rikonen / pghoard

PostgreSQL backup and restore service

PGHoard

pghoard is a PostgreSQL backup daemon and restore tooling that stores backup data in cloud object stores.

Features:

• Automatic periodic basebackups
• Automatic transaction log (WAL/xlog) backups (using either pg_receivexlog, archive_command or experimental PG native replication protocol support with walreceiver)
• Optional Standalone Hot Backup support
• Cloud object storage support (AWS S3, Google Cloud, OpenStack Swift, Azure, Ceph)
• Backup restoration directly from object storage, compressed and encrypted
• Point-in-time-recovery (PITR)
• Initialize a new standby from object storage backups, automatically configured as a replicating hot-standby

Fault-resilience and monitoring:

• Persists over temporary object storage connectivity issues by retrying transfers
• Verifies WAL file headers before upload (backup) and after download (restore), so that e.g. files recycled by PostgreSQL are ignored
• Automatic history cleanup (backups and related WAL files older than N days)
• "Archive sync" tool for detecting holes in WAL backup streams and fixing them
• "Archive cleanup" tool for deleting obsolete WAL files from the archive
• Keeps statistics updated in a file on disk (for monitoring tools)
• Creates alert files on disk on problems (for monitoring tools)

Performance:

• Parallel compression and encryption
• WAL pre-fetching on restore

Overview

PostgreSQL Point In Time Replication (PITR) consists of a having a database basebackup and changes after that point go into WAL log files that can be replayed to get to the desired replication point.

PGHoard supports multiple operating models. The basic mode where you have a separate backup machine, pghoard can simply connect with pg_receivexlog to receive WAL files from the database as they're written. Another model is to use pghoard_postgres_command as a PostgreSQL archive_command. There is also experimental support for PGHoard to use PostgreSQL's native replication protocol with the experimental walreceiver mode.

With both modes of operations PGHoard creates periodic basebackups using pg_basebackup that is run against the database in question.

The PostgreSQL write-ahead log (WAL) and basebackups are compressed with Snappy (default), Zstandard (configurable, level 3 by default) or LZMA (configurable, level 0 by default) in order to ensure good compression speed and relatively small backup size. For performance critical applications it is recommended to test compression algorithms to find the most suitable trade-off for the particular use-case. E.g. Snappy is fast but yields larger compressed files, Zstandard (zstd) on the other hand offers a very wide range of compression/speed trade-off.

Optionally, PGHoard can encrypt backed up data at rest. Each individual file is encrypted and authenticated with file specific keys. The file specific keys are included in the backup in turn encrypted with a master RSA private/public key pair.

PGHoard supports backing up and restoring from either a local filesystem or from various object stores (AWS S3, Azure, Ceph, Google Cloud and OpenStack Swift.)

In case you just have a single database machine, it is heavily recommended to utilize one of the object storage services to allow backup recovery even if the host running PGHoard is incapacitated.

Requirements

PGHoard can backup and restore PostgreSQL versions 9.3 and above. The daemon is implemented in Python and works with CPython version 3.5 or newer. The following Python modules are required:

• psycopg2 to look up transaction log metadata
• requests for the internal client-server architecture

Optional requirements include:

• azure for Microsoft Azure object storage
• botocore for AWS S3 (or Ceph-S3) object storage
• google-api-client for Google Cloud object storage
• cryptography for backup encryption and decryption (version 0.8 or newer required)
• snappy for Snappy compression and decompression
• zstandard for Zstandard (zstd) compression and decompression
• systemd for systemd integration
• swiftclient for OpenStack Swift object storage
• paramiko for sftp object storage

Developing and testing PGHoard also requires the following utilities: flake8, pylint and pytest.

PGHoard has been developed and tested on modern Linux x86-64 systems, but should work on other platforms that provide the required modules.

Vagrant

The Vagrantfile can be used to setup a vagrant development environment, consisting of two vagrant virtual machines.

1. Postgresql 9.4, python 3.5 and 3.6:

   vagrant up
   vagrant ssh postgres9
   cd /vagrant
   source ~/venv3/bin/activate
   make test
   source ~/venv3.6/bin/activate
   make test
   

2. Postgresql 10 and python 3.7:

   vagrant ssh postgres10
   cd /vagrant
   make test
   

Note: make deb does not work from vagrant at the moment, hopefully this will be resolved soon

Building

To build an installation package for your distribution, go to the root directory of a PGHoard Git checkout and run:

Debian:

make deb

This will produce a .deb package into the parent directory of the Git checkout.

