtrabackup stderr: Please upgrade PXB, if a new version is available. To continue with risk, use the option --no-server-version-check.
To continue with risk: Set --xtrabackup_backup_flags=--no-server-version-check
. Note this occurs when your MySQL server version is technically unsupported by xtrabackup
.
Using mysqlshell (EXPERIMENTAL) #
An alternative option to physical backups (xtrabackup
and builtin
) is to take logical backups.
Vitess supports MySQL Shell to dump and compress databases in a easy to recover plain text format
using tab-separated values. Physical backups have the potential to suffer from silent corruption
if pages are not often accessed and a logical backup will read and dump all the data on a tablet,
flushing out any issues that could spread on a physical backup.
Prerequisites #
The mysqlsh
binary should be present as it is used to handle the dump/load process. It is included
in the MySQL distribution and also included in vitess provided containers.
Supported Versions #
Note that MySQL >=8.0.21 is recommended, but not required. It allows you yo disable redo logs and the double write buffer to make restores faster.
How it works #
This engine differs from the other engines in vitess in the sense that they usually manage the entire backup workflow, including uploading the generated data to the respective backend.
Because MySQL Shell doesn't provide a single stream and has many optimisations for dumping data faster, it outputs data into a target directory in many files (often thousands). If you are using an object storage like S3, Google Cloud Storage, Ceph, etc, MySQL Shell will need additional configuration as it needs direct access to those bakend storages, while the other engines rely on Vitess to upload the backup output.
The backup location configured for Vitess will only be used to store the MANIFEST for the metadata, which will point to the actual location of the backup in your preferred destination. This will also be used during the restore process, as Vitess needs to pass the location of the backup to MySQL Shell to initiate the restore process.
Configuration #
mysqlshell
's configuration requires understanding some of the features of MySQL Shell. It is
recommended users are familiar with the dump/load options as well as the options related to their backend storage included in these links.
Required flags:
--backup_engine_implementation=mysqlshell
--mysql-shell-backup-location string
- This stores the location where MySQL Shell will store the backups. As mentioned above, this is different than the Vitess backup location.
--mysql-shell-flags="-js"
-js
is required to be in the mysql shell flags otherwise we can't run the necessary commands.- There you can also pass other seetings that you would add to connect to your database, including hostname, username and password.
- if you require a different way to pass credentials, you can look into the MySQL Shell's
--credential-store-helper
flag as that allows you to plug your own facility to fetch credentials without passing them via the command line
Optional flags:
--mysql-shell-dump-flags="{\"threads\": 4}"
- you can control how many threads and other options with this flag. See the MySQL docs for other options
--mysql-shell-load-flags="{\"updateGtidSet\": \"replace\", \"progressFile\": \"\", \"skipBinlog\": true}"
updateGtidSet
must be set to"replace"
so MySQL Shell updates the executed GTID at the end of the restoreprogressFile
must be empty indicate we are starting a new backup and not continuing an existing oneskipBinlog
must betrue
so we are not logging the statements to binary logs during the restoreloadUsers
defaults tofalse
, but it is not required. This is ideal if you rely on Vitess managing credentials as we will delete users by default during the dump process. If you perform your own credential management you might want to disable this.- these are all defaults in Vitess, but if you define this flag it must include the values above. other options can be found in the MySQL docs
--mysql-shell-should-drain=false
- This flag controls if the tablet should be drained during backup or not. If true, it will transition to
BACKUP
type and will return to the original type once done, otherwise it will continue to serve during the backup (similar to thextrabackup
engine)
- This flag controls if the tablet should be drained during backup or not. If true, it will transition to
--mysql-shell-speedup-restore=false
- If you are using MySQL >=8.0.21, this allows you to disable redo logs and the double write buffer for the duration of the restore, increasing the restore speed.
Limitations and Caveats to be aware #
Since MySQL needs to be running for the entire backup/restore process, we can't just replace all the files like we do with the physical backups. To account for this, Vitess handles things differently:
Databases #
All databases will be backed up, with the exception of information_schema
, mysql
, ndbinfo
, performance_schema
and sys
. This is a MySQL Shell limitation. During the restore process, Vitess will drop all databases in the host ahead of the restore before starting the mysqlsh
load process.
Users #
Since user grants are stored in the mysql
database, they are not backed up like the other databases. MySQL Shell works around that saving users separately and allow the user to pass the loadUsers: true
option during the load process to restore users to the state they were during the backup.
Internal users (mysql.sys@localhost
, mysql.session@localhost
, mysql.infoschema@localhost
) are not restored, as well as the current user logged in to MySQL. It is also possible to exclude some users from being imported by using adjusting your --mysql-shell-load-flags
if necessary.
Because MySQL Shell will fail the restore if any user already exists, Vitess will drop all users in MySQL (except the ones noted above) before the restore if you specify loadUsers: true
.
Create a full backup with vtctl #
Run the following vtctl command to create a backup:
vtctldclient --server=<vtctld_host>:<vtctld_port> Backup [--upgrade-safe=false] <tablet-alias>
If the engine is builtin
, replication will be stopped prior to shutting down mysqld for the backup.
If the engine is xtrabackup
, the tablet can continue to serve traffic while the backup is running.
Run the following vtctl command to backup a specific shard:
vtctldclient --server=<vtctld_host>:<vtctld_port> BackupShard [--allow_primary=false] [--upgrade-safe=false] <keyspace/shard>
Create an incremental backup with vtctl #
An incremental backup requires additional information: the point from which to start the backup. An incremental backup is taken by supplying --incremental-from-pos
to the Backup
or BackupShard
command. The argument may either indicate:
- A valid position.
- A name of a successful backup.
- Or, the value
auto
.
vtctldclient Backup --incremental-from-pos="MySQL56/0d7aaca6-1666-11ee-aeaf-0a43f95f28a3:1-53" zone1-0000000102
vtctldclient Backup --incremental-from-pos="0d7aaca6-1666-11ee-aeaf-0a43f95f28a3:1-53" zone1-0000000102
vtctldclient Backup --incremental-from-pos="2024-01-10.062022.zone1-0000000101 commerce/0" zone1-0000000102
vtctldclient Backup --incremental-from-pos="auto" zone1-0000000102
vtctldclient BackupShard --incremental-from-pos=auto commerce/0
When --incremental-from-pos
supplies a position, you may choose to use or to omit the MySQL56/
prefix (which you can find in the backup manifest's Position).
When --incremental-from-pos
indicates a backup name, that must be a successfully completed, existing backup. It may be either a full or an incremental backup.
When --incremental-from-pos="auto"
, Vitess chooses the position of the last successful backup as the starting point for the incremental backup. This is a convenient way to ensure a sequence of contiguous incremental backups.
An incremental backup backs up one or more MySQL binary log files. These binary log files may begin with the requested position, or with an earlier position. They will necessarily include the requested position. When the incremental backup begins, Vitess rotates the MySQL binary logs on the tablet, so that it does not back up an active log file.
If Vitess finds that the database made no writes since the requested backup/position, then the incremental backup is deemed empty and produces no artifacts, essentially becoming a no-op. The Backup/BackupShard
command exits with success code, but there is no MANIFEST
file created and no backup name.
An incremental backup fails when it is unable to find binary log files that covers the requested position. This can happen if the binary logs are purged earlier than the incremental backup was taken. It essentially means there's a gap in the changelog events. Note that while on one tablet the binary logs may be missing, another tablet may still have binary logs that cover the requested position.
Backing up Topology Server #
The Topology Server stores metadata (and not tablet data). It is recommended to create a backup using the method described by the underlying plugin: