Databases, DBMS Processes, Bolt, HTTP, Drivers, Configuration, Storage Locations, and Transaction Logs

Learning outcomes

AtlasMart can now connect to Neo4j, but a production operator needs to know what lies behind that successful query. This lesson draws the boundaries between the DBMS process, the neo4j database, the system database, Bolt and HTTP connectors, drivers, configuration inputs, persistent stores, logs, plugins, and transaction logs. The goal is observability without treating internal files as an application API.

DBMS process versus database

The DBMS is the running Neo4j management/runtime process. A database is a transactional data-management domain hosted by that DBMS. A standard current installation includes the system database for DBMS metadata/security/administration and a default user database named neo4j. Community supports one user database plus system; Enterprise extends multi-database administration.

A transaction cannot be casually treated as spanning arbitrary databases. Application code should specify the intended database rather than relying forever on a default. The driver example in Lesson 3 used database_="neo4j" deliberately. That makes routing and intent explicit and avoids surprises if a deployment’s default/home database changes.

CALL dbms.components() YIELD name, versions, edition RETURN name, versions, edition;SHOW CURRENT USER;MATCH (n) RETURN count(n) AS nodesInCurrentDatabase;

Enterprise learners may additionally inspect SHOW DATABASES; Community learners should not make that command a mandatory check.

Connectors: Browser is not Bolt and Bolt is not a driver

Neo4j exposes network connectors. The current defaults include HTTP on 7474, HTTPS on 7473 when enabled, and Bolt on 7687. Browser is a web client typically loaded through HTTP/HTTPS; it then connects to the database. Cypher Shell and official language drivers speak Bolt. A driver adds language-level connection pooling, sessions/transactions, routing, retries and result APIs around Bolt—it is not just a socket.

SHOW SETTINGS 'server.bolt.enabled', 'server.bolt.listen_address', 'server.bolt.advertised_address', 'server.http.enabled', 'server.http.listen_address' YIELD name, value RETURN name, value ORDER BY name;

Listen and advertised addresses solve different problems. A listener says where a server accepts traffic; an advertised address is what clients/routing may be told to use. Container NAT, DNS, clusters, proxies, and cloud networking can make them differ. The lab’s host-loopback mapping is Docker-level evidence in addition to Neo4j’s internal connector settings.

Configuration provenance: file, environment, and runtime evidence

For archive/server distributions, neo4j.conf is the principal configuration file. Docker can also derive settings from environment variables and mounted configuration. The safest operational practice is to know which source generated the effective setting and then confirm the runtime value. Editing a file without restarting or without understanding precedence does not prove the active DBMS changed.

docker exec atlasmart-neo4j sh -lc 'echo "$NEO4J_HOME"; sed -n "1,80p" "$NEO4J_HOME/conf/neo4j.conf"'docker exec atlasmart-neo4j cypher-shell -a neo4j://localhost:7687 -u neo4j -p atlasmart-course-2026 "SHOW SETTINGS 'db.query.default_language', 'server.bolt.enabled', 'server.http.enabled' YIELD name, value RETURN name, value ORDER BY name;"docker inspect atlasmart-neo4j --format '{{json .Config.Env}}' 

Do not paste a production docker inspect environment dump into tickets if it may contain secrets. This course environment contains only a disposable password. Production credentials should come from a secrets mechanism rather than plaintext Docker environment variables when practical.

Storage locations are operational evidence, not public data structures

Neo4j’s documented default locations separate configuration, data, logs, plugins, metrics/products and other runtime directories. In Docker, official mount points include /data, /logs, /conf, and /plugins. This chapter mounts only data and logs because no plugin or custom configuration is required yet.

docker exec atlasmart-neo4j sh -lc 'find /data -maxdepth 2 -type d -print | sort | head -80'docker exec atlasmart-neo4j sh -lc 'find /logs -maxdepth 1 -type f -printf "%f %s bytes\n" 2>/dev/null | sort'docker exec atlasmart-neo4j sh -lc 'find /data/transactions -maxdepth 2 -type f -printf "%p %s bytes\n" 2>/dev/null | head -40' 

Exact internal file names, store format, rotation, and directory contents can change by version and edition. The lesson observes them only to connect operations to documented concepts. An application must never depend on a particular store filename.

Transaction logs connect committed writes to recovery

Neo4j records transactional changes in transaction logs. They support recovery and other operational mechanisms. A successful transaction means the DBMS committed according to its transactional rules; it does not mean a separate off-host backup exists. Transaction logs are not a replacement for a backup policy, and replication—where available—is also not a backup.