Fedora:

make rpm

This will produce a .rpm package usually into rpm/RPMS/noarch/.

Python/Other:

python setup.py bdist_egg

This will produce an egg file into a dist directory within the same folder.

Installation

To install it run as root:

Debian:

dpkg -i ../pghoard*.deb

Fedora:

dnf install rpm/RPMS/noarch/*

On Linux systems it is recommended to simply run pghoard under systemd:

systemctl enable pghoard.service

and eventually after the setup section, you can just run:

systemctl start pghoard.service

Python/Other:

easy_install dist/pghoard-1.7.0-py3.6.egg

On systems without systemd it is recommended that you run pghoard under Supervisor or other similar process control system.

Setup

After this you need to create a suitable JSON configuration file for your installation.

0. Make sure PostgreSQL is configured to allow WAL archival and retrieval. postgresql.conf should have wal_level set to archive or higher and max_wal_senders set to at least 1 (archive_command mode) or at least 2 (pg_receivexlog and walreceiver modes), for example:

   wal_level = archive
   max_wal_senders = 4
   

   Note that changing wal_level or max_wal_senders settings requires restarting PostgreSQL.

1. Create a suitable PostgreSQL user account for pghoard:

   CREATE USER pghoard PASSWORD 'putyourpasswordhere' REPLICATION;
   

2. Edit the local pg_hba.conf to allow access for the newly created account to the replication database from the master and standby nodes. For example:

   # TYPE  DATABASE     USER     ADDRESS       METHOD
   host    replication  pghoard  127.0.0.1/32  md5
   

   After editing, please reload the configuration with either:

   SELECT pg_reload_conf();
   

   or by sending directly a SIGHUP to the PostgreSQL postmaster process.

3. Fill in the created user account and master/standby addresses into the configuration file pghoard.json to the section backup_sites.

4. Fill in the possible object storage user credentials into the configuration file pghoard.json under section object_storage in case you wish pghoard to back up into the cloud.

5. Now copy the same pghoard.json configuration to the standby node if there are any.

Other possible configuration settings are covered in more detail under the Configuration keys section of this README.

6. If all has been set up correctly up to this point, pghoard should now be ready to be started.

Backing up your database

PostgreSQL backups consist of full database backups, basebackups, plus write ahead logs and related metadata, WAL. Both basebackups and WAL are required to create and restore a consistent database (does not apply for standalone hot backups).

To enable backups with PGHoard the pghoard daemon must be running locally. The daemon will periodically take full basebackups of the database files to the object store. Additionally, PGHoard and PostgreSQL must be set up correctly to archive the WAL. There are two ways to do this:

The default option is to use PostgreSQL's own WAL-archive mechanism with pghoard by running the pghoard daemon locally and entering the following configuration keys in postgresql.conf:

archive_mode = on
archive_command = pghoard_postgres_command --mode archive --site default --xlog %f

This instructs PostgreSQL to call the pghoard_postgres_command whenever a new WAL segment is ready. The command instructs PGHoard to store the segment in its object store.

The other option is to set up PGHoard to read the WAL stream directly from PostgreSQL. To do this archive_mode must be disabled in postgresql.conf and pghoard.json must set active_backup_mode to pg_receivexlog in the relevant site, for example:

{
    "backup_sites": {
        "default": {
            "active_backup_mode": "pg_receivexlog",
            ...
         },
     },
     ...
 }

Note that as explained in the Setup section, postgresql.conf setting wal_level must always be set to archive, hot_standby or logical and max_wal_senders must allow 2 connections from PGHoard, i.e. it should be set to 2 plus the number of streaming replicas, if any.

While pghoard is running it may be useful to read the JSON state file pghoard_state.json that exists where json_state_file_path points. The JSON state file is human readable and is meant to describe the current state of pghoard 's backup activities.

Standalone Hot Backup Support

Pghoard has the option to enable standalone hot backups.

