Delta

Reads and writes Delta tables, including Delta Merge operations and Time travel.

Source

Source Parameters

Parameter	Description	Required
Location	File path for the Delta table	True
Read Timestamp	Time travel to a specific timestamp	False
Read Version	Time travel to a specific version of table	False

note

For time travel on Delta tables:

1. Only Read Timestamp OR Read Version can be selected, not both.
2. Timestamp should be between the first commit timestamp and the latest commit timestamp in the table.
3. Version needs to be an integer. Its value has to be between min and max version of table.

By default most recent version of each row is fetched if no time travel option is used.

info

To read more about Delta time travel and its use cases click here.

Example

Generated Code

Without time travel

Python

def ReadDelta(spark: SparkSession) -> DataFrame:
    return spark.read.format("delta").load("dbfs:/FileStore/data_engg/delta_demo/silver/orders")

Scala

object ReadDelta {

  def apply(spark: SparkSession): DataFrame = {
    spark.read.format("delta").load("dbfs:/FileStore/data_engg/delta_demo/silver/orders")
  }

}

Timestamp-based time travel

Python

def ReadDelta(spark: SparkSession) -> DataFrame:
    return spark.read.format("delta").option("timestampAsOf", "2022-05-05")\
        .load("dbfs:/FileStore/data_engg/delta_demo/silver/orders")

Scala

object ReadDelta {

  def apply(spark: SparkSession): DataFrame = {
    spark.read.format("delta").option("timestampAsOf", "2022-05-05")
        .load("dbfs:/FileStore/data_engg/delta_demo/silver/orders")
  }

}

Version-based time travel

Python

def readDelta(spark: SparkSession) -> DataFrame:
    return spark.read.format("delta").option("versionAsOf", "0")\
        .load("dbfs:/FileStore/data_engg/delta_demo/silver/orders")

Scala

object readDelta {

  def apply(spark: SparkSession): DataFrame = {
    spark.read.format("delta").option("versionAsOf", "0")
        .load("dbfs:/FileStore/data_engg/delta_demo/silver/orders")
  }

}

---

Target

Target Parameters

Parameter	Description	Required
Location	File path to write the Delta table to	True
Write mode	Write mode for DataFrame	True
Optimise write	If true, it optimizes Spark partition sizes based on the actual data	False
Overwrite table schema	If true, overwrites the schema of the Delta table with the schema of the incoming DataFrame	False
Merge schema	If true, then any columns that are present in the DataFrame but not in the target table are automatically added on to the end of the schema as part of a write transaction	False
Partition Columns	List of columns to partition the Delta table by	False
Overwrite partition predicate	If specified, then it selectively overwrites only the data that satisfies the given where clause expression.	False

Supported Write Modes

Write Mode	Description
overwrite	If data already exists, overwrite with the contents of the DataFrame
append	If data already exists, append the contents of the DataFrame
ignore	If data already exists, do nothing with the contents of the DataFrame. This is similar to a CREATE TABLE IF NOT EXISTS in SQL.
error	If data already exists, throw an exception.
merge	Insert, delete and update data using the Delta merge command.
SCD2 merge	It is a Delta merge operation that stores and manages both current and historical data over time.

Among these write modes overwrite, append, ignore and error works the same way as in case of parquet file writes. Merge will be explained with several examples in the following sections.

Target Example

Generated Code

Python

def writeDelta(spark: SparkSession, in0: DataFrame):
    return in0.write\
            .format("delta")\
            .option("optimizeWrite", True)\
            .option("mergeSchema", True)\
            .option("replaceWhere", "order_dt > '2022-01-01'")\
            .option("overwriteSchema", True)\
            .mode("overwrite")\
            .partitionBy("order_dt")\
            .save("dbfs:/FileStore/data_engg/delta_demo/silver/orders")

Scala

object writeDelta {

  def apply(spark: SparkSession, in: DataFrame): Unit = {
    in0.write
        .format("delta")
        .option("optimizeWrite", True)
        .option("mergeSchema", True)
        .option("replaceWhere", "order_dt > '2022-01-01'")
        .option("overwriteSchema", True)
        .mode("overwrite")
        .partitionBy("order_dt")
        .save("dbfs:/FileStore/data_engg/delta_demo/silver/orders")
  }

}

Delta MERGE

You can upsert data from a source DataFrame into a target Delta table by using the MERGE operation. Delta MERGE supports Inserts, Updates, and Deletes in a variety of use cases, and Delta is particularly suited to examine data with individual records that slowly change over time. Here we consider the most common types of slowly changing dimension (SCD) cases: SCD1, SCD2, and SCD3. Records are modified in one of the following ways: history is not retained (SCD1), history is retained at the row level (SCD2), or history is retained at the column level (SCD3).

