18.1 Integration Mental Model

1. Start with the workload

Do not start with the connector catalog.

Start with the workload:

Business requirement
      |
Freshness
Volume
Latency
Availability
Security
      |
Integration pattern
      |
Connection / adapter
      |
Runtime design

2. Four primary patterns

Federation

Query the source when the consumer needs data.

Best when:

• data is relatively stable,
• source query performance is acceptable,
• low-copy architecture matters,
• freshness must be immediate.

Risk:

• consumer workload becomes source workload.

Replication

Copy source data into Datasphere.

Best when:

• repeated analytics queries are expected,
• source protection matters,
• predictable performance is required,
• history or local transformation is valuable.

Risk:

• storage, freshness management and duplicate data.

ETL or transformation

Extract data, transform it and persist the result.

Best when:

• business logic is complex,
• multiple sources must be harmonized,
• transformation should be separated from source systems.

Push

An external application or ETL platform writes into an Open SQL schema.

Best when:

• the external platform owns extraction,
• a custom integration is required,
• application-driven ingestion is appropriate.

3. Decision matrix

Requirement	Federation	Replication	ETL	Push
Live access	Strong	Depends on freshness	Depends	Depends
Source protection	Weaker	Stronger	Stronger	Stronger
Complex transformation	Limited	Moderate	Strong	External
Predictable query performance	Depends on source	Strong	Strong	Strong
Minimal persistence	Strong	Weaker	Weaker	Depends

4. Anti-pattern

Source
  |
"Connect everything"
  |
Remote tables everywhere
  |
Complex views
  |
Source overloaded

A remote table is not automatically the best architecture.

5. Architect exercise

For these workloads choose the primary pattern:

• operational dashboard requiring current inventory,
• monthly finance reporting,
• 5 TB historical analytics,
• customer-service application lookup,
• cross-cloud enterprise data product.

For each decision state why the rejected alternatives are inferior.