Barman vs Medusa

Barman is an open-source Python administration tool for PostgreSQL disaster recovery, designed for organizations managing business-critical databases. It supports remote backups of multiple PostgreSQL servers and local or remote recovery, including point-in-time recovery. Backup management includes catalogs, incremental and parallel operations, retention policies, and WAL archiving or streaming. Synchronous WAL streaming supports zero data loss (RPO=0). Remote transfers can use rsync over SSH or pg_basebackup; recovery can run locally or over SSH. Barman integrates with PostgreSQL and standard archiving tools such as tar, and can compress WAL files with bzip2, gzip, or a custom method. GPG encryption for backups and WAL depends on configuration. Immutable storage is supported with a required grace period in the current implementation. Its experimental cloud-streaming method is limited to S3-compatible destinations and does not support encryption or direct restores in Barman. It runs on Linux and UNIX-like systems, uses an INI configuration file, and stores metadata locally. Barman costs nothing and is distributed under GPL-3.

Medusa is a command-line backup system for Apache Cassandra that runs on Linux in a self-hosted deployment. It supports full and differential backups, single-node backup and restore, and cluster-wide restores in place or to a remote location. Backup destinations include local storage, Google Cloud Storage, Azure Blob Storage, AWS S3, and compatible S3 services. It supports clusters using either single tokens or vnodes. Its verify command checks backup completion, manifest and file presence, and file hashes; MD5 checks can be enabled, while file size is checked by default. Configuration documentation covers client SSL settings and S3 server-side encryption with KMS or customer-provided keys. Retention settings can purge backups by age or count. Installation instructions cover pip and Debian packages, including offline installation, and describe installing Medusa on each Cassandra node. It requires Python 3.9 or newer and is licensed under Apache-2.0. A stated constraint is that it does not support Cassandra deployments with multiple data-folder directories.