SCD1

Let's take the simplest case to illustrate a MERGE condition.

Parameters

Parameter	Description	Required
Source alias	Alias to use for the source DataFrame	True
Target alias	Alias to use for existing target Delta table	True
Merge Condition	Condition to merge data from source DataFrame to target table, which would be used to perform update, delete, or insert actions as specified.	True
When Matched Update Action	Update the row from Source that already exists in Target (based on Merge Condition)	False
When Matched Update Condition	Optional additional condition for updating row. If specified then it must evaluate to true for the row to be updated.	False
When Matched Update Expressions	Optional expressions for setting the values of columns that need to be updated.	False
When Matched Delete Action	Delete rows if Merge Condition (and the optional additional condition) evaluates to true	False
When Matched Delete Condition	Optional additional condition for deleting row. If a condition is specified then it must evaluate to true for the row to be deleted.	False
When Not Matched Action	The action to perform if the row from Source is not present in Target (based on Merge Condition)	False
When Not Matched Condition	Optional condition for inserting row. If a condition is specified then it must evaluate to true for the row to be updated.	False
When Not Matched Expressions	Optional expressions for setting the values of columns that need to be updated.	False

note

1. At least one action out of update, delete or insert needs to be set.
2. Delete removes the data from the latest version of the Delta table but does not remove it from the physical storage until the old versions are explicitly vacuumed. See vacuum for details.
3. A merge operation can fail if multiple rows of the source DataFrame match and the merge attempts to update the same rows of the target Delta table. Deduplicate gem can be placed before target if duplicate rows at source are expected.

When possible, provide predicates on the partition columns for a partitioned Delta table as such predicates can significantly speed up the operations.

Example

Let's assume our initial customers table is as below:

And we have the below updates coming into customers table

Our output and configurations for SCD1 merge will look like below:

Generated Code

Python

def writeDeltaMerge(spark: SparkSession, in0: DataFrame):
    from delta.tables import DeltaTable, DeltaMergeBuilder

    if DeltaTable.isDeltaTable(spark, "dbfs:/FileStore/data_engg/delta_demo/silver/customers_scd1"):
        DeltaTable\
            .forPath(spark, "dbfs:/FileStore/data_engg/delta_demo/silver/customers_scd1")\
            .alias("target")\
            .merge(in0.alias("source"), (col("source.customer_id") == col("target.customer_id")))\
            .whenMatchedUpdateAll()\
            .whenNotMatchedInsertAll()\
            .execute()
    else:
        in0.write\
            .format("delta")\
            .mode("overwrite")\
            .save("dbfs:/FileStore/data_engg/delta_demo/silver/customers_scd1")

Scala

object writeDeltaMerge {

  def apply(spark: SparkSession, in: DataFrame): Unit = {
    import _root_.io.delta.tables._
    if (DeltaTable.isDeltaTable(spark, "dbfs:/FileStore/data_engg/delta_demo/silver/customers_scd1")) {
        DeltaTable
            .forPath(spark, "dbfs:/FileStore/data_engg/delta_demo/silver/customers_scd1")
            .as("target")
            .merge(in0.as("source"), (col("source.customer_id") === col("target.customer_id")))
            .whenMatched()
            .updateAll()
            .whenNotMatched()
            .insertAll()
            .execute()
    }
    else {
        in0.write
            .format("delta")
            .mode("overwrite")
            .save("dbfs:/FileStore/data_engg/delta_demo/silver/customers_scd1")
    }
  }

}

SCD2

Let's use the Delta log to capture the historical customer_zip_code at the row-level.

Parameters

Parameter	Description	Required
Key columns	List of key columns which would remain constant	True
Historic columns	List of columns which would change over time for which history needs to be maintained	True
From time column	Time from which a particular row became valid	True
To time column	Time till which a particular row was valid	True
Min/old-value flag	Column placeholder to store the flag as true for the first entry of a particular key	True
Max/latest flag	Column placeholder to store the flag as true for the last entry of a particular key	True
Flag values	Option to choose the min/max flag to be true/false or 0/1	True

Example

Using the same customer tables as in our merge example above, output and configurations for SCD2 merge will look like below:

