Data layer

ADB combines four distinct stores. Each one has a specific role and is consumed by a subset of the services.

flowchart LR
    subgraph Services
        persons[adb-persons]
        parts[adb-parts]
        contracts[adb-contracts]
        accounting[adb-accounting]
        files[adb-files]
        utilities[adb-utilities]
        views[adb-views]
        reports[adb-reports]
        aggregates[adb-aggregates]
    end

    subgraph Mongo[MongoDB Atlas]
        m_persons[(adb-persons)]
        m_parts[(adb-parts)]
        m_contracts[(adb-contracts)]
        m_accounting[(adb-accounting)]
        m_files[(adb-files)]
        m_utilities[(adb-utilities)]
        m_views[(adb-views)]
        m_reports[(adb-reports)]
    end

    s3[(AWS S3)]
    neo4j[(Neo4j)]

    persons --> m_persons
    parts --> m_parts
    contracts --> m_contracts
    accounting --> m_accounting
    files --> m_files
    files --> s3
    utilities --> m_utilities
    views --> m_views
    reports --> m_reports

    aggregates -.reads via HTTP.-> persons
    aggregates -.reads via HTTP.-> parts
    aggregates -.reads via HTTP.-> contracts

    adbgraph[adb-graph<br/>deprecated] --> neo4j

1. MongoDB Atlas

All business data lives in MongoDB Atlas, a managed multi-environment cluster (adb-development, adb-integration, adb-production). Each service has its own database (adb-<service>): no cross-service joins — data is propagated via HTTP or events. Exception: adb-aggregates has no own database — it fetches data directly from other services at query time

Databases and collections

adb-views hosts denormalized projections kept up to date by SQS listeners — this is intentional, to serve dashboards quickly.

Connection

Each service reads its connection string from the MONGO_DB_URI variable (format mongodb+srv://...). Credentials come from Kubernetes secrets.

Reactive stack

Access goes through Spring Data MongoDB Reactive (ReactiveMongoRepository). Services avoid any blocking call; BlockHound is enabled in cloud environments for adb-views and adb-accounting to detect blocking calls at runtime.

2. AWS S3

Two main buckets:

adb-files issues pre-signed S3 URLs for direct uploads and downloads from the browser (avoiding a round-trip through the Java service).

3. Neo4j

⚠️ Deprecated — adb-graph is shut down and Neo4j is no longer in use.

The adb-graph service (out-of-monorepo) maintains a graph of the relationships between persons, parts, contracts and tickets. It is fed by RabbitMQ events (persons.graph, parts.graph, contracts.graph, tickets.graph).

Helm/Terraform for Neo4j: see adb-charts/terraform/ and the scripts under adb-charts/scripts/cypher_r4.cypher.

4. Messaging — AWS SQS / SNS

Detailed in Service-to-service communication §4. Single region: eu-west-1. Queues are named by convention <env>-<service>-<event> (e.g. dev-views-contract-modified-queue).

Each service has:

• One SNS topic per emitted event type.
• One DLQ (<env>-<service>-dlq) for unprocessed messages.

Migrations / setup scripts

adb-infrastructure/mongodb-atlas/data-scripts/ contains around 37 migration and setup scripts. Tooling: MongoDB Atlas CLI + Realm CLI. No Terraform yet — managed manually or through the CLI.

Common environment variables