Data architecture

The data architecture describes how data are organized, related, stored and circulated to meet the needs of the organisation. A list of services does not explain who originates a data, what an authorized source is, what transformation changes its meaning and what consumers depend on it.

The architecture should connect today's state, goal and a viable transition. No need to migrate anything to solve a particular problem. If Horizon needs to reconcile bills daily, it starts with the flow that this reconciliation allows, its officials and controls. Add data with no consumer or defined purpose can increase cost and exposure without production of value.

Conceptual architecture - domains and relations -, logic - products, contracts and interfaces - and physics - services, formats and deployments - should be distinguished. The three views answer different questions and have to preserve the same meaning of the business.

The three columns describe the same client C7 with different level of detail. The conceptual view relates the client with his contracts C2 and C3. The logic defines "active client" as the person with at least one contract currently in place at the date of cut and identifies owner and version. The physics represents the result with identifier, state and date 30/09. The three views have to preserve the same meaning of business. ↗

Limits, Flows and Decisions

For every flow, you ask: who's production, who's consumption, what's his goal? What's the maximum latency of an update allowed? What's his name? What's his name about changing his scheme? It adds residence, volume, recovery and cost requirements as necessary. A liability-expressed master's limit a network boundary expressed connectivity. They're not equivalent.

Federar allows a source to be consulted by means of a connection, avoiding a previous load at destination for that access. It preserves availability, performance and permissions dependence from the source. To materialise creates a persistent result useful for performance, history or decoupling but requires to manage update and life cycle. Share grants access to a published package under conditions: does not remove the liability of the recipient.

To choose an alternative does not solve all future consultations. The monthly closure may require a reproducible version while an operational consultation requires the current state. To design both roads is reasonable if each copy and connection has a clear purpose.

Compare each pattern by data location, history maintenance and dependencies. To consult at source depends on the source, network and its permissions. Materializing preserves a result but requires an update and registration of versions and rules. Sharing adds conditions for the recipient and can be combined with both roads. Choose by use: Check the present and play a closure require different evidence. ↗

architecture with Oracle Cloud

Oracle AI Data Platform provides an environment for catalogue, engineering, workflows and AI assets. The Master Catalog from Oracle AI Data Platform register and organize metadata. The External catalogues Connecting compatible sources The access should be reviewed by considering Connection credentials and permissions from the source. Do not assume that every consultation will be run there with the personal identity of the consumer. Query External catalogues.

The operated Spark motor serves for production and processing. Oracle Autonomous AI Lakehouse allows to organize and consult cured information for analytics and other consumers. Oracle Cloud Infrastructure Object Storage can contain files and data from intermediate phases. They are complementary responsibilities: Oracle AI Data Platform isn't another name for Oracle Autonomous AI Lakehouse, and logging a file doesn't atcredit its quality.

Use Oracle AI Data Platform lineage to cover the sources, transformations and consumers registered before changing a product. It contrasts these dependencies with architecture design and warns those concerned. The vision retains the final capture of each process. To rebuild an earlier closure, you also save data and code versions, identifiers and evidence from execution. That's how you connect impact analysis with data product continuity.

To use Agent Flows and others AI functions, configure Oracle Autonomous AI Lakehouse 26ai or more. The agent's memory retains the context of the meeting; The persistent memory at Oracle AI Database 26ai of the diagram requires its own integration. The continuous processing uses compatible sources and destinations. Oracle Analytics Cloud accesses by means of a Configure connection.

The architecture starts with archives at Oracle Cloud Infrastructure Object Storage and organizes Landing, Bronze, Silver and Gold at Oracle AI Data Platform Workshop. Master Catalog registers metadata and the lineage allows to review dependencies. Agent Flows uses Oracle Cloud Infrastructure models Generative AI. The marked "Memory" connection offers persistent memory at Oracle AI Database 26ai, which requires its own integration. Session memory is an agent's function. The functions of AI require Oracle Autonomous AI Lakehouse 26ai or more. Streaming corresponds to continuous processing of supported sources and destinations. ↗

Explanated example

Horizonte needs two results. Customer Care consults existing contracts during an call Finances reproduces the September closure. For Customer Care an examination with the source and with a load and availability are evaluated. For Finance a version of the closure with date and known rules are preserved. "All data always in real time" does not necessarily satisfy the second case.

In technical light, Oracle AI Data Platform organizes connections and transformations. Oracle Autonomous AI Lakehouse publishes the closure set. The owner approves of the meaning of the court and the consumer knows its validity. If you change an relationship between client and contract, the dependencies are reviewed before changing the product. Adds the required freshness for Customer Care and maintenance period for Finance. Configure and test each path with respect to these criteria.

The contract is currently valid in September and suspended in October. Customer Care can be consulted with today's state while Finance needs to play the September 30 closure. The closing file keeps the data, applying rules and cutting date. That's how we can rebuild that result even if their origin changes. only check October's state does not explain the earlier closure. ↗

Common

• To merge catalogue with Universal storage. Methatates can describe assets that are followed at their sources.
• Assuming federation is free and has no overhead. The source, network and search engine are still at work.
• Draw arrows with no contracts or identities. It's hidden how every exchange is controlled and maintained.

Decision exercise

The director wants to rebuild the exact figure of the closure within 6 months. The team always propose to consult today's operational tables by Federation. What's missing from that proposal?