Glossary

Terms that mean something particular in Sentinel, or that are easy to confuse with a neighbouring idea. Each entry links to the page that explains it in depth.

A

Artifact : The file a backup produces: a dump, optionally compressed, optionally encrypted. An artifact is written to a storage backend and is accompanied by a manifest. See Backup.

additional_args : Engine-specific arguments passed through to the underlying dump tool. Parsed with shell-like tokenisation and validated before use. See the additional_args reference.

B

Backup job : A named entry under databases: in the configuration, describing one database to back up and where to put it. Distinct from a restore job. See Backup.

Baseline : The full backup a chain of incrementals is built on. Deleting it makes every incremental above it useless, which is why retention protects the active one.

Binary log : MySQL and MariaDB's record of changes, the basis of their incremental backups. Sentinel archives binary logs beside the artifact. See Incremental and PITR.

C

Chain : A baseline plus the ordered incrementals that depend on it. Chain depth is capped by max_chain_depth, after which a new full backup starts a new chain. Validate one with sentinel restore validate-chain. See Incremental and PITR.

Conflict strategy : What a restore does when data already exists at the target: error (the default), replace, or ignore. MySQL and MariaDB collapse the last two into the same behaviour. See Restore.

D

defaults_file : A MySQL or MariaDB option file, typically a my.cnf, whose [client] section supplies host, user and password for the whole pipeline rather than just the dump subprocess. See Credential handling.

Dispatcher : The component that fans a single event out to every configured notification channel. See Notifications.

Dry run : A rehearsal that reports what would happen without doing it. Note that sentinel restore dry-run does not invoke the restore planner, so it can pass for a job that cannot actually run (#152).

E

Envelope : The container format wrapping an encrypted artifact, identified by a SENC header. Version 2 chunks the plaintext and derives a per-artifact key. See Security and encryption.

Execution : One recorded run of a job, successful or not, stored in the history database.

F

Full backup : A complete copy, as opposed to an incremental one. Every chain starts with a full backup as its baseline. See Backup.

G

GFS : Grandfather-father-son. A retention scheme keeping the newest backup of each recent day, week, month and year, layered on top of the flat count and age rules. See Retention.

H

History database : The local SQLite file recording every execution. Its location is history_db_path, defaulting to ~/.sentinel/history.db. See Monitoring and execution history.

I

Incremental backup : A backup capturing only what changed since the previous one in its chain, using the engine's own change log. Works on all four engines. Incremental restore is a separate matter, and currently fails (#150). See Incremental and PITR.

L

Lock : A file taken for the duration of a job so two runs cannot interleave. Restores take one; backups currently do not (#163). See Locking and concurrency.

M

Manifest : The <artifact>.manifest.json sidecar recording an artifact's SHA-256, its encryption parameters, and its position in a chain. Written only when the job sets output: (#151). See Manifests and integrity.

Master key : The 32-byte key from which per-artifact encryption keys are derived. Provisioned through encryption_key_env or encryption_key_file. Losing it is unrecoverable. See Security and encryption.

O

Oplog : MongoDB's operation log, the basis of its incremental backups. Recovery granularity is bounded by the oplog window, which is how much history the server retains. See Incremental and PITR.

P

PITR : Point-in-time recovery: restoring to a specific moment rather than to a backup boundary. Advertised by Sentinel, but currently unplannable for any engine (#148). See Incremental and PITR.

Planner : The component deciding whether a requested restore is possible and how to assemble it, returning a plan or a reason code such as unsupported_database_type or missing_advanced_metadata. See Restore.

R

Redaction : Removing credentials from anything Sentinel logs, handled by the sanitize package. It covers what Sentinel writes, not what an engine tool prints. See Credential handling.

Restore job : A named entry under restores:, describing where to restore from and what to restore into. Disabled by default, unlike backup jobs. See Restore.

Retention : Deciding which old backups to delete, by count, by age, or by GFS tier. See Retention.

S

Secrets file : A file holding database credentials, referenced by defaults_file or mongo_secrets_file, and optionally encrypted at rest in an SSEC container. See Security and encryption.

Staging : Writing an artifact to local disk before uploading it to a remote backend, so hashing and encryption happen on a file Sentinel controls. Restores stage in the other direction, downloading before applying. See Storage backends and the MongoDB staging reference.

Storage backend : Where artifacts are written: local disk, S3, Google Cloud Storage, Google Drive, or Azure Blob. Restore reads from only the first three. See Storage backends.

W

WAL : PostgreSQL's write-ahead log, the basis of its incremental backups. Requires summarize_wal=on server-side on PostgreSQL 17 and later. See Incremental and PITR.

{/* sources: website/docs/concepts/, internal/config/types.go, internal/domain/restore/planner.go */}