Data Flow activity in Azure Data Factory

• 01/03/2021
• 6 minutes to read
• • k
  • J
  • p
  • j
  • l
  • +5

APPLIES TO: Azure Data Factory Azure Synapse Analytics

Use the Data Flow activity to transform and move data via mapping data flows. If you're new to data flows, see Mapping Data Flow overview

Syntax

{
    "name": "MyDataFlowActivity",
    "type": "ExecuteDataFlow",
    "typeProperties": {
      "dataflow": {
         "referenceName": "MyDataFlow",
         "type": "DataFlowReference"
      },
      "compute": {
         "coreCount": 8,
         "computeType": "General"
      },
      "traceLevel": "Fine",
      "runConcurrently": true,
      "continueOnError": true,      
      "staging": {
          "linkedService": {
              "referenceName": "MyStagingLinkedService",
              "type": "LinkedServiceReference"
          },
          "folderPath": "my-container/my-folder"
      },
      "integrationRuntime": {
          "referenceName": "MyDataFlowIntegrationRuntime",
          "type": "IntegrationRuntimeReference"
      }
}

Type properties

Dynamically size data flow compute at runtime

The Core Count and Compute Type properties can be set dynamically to adjust to the size of your incoming source data at runtime. Use pipeline activities like Lookup or Get Metadata in order to find the size of the source dataset data. Then, use Add Dynamic Content in the Data Flow activity properties.

Here is a brief video tutorial explaining this technique

Data Flow integration runtime

Choose which Integration Runtime to use for your Data Flow activity execution. By default, Data Factory will use the auto-resolve Azure Integration runtime with four worker cores and no time to live (TTL). This IR has a general purpose compute type and runs in the same region as your factory. You can create your own Azure Integration Runtimes that define specific regions, compute type, core counts, and TTL for your data flow activity execution.

For pipeline executions, the cluster is a job cluster, which takes several minutes to start up before execution starts. If no TTL is specified, this start-up time is required on every pipeline run. If you specify a TTL, a warm cluster pool will stay active for the time specified after the last execution, resulting in shorter start-up times. For example, if you have a TTL of 60 minutes and run a data flow on it once an hour, the cluster pool will stay active. For more information, see Azure integration runtime.

PolyBase

If you're using an Azure Synapse Analytics as a sink or source, you must choose a staging location for your PolyBase batch load. PolyBase allows for batch loading in bulk instead of loading the data row-by-row. PolyBase drastically reduces the load time into Azure Synapse Analytics.

Logging level

If you do not require every pipeline execution of your data flow activities to fully log all verbose telemetry logs, you can optionally set your logging level to "Basic" or "None". When executing your data flows in "Verbose" mode (default), you are requesting ADF to fully log activity at each individual partition level during your data transformation. This can be an expensive operation, so only enabling verbose when troubleshooting can improve your overall data flow and pipeline performance. "Basic" mode will only log transformation durations while "None" will only provide a summary of durations.

Sink properties

The grouping feature in data flows allow you to both set the order of execution of your sinks as well as to group sinks together using the same group number. To help manage groups, you can ask ADF to run sinks, in the same group, in parallel. You can also set the sink group to continue even after one of the sinks encounters an error.

The default behavior of data flow sinks is to execute each sink sequentially, in a serial manner, and to fail the data flow when an error is encountered in the sink. Additionally, all sinks are defaulted to the same group unless you go into the data flow properties and set different priorities for the sinks.

Parameterizing Data Flows

Parameterized datasets

If your data flow uses parameterized datasets, set the parameter values in the Settings tab.

Parameterized data flows

If your data flow is parameterized, set the dynamic values of the data flow parameters in the Parameters tab. You can use either the ADF pipeline expression language or the data flow expression language to assign dynamic or literal parameter values. For more information, see Data Flow Parameters.

Parameterized compute properties.

You can parameterize the core count or compute type if you use the auto-resolve Azure Integration runtime and specify values for compute.coreCount and compute.computeType.

Pipeline debug of Data Flow activity

To execute a debug pipeline run with a Data Flow activity, you must switch on data flow debug mode via the Data Flow Debug slider on the top bar. Debug mode lets you run the data flow against an active Spark cluster. For more information, see Debug Mode.

The debug pipeline runs against the active debug cluster, not the integration runtime environment specified in the Data Flow activity settings. You can choose the debug compute environment when starting up debug mode.

Monitoring the Data Flow activity

The Data Flow activity has a special monitoring experience where you can view partitioning, stage time, and data lineage information. Open the monitoring pane via the eyeglasses icon under Actions. For more information, see Monitoring Data Flows.

Use Data Flow activity results in a subsequent activity

The data flow activity outputs metrics regarding the number of rows written to each sink and rows read from each source. These results are returned in the output section of the activity run result. The metrics returned are in the format of the below json.

{
    "runStatus": {
        "metrics": {
            "<your sink name1>": {
                "rowsWritten": <number of rows written>,
                "sinkProcessingTime": <sink processing time in ms>,
                "sources": {
                    "<your source name1>": {
                        "rowsRead": <number of rows read>
                    },
                    "<your source name2>": {
                        "rowsRead": <number of rows read>
                    },
                    ...
                }
            },
            "<your sink name2>": {
                ...
            },
            ...
        }
    }
}

For example, to get to number of rows written to a sink named 'sink1' in an activity named 'dataflowActivity', use @activity('dataflowActivity').output.runStatus.metrics.sink1.rowsWritten.

To get the number of rows read from a source named 'source1' that was used in that sink, use @activity('dataflowActivity').output.runStatus.metrics.sink1.sources.source1.rowsRead.

Next steps

See control flow activities supported by Data Factory:

• If Condition Activity
• Execute Pipeline Activity
• For Each Activity
• Get Metadata Activity
• Lookup Activity
• Web Activity
• Until Activity