Qeasy Cloud
Get Started

Sales Delivery Query Strategy in Practice: Incremental Pull Configuration from Kingdee Cloud to Qeasy

· 陈洁琳· Integration Solutions· 25 views· 4 min read

What This Strategy Solves

In a real project with a retail enterprise, Kingdee Cloud Sky serves as the back-end ERP, and sales orders and outbound flows need to be pulled out periodically by Qeasy (Qingyiyun Data Integration Platform) for downstream BI, warehouse dashboards, and reconciliation systems. What we built is the "Query Kingdee Sales Delivery - Latest" strategy: the source is Kingdee Cloud Sky, and the target is a no-op write action on the Qeasy integration platform (datahub). In essence, the data is staged in the integration platform and consumed by downstream strategies or systems.

It addresses three things: first, pulling data precisely by document type (standard outbound, consignment, VMI, B2C, etc., so they don't bleed into each other); second, running lightweight high-frequency polling so downstream systems don't hammer the source in turn; third, preparing a clean, traceable intermediate result for subsequent write strategies.

Data Flow and Field Mapping

Data flows from Kingdee Cloud Sky through Qeasy's transit layer, then to the target no-op landing point. The table below is the key field alignment we finalized on-site with the implementation engineer.

Business meaningSource field (Kingdee Cloud)Target field (datahub no-op)Notes
Document numberFBillNo0Number field
Document typeFBillTypeID.FNumber—Enums like XSCKD01_SYS
Entity line IDFEntity_FENTRYID—Line-level primary key
Document FIDFID—Header primary key
Business dateFDate—yyyy-MM-dd
Sales orgFSaleOrgId.FNumber—Org code

Setting number to FBillNo and id to FEntity_FENTRYID on the source side gives Qeasy a stable primary key for idempotent deduplication, preventing the same line from being delivered twice.

How to Configure on Qeasy

Pick Kingdee Cloud Sky as the source platform, choose the executeBillQuery API, POST method, effect marked as QUERY. We usually turn on autoFillResponse so the response structure is auto-applied to Qeasy's response model, reducing the chance of manual table-building errors.

For the target, pick Qeasy's integration platform (datahub), API is "no-op write", effect is EXECUTE. This is not really "doing nothing" — it lands the upstream pulled data into the integration platform's intermediate store as input for downstream strategies. This is a common "intermediate staging" pattern among Qeasy customers.

For code mapping, Qeasy offers centralized management: the XSCKD01_SYS enum values for document types should not be scattered across field conversions — keep them in a single mapping table. When Kingdee adds a new document type later, you only update one place.

Implementation Steps

We split the rollout into three stages and only opened it up once it stabilized.

Stage 1: Incremental starting point. Don't start the first trigger by pulling a full load from 1970. Set a business start date (e.g., the 1st of the current month), use FDate >= start date as the incremental baseline, and only backfill history after the first round stabilizes.

Stage 2: Full-load trigger. Once the incremental run is stable, schedule a manual full load during a nighttime window to backfill historical outbound documents. Qeasy strategies support a mix of manual trigger and auto schedule, which is convenient.

Stage 3: Schedule frequency. The source crontab is set to */7 * * * *, one round every 7 minutes. The target crontab is 1 1 1 1 1 (not scheduled). This is a typical "high-frequency pull, low-frequency write" pairing — light intake upstream, on-demand downstream, so the source isn't overwhelmed.

Before going live, run a dry-run on Qeasy to verify the returned fields match expectations, especially FEntity_FENTRYID — without it, downstream deduplication is dead.

Pitfalls and Lessons

  1. Treating "no-op" as truly doing nothing. In reality it's an EXECUTE channel and the data has already landed in datahub. Newcomers often miss the downstream subscription config here and assume the data is lost because it "didn't write out".
  2. Memorizing document-type enums by heart. XSCKD01_SYS through XSCKD08_SYS mean very different things. Without centralized mapping, three months later nobody will remember which is consignment and which is VMI. Always use centralized code-mapping management.
  3. Picking the incremental starting point arbitrarily. Choosing "the last 30 days" looks safe, but if the customer batch-enters orders at the start of the month, old orders from 30 days ago will be missed. Use the business start date as the baseline, not a fixed window.
  4. Using a 7-minute schedule as if it were real-time. A 7-minute latency must be acknowledged in downstream reconciliation scenarios — don't promise second-level freshness. If higher timeliness is needed, upgrade to Kingdee's real-time document subscription API, don't fiddle with the schedule.
  5. Primary key drift after turning off idCheck. We insist on idCheck: true so Qeasy uses FEntity_FENTRYID for idempotency checks. Once it's off, line-level updates become "new line appended", and the books stop matching.

Applicable and Non-Applicable Scenarios

Applicable: scenarios where sales outbound flows are pulled in batches by document type, an intermediate staging layer is needed for multiple downstream consumers, and minute-level latency is acceptable — retail, e-commerce, 3PL.

Not applicable: scenarios requiring second-level push (use Kingdee's real-time document subscription API instead), extremely large document volumes that must be streamed, or downstream systems that don't accept an intermediate layer and require a direct connection to the source.

Original content. Please credit the source when reposting: https://www.qeasy.cloud/insights/solutions/strat-kingdee-cloud-erp-3113-n5b7474c4-d2f8bec3

Comments