apache-spark - Spark incremental loading overwrite old record

Question

I have a requirement to do the incremental loading to a table by using Spark (PySpark)

Here's the example:

Day 1

id | value
-----------
1  | abc
2  | def

Day 2

id | value
-----------
2  | cde
3  | xyz

Expected result

id | value
-----------
1  | abc
2  | cde
3  | xyz

This can be done easily in relational database,
Wondering whether this can be done in Spark or other transformational tool, e.g. Presto?

标签： apache-sparkpysparketlpresto

Answer 1

Here you go! First Dataframe:

 >>> list1 = [(1, 'abc'),(2,'def')]
 >>> olddf = spark.createDataFrame(list1, ['id', 'value'])
 >>> olddf.show();
 +---+-----+
 | id|value|
 +---+-----+
 |  1|  abc|
 |  2|  def|
 +---+-----+

Second Dataframe:

>>> list2 = [(2, 'cde'),(3,'xyz')]
>>> newdf = spark.createDataFrame(list2, ['id', 'value'])
>>> newdf.show();
+---+-----+
| id|value|
+---+-----+
|  2|  cde|
|  3|  xyz|
+---+-----+

Now join and merge these two datafame using full outer join and use coalesce function while select and can replace the null values wih user defined values.

from pyspark.sql.functions import *

>>> df = olddf.join(newdf, olddf.id == newdf.id,'full_outer').select(coalesce(olddf.id,newdf.id).alias("id"),coalesce(newdf.value,olddf.value).alias("value"))
>>> df.show();
+---+-----+
| id|value|
+---+-----+
|  1|  abc|
|  3|  xyz|
|  2|  cde|
+---+-----+

I hope this should solve your problem. :-)