Some months ago I presented save modes in Spark SQL. However, this post was limited to their use in files. I was quite surprised to observe some specific behavior of them for RDBMS sinks. Especially for SaveMode.Overwrite.
Some weeks ago I presented correlated scalar subqueries in the example of PostgreSQL. However they can also be found in the Big Data processing systems, as for instance BigQuery or Apache Spark SQL.
In programming a simple is often the synonymous of understandable and maintainable. However it doesn't always mean efficient. One of examples of this thesis is nested loop join that is also present in Apache Spark SQL.
Prior to Spark 2.2.0 release, the data processing was based on a set of heuristic rules ignoring the typology of the data. But the most recent release brought a tool well known from the RDBMS world that is a Cost-Based Optimizer.
DataFrame can either be loaded and saved. And Spark SQL provides, as for a lot other points, different strategies to deal with data persistence.
It's time to continue the exploration of operator optimizations of logic plans in Spark SQL. After the first part describing optimizations from A to L, this post covers remaining letters.
Pushdown predicate is one of the most popular optimizations in Spark SQL. But it's not the single one and their main list is defined in org.apache.spark.sql.catalyst.optimizer.Optimizer abstract class.
User Defined Functions are not the single way to extend Spark SQL. The second solution is offered by User Defined Aggregate Functions.
Spark SQL has a lot of "hidden" features making it an efficient processing tool. One of them are statistics.
The optimizer in Spark SQL helps to improve the performance of processing pipelines. One of its techniques is predicate pushdown.
Chain of responsibility design pattern is one of my favorite's alternatives to avoid too many nested calls. Some days ago I was wondering if it could be used instead of nested calls of multiple UDFs applied in column level in Spark SQL. And the response was affirmative.
User Defined Types (UDT) described in one of previous posts aren't the single customization possibility in Apache Spark SQL. The other possibility are User Defined Functions (UDF).
Spark SQL reading from RDBMS is based on classic JDBC drivers. Thus it supports some of their options, as fetchsize described in sections below.
After discovering two methods used to join DataFrames, broadcast and hashing, it's time to talk about the third possibility - sort-merge join.
Shuffle consists on moving data with the same key to the one executor in order to execute some specific processing on it. We could think that it concerns only *ByKey operations but it's not necessarily true.
Joining DataFrames can be a performance-sensitive task. After all, it involves matching data from two data sources and keeping matched results in a single place. As you can deduce, the first thinking goes towards shuffle join operation. However, it's not the single strategy implemented in Spark SQL. For some specific use cases another type called broadcast join can be preferred.
Spark SQL reflects the most of concepts related to relational databases as possible. One of them are joins that can be defined in one of 7 forms.
Without any explicit definition, Spark SQL won't partition any data, i.e. all rows will be processed by one executor. It's not optimal since Spark was designed to parallel and distributed processing.
Structured data processing takes more and more place in Apache Spark project. Structured streaming is one of the proofs. But how does Spark SQL work - and particularly, how does it load data from sources of structured data as RDMBS ?
Even if it's always better to explicit things, in programming we have often the possibility to let the computer to guess. Spark SQL also has this level of intelligence, for example during schema resolving.