Generated Code

Python

def writeDeltaSCD2(spark: SparkSession, in0: DataFrame):
    from delta.tables import DeltaTable, DeltaMergeBuilder

    if DeltaTable.isDeltaTable(spark, "dbfs:/FileStore/data_engg/delta_demo/silver/customers_scd2"):
        existingTable = DeltaTable.forPath(
            spark, "dbfs:/FileStore/data_engg/delta_demo/silver/customers_scd2"
        )
        updatesDF = in0.withColumn("minFlag", lit("true")).withColumn(
            "maxFlag", lit("true")
        )
        existingDF = existingTable.toDF()
        updateColumns = updatesDF.columns
        stagedUpdatesDF = (
            updatesDF.join(existingDF, ["customer_id"])
            .where(
                (
                    (existingDF["maxFlag"] == lit("true"))
                    & (
                        (
                            (
                                existingDF["customer_zip_code"]
                                != updatesDF["customer_zip_code"]
                            )
                            | (
                                existingDF["customer_city"]
                                != updatesDF["customer_city"]
                            )
                        )
                        | (existingDF["customer_state"] != updatesDF["customer_state"])
                    )
                )
            )
            .select(*[updatesDF[val] for val in updateColumns])
            .withColumn("minFlag", lit("false"))
            .withColumn("mergeKey", lit(None))
            .union(updatesDF.withColumn("mergeKey", concat("customer_id")))
        )
        existingTable.alias("existingTable").merge(
            stagedUpdatesDF.alias("staged_updates"),
            concat(existingDF["customer_id"]) == stagedUpdatesDF["mergeKey"],
        ).whenMatchedUpdate(
            condition=(
                (existingDF["maxFlag"] == lit("true"))
                & (
                    (
                        (
                            existingDF["customer_zip_code"]
                            != stagedUpdatesDF["customer_zip_code"]
                        )
                        | (
                            existingDF["customer_city"]
                            != stagedUpdatesDF["customer_city"]
                        )
                    )
                    | (
                        existingDF["customer_state"]
                        != stagedUpdatesDF["customer_state"]
                    )
                )
            ),
            set={"maxFlag": "false", "end_date": "staged_updates.updated_dt"},
        )\
        .whenNotMatchedInsertAll()\
        .execute()
    else:
        in0.write\
            .format("delta")\
            .mode("overwrite")\
            .save("dbfs:/FileStore/data_engg/delta_demo/silver/customers_scd2")

Scala

object writeDeltaSCD2 {

  def apply(spark: SparkSession, in: DataFrame): Unit = {
    import _root_.io.delta.tables._
    if (
      DeltaTable.isDeltaTable(
        spark,
        "dbfs:/FileStore/data_engg/delta_demo/silver/customers_scd2"
      )
    ) {
      val updatesDF = in
        .withColumn("minFlag", lit("true"))
        .withColumn("maxFlag", lit("true"))
      val existingTable: DeltaTable = DeltaTable.forPath(
        spark,
        "dbfs:/FileStore/data_engg/delta_demo/silver/customers_scd2"
      )
      val existingDF: DataFrame = existingTable.toDF
      val stagedUpdatesDF = updatesDF
        .join(existingDF, List("customer_id"))
        .where(
          existingDF.col("maxFlag") === lit("true") && List(
            existingDF.col("customer_zip_code") =!= updatesDF
              .col("customer_zip_code"),
            existingDF.col("customer_city") =!= updatesDF
              .col("customer_city"),
            existingDF.col("customer_state") =!= updatesDF
              .col("customer_state")
          ).reduce((c1, c2) => c1 || c2)
        )
        .select(updatesDF.columns.map(x => updatesDF.col(x)): _*)
        .withColumn("minFlag", lit("false"))
        .withColumn("mergeKey", lit(null))
        .union(updatesDF.withColumn("mergeKey", concat(col("customer_id"))))
      existingTable
        .as("existingTable")
        .merge(
          stagedUpdatesDF.as("staged_updates"),
          concat(existingDF.col("customer_id")) === stagedUpdatesDF(
            "mergeKey"
          )
        )
        .whenMatched(
          existingDF.col("maxFlag") === lit("true") && List(
            existingDF.col("customer_zip_code") =!= stagedUpdatesDF
              .col("customer_zip_code"),
            existingDF.col("customer_city") =!= stagedUpdatesDF
              .col("customer_city"),
            existingDF.col("customer_state") =!= stagedUpdatesDF
              .col("customer_state")
          ).reduce((c1, c2) => c1 || c2)
        )
        .updateExpr(
          Map("maxFlag" → "false", "end_date" → "staged_updates.updated_dt")
        )
        .whenNotMatched()
        .insertAll()
        .execute()
    } else {
      in0.write
        .format("delta")
        .mode("overwrite")
        .save("dbfs:/FileStore/data_engg/delta_demo/silver/orders")
    }

  }

}

SCD3

Using the same customer tables as in our merge example above, output and configurations for SCD3 merge will look like below. Let's track change for customer_zip_code by adding a column to show the previous value.

---

info

To check out our blogpost on making data lakehouse easier using Delta with Prophecy click here.