Κοινή χρήση μέσω

Use foreachBatch to write to arbitrary data sinks

  • Άρθρο

This article discusses using foreachBatch with Structured Streaming to write the output of a streaming query to data sources that do not have an existing streaming sink.

The code pattern streamingDF.writeStream.foreachBatch(...) allows you to apply batch functions to the output data of every micro-batch of the streaming query. Functions used with foreachBatch take two parameters:

  • A DataFrame that has the output data of a micro-batch.
  • The unique ID of the micro-batch.

You must use foreachBatch for Delta Lake merge operations in Structured Streaming. See Upsert from streaming queries using foreachBatch.

Apply additional DataFrame operations

Many DataFrame and Dataset operations are not supported in streaming DataFrames because Spark does not support generating incremental plans in those cases. Using foreachBatch() you can apply some of these operations on each micro-batch output. For example, you can use foreachBath() and the SQL MERGE INTO operation to write the output of streaming aggregations into a Delta table in update mode. See more details in MERGE INTO.

Important

  • foreachBatch() provides only at-least-once write guarantees. However, you can use the batchId provided to the function as way to deduplicate the output and get an exactly-once guarantee. In either case, you will have to reason about the end-to-end semantics yourself.
  • foreachBatch() does not work with the continuous processing mode as it fundamentally relies on the micro-batch execution of a streaming query. If you write data in continuous mode, use foreach() instead.

An empty dataframe can be invoked with foreachBatch() and user code needs to be resilient to allow for proper operation. An example is shown here:

  .foreachBatch(
          (outputDf: DataFrame, bid: Long) => {
             // Process valid data frames only
             if (!outputDf.isEmpty) {
                // business logic
             }
         }
  ).start()

Behavior changes for foreachBatch in Databricks Runtime 14.0

In Databricks Runtime 14.0 and above on compute configured with shared access mode, the following behavior changes apply:

  • print() commands write output to the driver logs.
  • You cannot access the dbutils.widgets submodule inside the function.
  • Any files, modules, or objects referenced in the function must be serializable and available on Spark.

Reuse existing batch data sources

Using foreachBatch(), you can use existing batch data writers for data sinks that might not have Structured Streaming support. Here are a few examples:

Many other batch data sources can be used from foreachBatch(). See Connect to data sources.

Write to multiple locations

If you need to write the output of a streaming query to multiple locations, Databricks recommends using multiple Structured Streaming writers for best parallelization and throughput.

Using foreachBatch to write to multiple sinks serializes the execution of streaming writes, which can increase latency for each micro-batch.

If you do use foreachBatch to write to multiple Delta tables, see Idempotent table writes in foreachBatch.