To do this archive_mode must be disabled in postgresql.conf and pghoard.json must set active_backup_mode to standalone_hot_backup in the relevant site, for example:

{
    "backup_sites": {
        "default": {
            "active_backup_mode": "standalone_hot_backup",
            ...
         },
     },
     ...
 }

For more information refer to the postgresql documentation https://www.postgresql.org/docs/9.5/continuous-archiving.html#BACKUP-STANDALONE

Restoring databases

You can list your database basebackups by running:

pghoard_restore list-basebackups --config /var/lib/pghoard/pghoard.json

Basebackup                       Size  Start time            Metadata
-------------------------------  ----  --------------------  ------------
default/basebackup/2016-04-12_0  8 MB  2016-04-12T07:31:27Z  {'original-file-size': '48060928',
                                                              'start-wal-segment': '000000010000000000000012',
                                                              'compression-algorithm': 'snappy'}

If we'd want to restore to the latest point in time we could fetch the required basebackup by running:

pghoard_restore get-basebackup --config /var/lib/pghoard/pghoard.json \
    --target-dir /var/lib/pgsql/9.5/data --restore-to-master

Basebackup complete.
You can start PostgreSQL by running pg_ctl -D foo start
On systemd based systems you can run systemctl start postgresql
On SYSV Init based systems you can run /etc/init.d/postgresql start

Note that the target-dir needs to be either an empty or non-existent directory in which case PGHoard will automatically create it.

After this we'd proceed to start both the PGHoard server process and the PostgreSQL server normally by running (on systemd based systems, assuming PostgreSQL 9.5 is used):

systemctl start pghoard
systemctl start postgresql-9.5

Which will make PostgreSQL start recovery process to the latest point in time. PGHoard must be running before you start up the PostgreSQL server. To see other possible restoration options please run:

pghoard_restore --help

Commands

Once correctly installed, there are six commands available:

pghoard is the main daemon process that should be run under a service manager, such as systemd or supervisord. It handles the backup of the configured sites.

pghoard_restore is a command line tool that can be used to restore a previous database backup from either pghoard itself or from one of the supported object stores. pghoard_restore can also configure recovery.conf to use pghoard_postgres_command as the WAL restore_command in recovery.conf.

pghoard_archive_cleanup can be used to clean up any orphan WAL files from the object store. After the configured number of basebackups has been exceeded (configuration key basebackup_count), pghoard deletes the oldest basebackup and all WAL associated with it. Transient object storage failures and other interruptions can cause the WAL deletion process to leave orphan WAL files behind, they can be deleted with this tool.

pghoard_archive_sync can be used to see if any local files should be archived but haven't been or if any of the archived files have unexpected content and need to be archived again. The other usecase it has is to determine if there are any gaps in the required files in the WAL archive from the current WAL file on to to the latest basebackup's first WAL file.

pghoard_create_keys can be used to generate and output encryption keys in the pghoard configuration format.

pghoard_postgres_command is a command line tool that can be used as PostgreSQL's archive_command or recovery_command. It communicates with pghoard 's locally running webserver to let it know there's a new file that needs to be compressed, encrypted and stored in an object store (in archive mode) or it's inverse (in restore mode.)

Configuration keys

active (default true)

Can be set on a per backup_site level to false to disable the taking of new backups and to stop the deletion of old ones.

active_backup_mode (default pg_receivexlog)

Can be either pg_receivexlog or archive_command. If set to pg_receivexlog, pghoard will start up a pg_receivexlog process to be run against the database server. If archive_command is set, we rely on the user setting the correct archive_command in postgresql.conf. You can also set this to the experimental walreceiver mode whereby pghoard will start communicating directly with PostgreSQL through the replication protocol. (Note requires an unreleased version of psycopg2 library)

alert_file_dir (default backup_location if set else os.getcwd())

Directory in which alert files for replication warning and failover are created.

backup_location (no default)

Place where pghoard will create its internal data structures for local state data and the actual backups. (if no object storage is used)

backup_sites (default {})

This object contains names and configurations for the different PostgreSQL clusters (here called sites) from which to take backups. The configuration keys for sites are listed below.

