Features
- Add support for Postgres 17+ incremental backups. This major feature is
composed of several small changes:
- Add `--incremental` command-line option to `barman backup` command. This is
used to specify the parent backup when taking an incremental backup. The
parent can be either a full backup or another incremental backup.
- Add `latest-full` shortcut backup ID. Along with `latest`, this can be used
as a shortcut to select the parent backup for an incremental backup. While
`latest` takes the latest backup independently if it is full or incremental,
`latest-full` takes the latest full backup.
- `barman keep` command can only be applied to full backups when
`backup_method = postgres`. If a full backup has incremental backups that
depend on it, all of the incrementals are also kept by Barman.
- When deleting a backup all the incremental backups depending on it, if any,
are also removed.
- Retention policies do not take incremental backups into consideration. As
incremental backups cannot be recovered without having the complete chain of
backups available up to the full backup, only full backups account for
retention policies.
- `barman recover` needs to combine the full backup with the chain of incremental
backups when recovering. The new CLI option `--local-staging-path`, and the
corresponding `local_staging_path` configuration option, are used to specify
the path in the Barman host where the backups will be combined when recovering
an incremental backup.
- Changes to `barman show-backup` output:
- Add the “Estimated cluster size” field. It's useful to have an estimation
of the data directory size of a cluster when restoring a backup. It’s
particularly useful when recovering compressed backups or incremental
backups, situations where the size of the backup doesn’t reflect the size of the
data directory in Postgres. In JSON format, this is stored as
`cluster_size`.
- Add the “WAL summarizer” field. This field shows if `summarize_wal` was
enabled in Postgres at the time the backup was taken. In JSON format, this
is stored as `server_information.summarize_wal`. This field is omitted for
Postgres 16 and older.
- Add “Data checksums” field. This shows if `data_checkums` was enabled in
Postgres at the time the backup was taken. In JSON format, this is stored as
`server_information.data_checksums`.
- Add the “Backup method” field. This shows the backup method used for this
backup. In JSON format, this is stored as
`base_backup_information.backup_method`.
- Rename the field “Disk Usage” as “Backup Size”. The latter provides a more
comprehensive name which represents the size of the backup in the Barman
host. The JSON field under `base_backup_information` was also renamed from
`disk_usage` to `backup_size`.
- Add the “WAL size” field. This shows the size of the WALs required by the
backup. In JSON format, this is stored as
`base_backup_information.wal_size`.
- Refactor the field “Incremental size”. It is now named “Resources saving”
and it now shows an estimation of resources saved when taking incremental
backups with `rsync` or `pg_basebackup`. It compares the backup size with
the estimated cluster size to estimate the amount of disk and network
resources that were saved by taking an incremental backup. In JSON format,
the field was renamed from `incremental_size` to `resource_savings` under `base_backup_information`.
- Add the `system_id` field to the JSON document. This field contains the
system identifier of Postgres. It was present in console format, but was
missing in JSON format.
- Add fields related with Postgres incremental backups:
- “Backup type”: indicates if the Postgres backup is full or incremental. In
JSON format, this is stored as `backup_type` under `base_backup_information`.
- “Root backup”: the ID of the full backup that is the root of a chain of
one or more incremental backups. In JSON format, this is stored as
`catalog_information.root_backup_id`.
- “Parent backup”: the ID of the full or incremental backup from which this
incremental backup was taken. In JSON format, this is stored as
`catalog_information.parent_backup_id`.
- “Children Backup(s)”: the IDs of the incremental backups that were taken
with this backup as the parent. In JSON format, this is stored as
`catalog_information.children_backup_ids`.
- “Backup chain size”: the number of backups in the chain from this
incremental backup up to the root backup. In JSON format, this is
stored as `catalog_information.chain_size`.
- Changes to `barman list-backup` output:
- It now includes the backup type in the JSON output, which can be either
`rsync` for backups taken with rsync, `full` or `incremental` for backups
taken with `pg_basebackup`, or `snapshot` for cloud snapshots. When printing
to the console the backup type is represented by the corresponding labels
`R`, `F`, `I` or `S`.
- Remove tablespaces information from the output. That was bloating the
output. Tablespaces information can still be found in the output of
`barman show-backup`.
- Always set a timestamp with a time zone when configuring
`recovery_target_time` through `barman recover`. Previously, if no time zone
was explicitly set through `--target-time`, Barman would configure
`recovery_target_time` without a time zone in Postgres. Without a time zone,
Postgres would assume whatever is configured through `timezone` GUC in
Postgres. From now on Barman will issue a warning and configure
`recovery_target_time` with the time zone of the Barman host if no time zone
is set by the user through `--target-time` option.
- When recovering a backup with the “no get wal” approach and `--target-lsn` is set,
copy only the WAL files required to reach the configured target. Previously
Barman would copy all the WAL files from its archive to Postgres.
- When recovering a backup with the “no get wal” approach and `--target-immediate`
is set, copy only the WAL files required to reach the consistent point.
Previously Barman would copy all the WAL files from its archive to Postgres.
- `barman-wal-restore` now moves WALs from the spool directory to `pg_wal`
instead of copying them. This can improve performance if the spool directory
and the `pg_wal` directory are in the same partition.
- `barman check-backup` now shows the reason why a backup was marked as `FAILED`
in the output and logs. Previously for a user to know why the backup was
marked as `FAILED`, they would need to run `barman show-backup` command.
- Add configuration option `aws_await_snapshots_timeout` and the corresponding
`--aws-await-snapshots-timeout` command-line option on `barman-cloud-backup`.
This specifies the timeout in seconds to wait for snapshot backups to reach
the completed state.
- Add a keep-alive mechanism to rsync-based backups. Previously the Postgres
session created by Barman to run `pg_backup_start()` and `pg_backup_stop()` would
stay idle for as long as the base backup copy would take. That could lead to a
firewall or router dropping the connection because it was idle for a long
time. The keep-alive mechanism sends heartbeat queries to Postgres
through that connection, thus reducing the likelihood of a connection
getting dropped. The interval between heartbeats can be controlled through the new
configuration option `keepalive_interval` and the corresponding CLI
option `--keepalive-interval` of the `barman backup` command.
Bug fixes
- When recovering a backup with the “no get wal” approach and `--target-time`
set, copy all WAL files. Previously Barman would attempt to “guess” the WAL
files required by Postgres to reach the configured target time. However,
the mechanism was not robust enough as it was based on the stats of the WAL
file in the Barman host (more specifically the creation time). For example:
if there were archiving or streaming lag between Postgres and Barman, that
could be enough for recovery to fail because Barman would miss to copy all
the required WAL files due to the weak check based on file stats.
- Pin `python-snappy` to `0.6.1` when running Barman through Python 3.6 or
older. Newer versions of `python-snappy` require `cramjam` version `2.7.0` or
newer, and these are only available for Python 3.7 or newer.
- `barman receive-wal` now exits with code `1` instead of `0` in the following
cases:
- Being unable to run with `--reset` flag because `pg_receivewal` is
running.
- Being unable to start `pg_receivewal` process because it is already
running.
- Fix and improve information about Python in `barman diagnose` output:
- The command now makes sure to use the same Python interpreter under which
Barman is installed when outputting the Python version through
`python_ver` JSON key. Previously, if an environment had multiple Python
installations and/or virtual environments, the output could eventually be
misleading, as it could be fetched from a different Python interpreter.
- Added a `python_executable` key to the JSON output. That contains the path
to the exact Python interpreter being used by Barman.