Create one clearly identifiable write, then compare transaction-log directory metadata before and after. File sizes can be buffered/rotated and therefore should not be treated as a one-write-equals-N-bytes formula.

docker exec atlasmart-neo4j sh -lc 'find /data/transactions/neo4j -maxdepth 1 -type f -printf "%f %s bytes %TY-%Tm-%TdT%TH:%TM:%TS\n" 2>/dev/null | sort' 

CYPHER 25MERGE (p:Product {productId:'P-2001'})SET p.name='Smart Shelf Sensor',    p.category='Store IoT',    p.updatedAt=datetime('2026-09-09T09:00:00Z')RETURN p.productId, p.name, p.updatedAt;

docker exec atlasmart-neo4j sh -lc 'find /data/transactions/neo4j -maxdepth 1 -type f -printf "%f %s bytes %TY-%Tm-%TdT%TH:%TM:%TS\n" 2>/dev/null | sort' 

The accepted evidence is that the graph query returns the committed state and the transaction-log area exists/changes over normal operation. Do not claim that one timestamp or byte delta uniquely identifies that transaction. Background activity, checkpoints, rotation, and buffering make that inference unsound.

What Neo4j is doing during the write

The driver/Cypher Shell sends the statement over Bolt. Neo4j executes it in a transaction, obtains the necessary locks/concurrency control, evaluates the MERGE match/create semantics, updates graph state and relevant indexes/constraints, records transactional changes, and commits or rolls back as a unit. MERGE is not a universal SQL-style UPSERT: its match pattern and constraints determine what can match/create, and concurrent writers require a deliberate uniqueness model. Later chapters analyze these details.

Memory also has distinct surfaces: JVM heap for runtime objects/query execution, page cache for store pages, native/off-heap allocations, and—when used later—Graph Data Science memory. Increasing one pool blindly can starve another. Chapter 01 records the vocabulary; performance engineering later measures it.

Deliberately wrong approach: manipulate files because they are visible

A common local-lab mistake is to find a large file under /data/transactions or a store file under /data/databases and delete it manually. That bypasses Neo4j’s recovery, retention, checkpoint, and consistency mechanisms. The same anti-pattern applies to copying a subset of live files and calling it a backup.

The repair is to manage transaction-log retention through supported configuration and to use supported dump/backup/restore tooling appropriate to the edition. If a lab needs destructive file experiments in a later internals chapter, it must operate on a stopped disposable copy with an explicit reset path.

Production judgment

For self-managed production, document NEO4J_HOME/NEO4J_CONF strategy, persistent-volume ownership, data/log separation, filesystem permissions, secrets, TLS, connector exposure, log collection, transaction-log retention, backup paths, disk-headroom alarms, and upgrade-compatible layouts. Windows, packages, archives, containers, and orchestrators place files differently; avoid runbooks that assume Docker paths everywhere.

For Aura, most of these server-file details disappear from the customer control surface, but the application still must use the correct database, secure URI, driver settings, query timeouts, credentials, and service-supported recovery/observability mechanisms.

Summary and next step

You can now locate the boundary between supported client interfaces and implementation storage, distinguish DBMS from database, explain HTTP/Bolt/driver roles, inspect effective configuration, and connect a committed write to transaction-log state without over-interpreting internal files.

Next, turn these pieces into a repeatable course lab with a richer AtlasMart fixture, administrative-account evidence, metrics, failure checks, and safe reset procedures.

Authoritative references

• Current Neo4j versions — Official current-release and LTS patch snapshot.
• Neo4j Operations Manual — Authoritative self-managed operational documentation for the current release.
• Neo4j system requirements — Supported operating systems and JVM requirements.
• Cypher Manual — Current Cypher 25 reference and language semantics.
• Configure the Cypher default version — Cypher 5 versus Cypher 25 default and override behavior.
• Neo4j in Docker — Official image tags, ports, editions, and Docker starting point.
• Default file locations — Distribution-specific config/data/log/plugin/product locations.
• Neo4j transaction logs — Transaction-log purpose and management.
• Neo4j ports — Connector and administrative port reference.
• Configure network connectors — Listen/advertised address and protocol configuration.
• Docker configuration — Environment/configuration behavior for official images.
• Docker volumes — Official mount points and persistence behavior.