Logistics and warehouse on Databricks

Shipment exception and on-time recovery on Databricks

Finds deliveries that will miss the customer's requested date, shows the cause, offers expedite, split or a new date, and records the customer's confirmation on the order.

SCREENSDATA HANDLERSDATABRICKS · READ IN PLACES01 Deliveries at riskS02 Delivery detailS03 Recovery optionsS04 Customer messageS05 Carrier portalS06 Boarddh_deliveryDeliverysilver.deliverydh_sales_order_lineSales order linesilver.sales_order_linedh_stock_positionStock positionsilver.stock_position · gold.stock_dailydh_production_orderProduction ordersilver.production_orderdh_customer_masterCustomer mastersilver.customerWRITE BACK TO DATABRICKSDelivery → MERGE INTO gold.lima_app_decisionsSales order line → MERGE INTO gold.lima_app_decisionsrows stay in Databricks · access rules applied at read time · each write acknowledged and logged
The screens of the app, the data handler behind each, and the Databricks object it reads in place. Writes below the line go through MERGE into a Lima-owned table, with the operational write to the ERP of record.
5
objects read in place from Databricks
catalogue
6
screens on those rows
catalogue
5 days
to production-grade
estimate
2
objects written back
catalogue
0
rows copied to build it
product
What the app reads

The Databricks objects behind each handler

Delta tables in Unity Catalog are read in place through a SQL warehouse: the app sends only the filter and the columns it needs. The Backbone landing model (bronze, silver and gold) is the usual source, so history and aggregates are already prepared and lineage is registered in Unity Catalog.

ObjectRead fromFields the app usesFreshness
Deliverysilver.deliveryroute (route) · planned_gi_date (planned gi) · actual_gi_date (actual gi) · delivery_id (number)prepared by the Backbone; refreshed on its schedule
Sales order linesilver.sales_order_linerequested_date (requested date) · so_number (number) · so_line (line) · customer_id (customer)prepared by the Backbone; refreshed on its schedule
Stock positionsilver.stock_position · gold.stock_dailymaterial_id (material) · plant (plant) · storage_location (sloc) · on_hand_qty (on hand)prepared by the Backbone; refreshed on its schedule
Production ordersilver.production_orderorder_id (number) · material_id (material) · order_qty (qty) · start_date (start)prepared by the Backbone; refreshed on its schedule
Customer mastersilver.customercustomer_id (id) · name (name) · country (country) · sales_org (sales org)prepared by the Backbone; refreshed on its schedule
The trigger

The start criterion in Databricks's own fields

In plain words: An open delivery whose planned goods issue date has passed with actual goods issue empty, or whose planned goods issue plus the transit days of the route falls after the requested date on the order line.

The criterion runs in Databricks through the data handler, so only the matching rows return. The editor shows the rows it selects, and their count, as it is written.

# start criterion in Databricks fields
actual_gi_date = '' and (planned_gi_date < today or planned_gi_date + route_days(route) > requested_date)

# the same criterion in canonical fields
{delivery.actual_gi} = '' and ({delivery.planned_gi} < today or {delivery.planned_gi} + route_days({delivery.route}) > {so.requested_date})
What the app writes back

Through MERGE and the system of record

Decisions are recorded in a Lima-owned gold table with MERGE INTO, so notebooks and dashboards see them next to the history. The operational write goes to the system of record through its released interface, and the app records the acknowledgement. For this app the operational change (planned goods issue, route or carrier changed) goes to SAP S/4HANA or Oracle Fusion through its released interface, and the pair pages for those systems show the exact call.

ObjectWhat changesInterface
Deliveryplanned goods issue, route or carrier changed; delivery split into a partial delivery through the delivery change interfaceMERGE INTO gold.lima_app_decisions
Sales order lineconfirmed date and the customer's confirmation written as a note on the order lineMERGE INTO gold.lima_app_decisions
Acknowledged and logged

Each write stores the identifier Databricks returns beside the decision, the person and the time, so the log answers who changed what, and the reconciliation shows it arrived.

Access and prerequisites

What is in place before the first build

Unity Catalog grants and row filters are mapped to Lima access policies once, so a person sees the same rows in the app that they see in a notebook.

  • A service principal with SELECT on the silver and gold schemas and MODIFY on the Lima decision schema
  • A SQL warehouse (a small serverless warehouse serves most screens)
  • The Backbone landing model in place for the objects the app uses, or your own silver tables registered
  • The system of record connected for the operational write
Register the handlers5 objects from the catalog; each shows its rows at once.
Why the lake

History and aggregates, read where they land

This app reads delivery from Databricks because the screens need history at scale: months of rows, aggregates by period, several versions. The Backbone lands the objects in the conformed model and refreshes them on a schedule, and the handler reads them in place with the filter and the columns pushed down. The person's grants in Databricks decide what they see.

The operational rows the person acts on can come live from the ERP in the same app: one handler per object per system, and a screen that shows both.

Readsconformed and history tables, in placeDatabricks
Decision recordMERGE into a Lima-owned tablebeside the history
Operational writeto the ERP of record through its interfaceacknowledged
GrantsDatabricks roles and row policies honouredat read time
Nearby

This app on other systems, and other apps on Databricks

Shipment exception and on-time recovery on

Other apps on Databricks

Terms on this page

On time in fullA delivery that arrives on the customer's requested date with the full quantity; the board counts the share of deliveries that meet both.
Goods issueThe posting that records goods leaving the warehouse; the planned goods issue date is when it should happen.
Read in placeThe rows stay in Databricks; the handler sends the filter and the columns and Databricks returns the matching rows.
Questions

Does this app copy data out of Databricks?

The screens read through handlers that leave the rows in Databricks. The conformed tables it reads were landed there by the Backbone from the ERP; the app reads them there.

What does Databricks need to allow?

The prerequisites above: an integration identity with read rights on the objects listed and write rights on the interfaces named, connectivity, and the scopes agreed with the process owner.

Can I run it on another system as well?

Yes. The screens, criteria and roles stay; the handlers are re-bound to the other system's objects, and the app's other system pages show them.

See this app on your data

Bring the process and the systems it touches. In the first session we register the data handlers, draw the levels or the lanes and run the first screen on your live rows.