MetricSign
EN|NLRequest Access
High severitydata source

Power BI Refresh Error:
DF-SAPODP-OOM

What does this error mean?

The Self-Hosted Integration Runtime ran out of memory (OOM — Out Of Memory) while processing an SAP ODP extraction. The IR host did not have enough available RAM to buffer the data being received from SAP, and the process was killed by the operating system.

Common causes

  • 1The ODP extraction is pulling a very large dataset that exceeds the available memory on the IR host
  • 2Multiple large pipelines ran concurrently on the same IR host, exhausting shared memory
  • 3The IR host machine has insufficient RAM for the volume of data being extracted
  • 4The extraction retrieves a full load of a large SAP table without any row filtering, causing excessive memory consumption

How to fix it

  1. 1Reduce the extraction batch size by configuring the rowsPerPage setting in the ADF SAP ODP source — this limits how many rows the connector buffers in memory at one time.
  2. 2Add row-level filters in the ADF source to reduce the total data volume — filter by date range, company code, or plant to extract only the records needed.
  3. 3If possible, switch from full extraction to delta extraction for subsequent runs — delta runs process only changed records and use far less memory.
  4. 4Upgrade the Self-Hosted IR host machine to one with more RAM — 16GB is a practical minimum for large SAP extractions; 32GB is recommended for high-volume workloads.
  5. 5Schedule large SAP ODP pipelines to run sequentially rather than concurrently to avoid memory contention on the IR host.
  6. 6Monitor IR host memory usage during runs using Windows Task Manager or Performance Monitor (perfmon) to confirm OOM as the root cause.

Frequently asked questions

What is rowsPerPage and how do I configure it?

RowsPerPage controls rows per RFC call — lowering it reduces peak memory. In the SAP ODP source, go to Source options and set rowsPerPage. Start with 50000 and adjust based on memory pressure.

Can I run SAP ODP extractions in parallel across multiple IR hosts?

Yes — ADF distributes concurrent runs across SHIR cluster nodes, but each ODP extraction runs on a single node. To parallelize large extractions, split the dataset and run multiple pipelines each targeting a subset.

Does switching to delta extraction always solve OOM?

For recurring daily runs, yes — delta processes only changed records and is far smaller than a full load. The initial full load may require batching or IR upgrades; after that, delta handles subsequent runs.

Will downstream Power BI datasets be affected?

Yes — the pipeline is killed mid-run and the target table may have incomplete data. Power BI datasets will show either stale data from the previous run or partial data from the failed run.

Official documentation: https://learn.microsoft.com/en-us/azure/data-factory/data-flow-troubleshoot-guide

Other data source errors