• compression WAL/basebackup compression parameters

• algorithm default "snappy" if available, otherwise "lzma" or "zstd"
• level default "0" compression level for "lzma" or "zstd" compression
• thread_count (default max(cpu_count, 5)) number of parallel compression threads

hash_algorithm (default "sha1")

The hash algorithm used for calculating checksums for WAL or other files. Must be one of the algorithms supported by Python's hashlib.

http_address (default "127.0.0.1")

Address to bind the PGHoard HTTP server to. Set to an empty string to listen to all available addresses.

http_port (default 16000)

HTTP webserver port. Used for the archive command and for fetching of basebackups/WAL's when restoring if not using an object store.

json_state_file_path (default "/var/lib/pghoard/pghoard_state.json")

Location of a JSON state file which describes the state of the pghoard process.

log_level (default "INFO")

Determines log level of pghoard.

maintenance_mode_file (default "/var/lib/pghoard/maintenance_mode_file")

If a file exists in this location, no new backup actions will be started.

pg_receivexlog

When active backup mode is set to "pg_receivexlog" this object may optionally specify additional configuration options. The currently available options are all related to monitoring disk space availability and optionally pausing xlog/WAL receiving when disk space goes below configured threshold. This is useful when PGHoard is configured to create its temporary files on a different volume than where the main PostgreSQL data directory resides. By default this logic is disabled and the minimum free bytes must be configured to enable it.

pg_receivexlog.disk_space_check_interval (default 10)

How often to check available disk space.

pg_receivexlog.min_disk_free_bytes (default undefined)

Minimum bytes (as an integer) that must be available in order to keep on receiving xlogs/WAL from PostgreSQL. If available disk space goes below this limit a STOP signal is sent to the pg_receivexlog / pg_receivewal application.

pg_receivexlog.resume_multiplier (default 1.5)

Number of times the min_disk_free_bytes bytes of disk space that is required to start receiving xlog/WAL again (i.e. send the CONT signal to the pg_receivexlog / pg_receivewal process). Multiplier above 1 should be used to avoid stopping and continuing the process constantly.

restore_prefetch (default transfer.thread_count)

Number of files to prefetch when performing archive recovery. The default is the number of Transfer Agent threads to try to utilize them all.

statsd (default: disabled)

Enables metrics sending to a statsd daemon that supports Telegraf or DataDog syntax with tags.

The value is a JSON object:

{
    "host": "<statsd address>",
    "port": <statsd port>,
    "format": "<statsd message format>",
    "tags": {
        "<tag>": "<value>"
    }
}

format (default: "telegraf")

Determines statsd message format. Following formats are supported:

• telegraf Telegraf spec

• datadog DataDog spec

The tags setting can be used to enter optional tag values for the metrics.

pushgateway (default: disabled)

Enables metrics sending to a Prometheus Pushgateway with tags.

The value is a JSON object:

{
    "endpoint": "<pushgateway address>",
    "tags": {
        "<tag>": "<value>"
    }
}

The tags setting can be used to enter optional tag values for the metrics.

prometheus (default: disabled)

Expose metrics through a Prometheus endpoint.

The value is a JSON object:

{
    "tags": {
        "<tag>": "<value>"
    }
}

The tags setting can be used to enter optional tag values for the metrics.

syslog (default false)

Determines whether syslog logging should be turned on or not.

syslog_address (default "/dev/log")

Determines syslog address to use in logging (requires syslog to be true as well)

syslog_facility (default "local2")

Determines syslog log facility. (requires syslog to be true as well)

• transfer WAL/basebackup transfer parameters

• thread_count (default max(cpu_count, 5)) number of parallel uploads/downloads

upload_retries_warning_limit (default 3)

After this many failed upload attempts for a single file, create an alert file.

tar_executable (default "pghoard_gnutaremu")

The tar command to use for restoring basebackups. This must be GNU tar because some advanced switches like --transform are needed. If this value is not defined (or is explicitly set to "pghoard_gnutaremu"), Python's internal tarfile implementation is used. The Python implementation is somewhat slower than the actual tar command and in environments with fast disk IO (compared to available CPU capacity) it is recommended to set this to "tar".

