Backup Types, Online Backup Concepts, Transaction Logs, Retention, and Recovery Objectives

Learning outcomes

Reproducible AtlasMart setup

The drill adds one tiny course-owned marker so recovery correctness has a deterministic invariant even if your earlier AtlasMart graph has grown. The marker is not a substitute for business reconciliation; it is a canary.

CYPHER 25
CREATE CONSTRAINT recovery_marker_id IF NOT EXISTS
FOR (m:RecoveryMarker) REQUIRE m.markerId IS UNIQUE;
MERGE (c:Customer {customerId:'C-1001'})
ON CREATE SET c.name = 'Mina Rahimi'
MERGE (m:RecoveryMarker {markerId:'DR-CH16-001'})
SET m.createdAt = datetime('2026-09-09T17:00:00Z'),
    m.expectedState = 'before-dump',
    m.exercise = 'chapter-16'
MERGE (c)-[:HAS_RECOVERY_MARKER]->(m);
MATCH (m:RecoveryMarker {markerId:'DR-CH16-001'})
RETURN m.markerId, m.expectedState;

1. Start with the recovery objective, not the command

AtlasMart can survive a server restart and still fail a disaster: an operator can delete data correctly on every cluster member, ransomware can encrypt every writable replica, or an application bug can commit bad state. Replication keeps copies available; a backup preserves an independent recoverable state. The design question is therefore: how much committed work may the business lose, and how long may recovery take?

2. Community offline dump vs Enterprise online backup

neo4j-admin database dump requires the Community DBMS/database to be offline. It produces a single archive of database contents. It is operationally simple and free, but downtime contributes directly to RTO and the artifact is a point-in-time full copy rather than a continuously replayable backup chain.

Enterprise neo4j-admin database backup talks to a configured online backup service. The first artifact is full; subsequent runs can produce differential artifacts containing transaction logs. A valid chain is the full plus contiguous differentials. Since Neo4j 2026.02, the first differential may overlap the full, but subsequent parent/child coverage still matters.

3. Transaction-log retention is a recoverability dependency

db.tx_log.rotation.retention_policy controls how long logical transaction logs remain before safe pruning after checkpoints. The current default is 2 days 2G. The setting is dynamic. Manual deletion of transaction-log files is unsupported because pruning must respect checkpoint/recovery state.

SHOW SETTINGS YIELD name, value, dynamic
WHERE name = 'db.tx_log.rotation.retention_policy'
RETURN name, value, dynamic;

4. Produce the Community artifact without touching the source store

Stop the course container first. Then mount its data volume read/write into the official admin image and write the dump to a host backup directory. Exact Docker filesystem ownership varies by host; fix permissions deliberately rather than using world-writable backup directories.

mkdir -p ./neo4j-dr/backups
docker stop atlasmart-neo4j
docker run --rm \
  -v atlasmart-neo4j-data:/data \
  -v "$PWD/neo4j-dr/backups:/backups" \
  neo4j/neo4j-admin:2026.07.1 \
  neo4j-admin database dump neo4j --to-path=/backups
ls -lh ./neo4j-dr/backups/neo4j.dump
sha256sum ./neo4j-dr/backups/neo4j.dump

5. Inspect and consistency-check before calling it “good”

An artifact that merely exists can still be unusable or incomplete. Record its hash and run a consistency check while it is still isolated.

docker run --rm \
  -v "$PWD/neo4j-dr/backups:/backups" \
  neo4j/neo4j-admin:2026.07.1 \
  neo4j-admin database check --from-path=/backups/neo4j.dump neo4j

6. The misleading shortcut: “the cluster is my backup”

A replicated topology can protect availability from a member failure, but it normally replicates legitimate destructive transactions too. A copied live data directory can also be crash-inconsistent unless the mechanism is designed for it. The repair is independent, access-controlled artifacts plus routine restore drills.

7. Recovery-objective worksheet

Production judgment

Summary and next step

Backup design begins with RPO/RTO and failure classes, not with a command. Lesson 2 restores the Community dump into an isolated volume, validates the graph and explains exactly where Enterprise point-in-time recovery changes the mechanism.

Authoritative references

• Current Neo4j versions — Current server and 5.26 LTS release snapshot.
• Backup and restore — Edition-aware entry point for dump/load, online backup, restore and planning.
• Backup and restore planning — RPO/RTO, backup mode, storage location, cadence and retention planning.
• Back up an offline database — neo4j-admin database dump semantics and Community offline boundary.
• Restore a database dump — neo4j-admin database load semantics, overwrite rules and edition differences.
• Back up an online database — Enterprise full/differential backup artifacts and chain semantics.
• Restore a database backup — Enterprise recovery of backup chains and restore-until predicates.
• Check database consistency — neo4j-admin database check for stores, dumps and recovered full backups.
• Transaction logging — Transaction-log retention, checkpointing and pruning behavior.
• Store formats — Current aligned/block formats, limits and legacy-format deprecation.
• Migrate a database — neo4j-admin database migrate and store-format migration boundaries.
• System requirements — Supported Java/runtime and platform requirements for current Neo4j.
• APOC installation — APOC/server release compatibility and restart/deployment coupling.
• GDS compatibility — Graph Data Science and Neo4j version compatibility matrix.
• Python driver installation — Current 6.x driver/server compatibility baseline.