Backup site configuration

The following options control the behavior of each backup site. A backup site means an individual PostgreSQL installation ("cluster" in PostgreSQL terminology) from which to take backups.

basebackup_age_days_max (default undefined)

Maximum age for basebackups. Basebackups older than this will be removed. By default this value is not defined and basebackups are deleted based on total count instead.

basebackup_chunks_in_progress (default 5)

How many basebackup chunks can there be simultaneously on disk while it is being taken. For chunk size configuration see basebackup_chunk_size.

basebackup_chunk_size (default 2147483648)

In how large backup chunks to take a local-tar basebackup. Disk space needed for a successful backup is this variable multiplied by basebackup_chunks_in_progress.

basebackup_compression_threads (default 0)

Number of threads to use within compression library during basebackup. Only applicable when using compression library that supports internal multithreading, namely zstd at the moment. Default value 0 means not to use multithreading.

basebackup_count (default 2)

How many basebackups should be kept around for restoration purposes. The more there are the more diskspace will be used. If basebackup_max_age is defined this controls the maximum number of basebackups to keep; if backup interval is less than 24 hour or extra backups are created there can be more than one basebackup per day and it is often desirable to set basebackup_count to something slightly higher than the max age in days.

basebackup_count_min (default 2)

Minimum number of basebackups to keep. This is only effective when basebackup_age_days_max has been defined. If for example the server is powered off and then back on a month later, all existing backups would be very old. However, in that case it is usually not desirable to immediately delete all old backups. This setting allows specifying a minimum number of backups that should always be preserved regardless of their age.

basebackup_hour (default undefined)

The hour of day during which to start new basebackup. If backup interval is less than 24 hours this is the base hour used to calculate the hours at which backup should be taken. E.g. if backup interval is 6 hours and this value is set to 1 backups will be taken at hours 1, 7, 13 and 19. This value is only effective if also basebackup_interval_hours and basebackup_minute are set.

basebackup_interval_hours (default 24)

How often to take a new basebackup of a cluster. The shorter the interval, the faster your recovery will be, but the more CPU/IO usage is required from the servers it takes the basebackup from. If set to a null value basebackups are not automatically taken at all.

basebackup_minute (default undefined)

The minute of hour during which to start new basebackup. This value is only effective if also basebackup_interval_hours and basebackup_hour are set.

basebackup_mode (default "basic")

The way basebackups should be created. The default mode, basic runs pg_basebackup and waits for it to write an uncompressed tar file on the disk before compressing and optionally encrypting it. The alternative mode pipe pipes the data directly from pg_basebackup to PGHoard's compression and encryption processing reducing the amount of temporary disk space that's required.

Neither basic nor pipe modes support multiple tablespaces.

Setting basebackup_mode to local-tar avoids using pg_basebackup entirely when pghoard is running on the same host as the database. PGHoard reads the files directly from $PGDATA in this mode and compresses and optionally encrypts them. This mode allows backing up user tablespaces.

Note that the local-tar backup mode can not be used on replica servers prior to PostgreSQL 9.6 unless the pgespresso extension is installed.

basebackup_threads (default 1)

How many threads to use for tar, compress and encrypt tasks. Only applies for local-tar basebackup mode. Only values 1 and 2 are likely to be sensible for this, with higher thread count speed improvement is negligible and CPU time is lost switching between threads.

encryption_key_id (no default)

Specifies the encryption key used when storing encrypted backups. If this configuration directive is specified, you must also define the public key for storing as well as private key for retrieving stored backups. These keys are specified with encryption_keys dictionary.

encryption_keys (no default)

This key is a mapping from key id to keys. Keys in turn are mapping from public and private to PEM encoded RSA public and private keys respectively. Public key needs to be specified for storing backups. Private key needs to be in place for restoring encrypted backups.

You can use pghoard_create_keys to generate and output encryption keys in the pghoard configuration format.

object_storage (no default)

Configured in backup_sites under a specific site. If set, it must be an object describing a remote object storage. The object must contain a key storage_type describing the type of the store, other keys and values are specific to the storage type.

The following object storage types are supported:

• local makes backups to a local directory, see pghoard-local-minimal.json for example. Required keys:

• directory for the path to the backup target (local) storage directory

• sftp makes backups to a sftp server, required keys:

• server
• port
• username
• password or private_key

• google for Google Cloud Storage, required configuration keys:

• project_id containing the Google Storage project identifier
• bucket_name bucket where you want to store the files
• credential_file for the path to the Google JSON credential file

• s3 for Amazon Web Services S3, required configuration keys:

• aws_access_key_id for the AWS access key id
• aws_secret_access_key for the AWS secret access key
• region S3 region of the bucket
• bucket_name name of the S3 bucket

Optional keys for Amazon Web Services S3:

• encrypted if True, use server-side encryption. Default is False.

• s3 for other S3 compatible services such as Ceph, required configuration keys:

• aws_access_key_id for the AWS access key id
• aws_secret_access_key for the AWS secret access key
• bucket_name name of the S3 bucket
• host for overriding host for non AWS-S3 implementations
• port for overriding port for non AWS-S3 implementations
• is_secure for overriding the requirement for https for non AWS-S3
• is_verify_tls for configuring tls verify for non AWS-S3 implementations

• azure for Microsoft Azure Storage, required configuration keys:

• account_name for the name of the Azure Storage account
• account_key for the secret key of the Azure Storage account
• bucket_name for the name of Azure Storage container used to store objects
• azure_cloud Azure cloud selector, "public" (default) or "germany"

• swift for OpenStack Swift, required configuration keys:

• user for the Swift user ('subuser' in Ceph RadosGW)
• key for the Swift secret_key
• auth_url for Swift authentication URL
• container_name name of the data container
• Optional configuration keys for Swift:

• auth_version - 2.0 (default) or 3.0 for keystone, use 1.0 with Ceph Rados GW.
• segment_size - defaults to 1024**3 (1 gigabyte). Objects larger than this will be split into multiple segments on upload. Many Swift installations require large files (usually 5 gigabytes) to be segmented.
• tenant_name
• region_name
• user_id - for auth_version 3.0
• user_domain_id - for auth_version 3.0
• user_domain_name - for auth_version 3.0
• tenant_id - for auth_version 3.0
• project_id - for auth_version 3.0
• project_name - for auth_version 3.0
• project_domain_id - for auth_version 3.0
• project_domain_name - for auth_version 3.0
• service_type - for auth_version 3.0
• endpoint_type - for auth_version 3.0

nodes (no default)

Array of one or more nodes from which the backups are taken. A node can be described as an object of libpq key: value connection info pairs or libpq connection string or a postgres:// connection uri. If for example you'd like to use a streaming replication slot use the syntax {... "slot": "slotname"}.

pg_bin_directory (default: find binaries from well-known directories)

Site-specific option for finding pg_basebackup and pg_receivexlog commands matching the given backup site's PostgreSQL version. If a value is not supplied PGHoard will attempt to find matching binaries from various well-known locations. In case pg_data_directory is set and points to a valid data directory the lookup is restricted to the version contained in the given data directory.

pg_data_directory (no default)

This is used when the local-tar basebackup_mode is used. The data directory must point to PostgreSQL's $PGDATA and must be readable by the pghoard daemon.

prefix (default: site name)

Path prefix to use for all backups related to this site. Defaults to the name of the site.

Alert files

Alert files are created whenever an error condition that requires human intervention to solve. You're recommended to add checks for the existence of these files to your alerting system.

authentication_error

There has been a problem in the authentication of at least one of the PostgreSQL connections. This usually denotes a wrong username and/or password.

configuration_error

There has been a problem in the authentication of at least one of the PostgreSQL connections. This usually denotes a missing pg_hba.conf entry or incompatible settings in postgresql.conf.

upload_retries_warning

Upload of a file has failed more times than upload_retries_warning_limit. Needs human intervention to figure out why and to delete the alert once the situation has been fixed.

version_mismatch_error

Your local PostgreSQL client versions of pg_basebackup or pg_receivexlog do not match with the servers PostgreSQL version. You need to update them to be on the same version level.

version_unsupported_error

Server PostgreSQL version is not supported.

License

PGHoard is licensed under the Apache License, Version 2.0. Full license text is available in the LICENSE file and at http://www.apache.org/licenses/LICENSE-2.0.txt

Credits

PGHoard was created by Hannu Valtonen <[e-mail]> for Aiven and is now maintained by Aiven developers <[e-mail]>.

Recent contributors are listed on the GitHub project page, https://github.com/aiven/pghoard/graphs/contributors

Contact

Bug reports and patches are very welcome, please post them as GitHub issues and pull requests at https://github.com/aiven/pghoard . Any possible vulnerabilities or other serious issues should be reported directly to the maintainers <[e-mail]>.

Copyright

Copyright (C) 2015 Aiven Ltd