| |
| <!DOCTYPE html> |
| <!--[if lt IE 7]> <html class="no-js lt-ie9 lt-ie8 lt-ie7"> <![endif]--> |
| <!--[if IE 7]> <html class="no-js lt-ie9 lt-ie8"> <![endif]--> |
| <!--[if IE 8]> <html class="no-js lt-ie9"> <![endif]--> |
| <!--[if gt IE 8]><!--> <html class="no-js"> <!--<![endif]--> |
| <head> |
| <meta charset="utf-8"> |
| <meta http-equiv="X-UA-Compatible" content="IE=edge,chrome=1"> |
| <title>Migration Guide: SQL, Datasets and DataFrame - Spark 3.0.0-preview2 Documentation</title> |
| |
| |
| |
| |
| <link rel="stylesheet" href="css/bootstrap.min.css"> |
| <style> |
| body { |
| padding-top: 60px; |
| padding-bottom: 40px; |
| } |
| </style> |
| <meta name="viewport" content="width=device-width"> |
| <link rel="stylesheet" href="css/bootstrap-responsive.min.css"> |
| <link rel="stylesheet" href="css/main.css"> |
| |
| <script src="js/vendor/modernizr-2.6.1-respond-1.1.0.min.js"></script> |
| |
| <link rel="stylesheet" href="css/pygments-default.css"> |
| |
| |
| <!-- Google analytics script --> |
| <script type="text/javascript"> |
| var _gaq = _gaq || []; |
| _gaq.push(['_setAccount', 'UA-32518208-2']); |
| _gaq.push(['_trackPageview']); |
| |
| (function() { |
| var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true; |
| ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js'; |
| var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s); |
| })(); |
| </script> |
| |
| |
| </head> |
| <body> |
| <!--[if lt IE 7]> |
| <p class="chromeframe">You are using an outdated browser. <a href="https://browsehappy.com/">Upgrade your browser today</a> or <a href="http://www.google.com/chromeframe/?redirect=true">install Google Chrome Frame</a> to better experience this site.</p> |
| <![endif]--> |
| |
| <!-- This code is taken from http://twitter.github.com/bootstrap/examples/hero.html --> |
| |
| <div class="navbar navbar-fixed-top" id="topbar"> |
| <div class="navbar-inner"> |
| <div class="container"> |
| <div class="brand"><a href="index.html"> |
| <img src="img/spark-logo-hd.png" style="height:50px;"/></a><span class="version">3.0.0-preview2</span> |
| </div> |
| <ul class="nav"> |
| <!--TODO(andyk): Add class="active" attribute to li some how.--> |
| <li><a href="index.html">Overview</a></li> |
| |
| <li class="dropdown"> |
| <a href="#" class="dropdown-toggle" data-toggle="dropdown">Programming Guides<b class="caret"></b></a> |
| <ul class="dropdown-menu"> |
| <li><a href="quick-start.html">Quick Start</a></li> |
| <li><a href="rdd-programming-guide.html">RDDs, Accumulators, Broadcasts Vars</a></li> |
| <li><a href="sql-programming-guide.html">SQL, DataFrames, and Datasets</a></li> |
| <li><a href="structured-streaming-programming-guide.html">Structured Streaming</a></li> |
| <li><a href="streaming-programming-guide.html">Spark Streaming (DStreams)</a></li> |
| <li><a href="ml-guide.html">MLlib (Machine Learning)</a></li> |
| <li><a href="graphx-programming-guide.html">GraphX (Graph Processing)</a></li> |
| <li><a href="sparkr.html">SparkR (R on Spark)</a></li> |
| </ul> |
| </li> |
| |
| <li class="dropdown"> |
| <a href="#" class="dropdown-toggle" data-toggle="dropdown">API Docs<b class="caret"></b></a> |
| <ul class="dropdown-menu"> |
| <li><a href="api/scala/index.html#org.apache.spark.package">Scala</a></li> |
| <li><a href="api/java/index.html">Java</a></li> |
| <li><a href="api/python/index.html">Python</a></li> |
| <li><a href="api/R/index.html">R</a></li> |
| <li><a href="api/sql/index.html">SQL, Built-in Functions</a></li> |
| </ul> |
| </li> |
| |
| <li class="dropdown"> |
| <a href="#" class="dropdown-toggle" data-toggle="dropdown">Deploying<b class="caret"></b></a> |
| <ul class="dropdown-menu"> |
| <li><a href="cluster-overview.html">Overview</a></li> |
| <li><a href="submitting-applications.html">Submitting Applications</a></li> |
| <li class="divider"></li> |
| <li><a href="spark-standalone.html">Spark Standalone</a></li> |
| <li><a href="running-on-mesos.html">Mesos</a></li> |
| <li><a href="running-on-yarn.html">YARN</a></li> |
| <li><a href="running-on-kubernetes.html">Kubernetes</a></li> |
| </ul> |
| </li> |
| |
| <li class="dropdown"> |
| <a href="api.html" class="dropdown-toggle" data-toggle="dropdown">More<b class="caret"></b></a> |
| <ul class="dropdown-menu"> |
| <li><a href="configuration.html">Configuration</a></li> |
| <li><a href="monitoring.html">Monitoring</a></li> |
| <li><a href="tuning.html">Tuning Guide</a></li> |
| <li><a href="job-scheduling.html">Job Scheduling</a></li> |
| <li><a href="security.html">Security</a></li> |
| <li><a href="hardware-provisioning.html">Hardware Provisioning</a></li> |
| <li><a href="migration-guide.html">Migration Guide</a></li> |
| <li class="divider"></li> |
| <li><a href="building-spark.html">Building Spark</a></li> |
| <li><a href="https://spark.apache.org/contributing.html">Contributing to Spark</a></li> |
| <li><a href="https://spark.apache.org/third-party-projects.html">Third Party Projects</a></li> |
| </ul> |
| </li> |
| </ul> |
| <!--<p class="navbar-text pull-right"><span class="version-text">v3.0.0-preview2</span></p>--> |
| </div> |
| </div> |
| </div> |
| |
| <div class="container-wrapper"> |
| |
| |
| |
| <div class="left-menu-wrapper"> |
| <div class="left-menu"> |
| <h3><a href="migration-guide.html">Migration Guide</a></h3> |
| |
| <ul> |
| |
| <li> |
| <a href="core-migration-guide.html"> |
| |
| Spark Core |
| |
| </a> |
| </li> |
| |
| |
| |
| <li> |
| <a href="sql-migration-guide.html"> |
| |
| <b>SQL, Datasets and DataFrame</b> |
| |
| </a> |
| </li> |
| |
| |
| |
| <li> |
| <a href="ss-migration-guide.html"> |
| |
| Structured Streaming |
| |
| </a> |
| </li> |
| |
| |
| |
| <li> |
| <a href="ml-migration-guide.html"> |
| |
| MLlib (Machine Learning) |
| |
| </a> |
| </li> |
| |
| |
| |
| <li> |
| <a href="pyspark-migration-guide.html"> |
| |
| PySpark (Python on Spark) |
| |
| </a> |
| </li> |
| |
| |
| |
| <li> |
| <a href="sparkr-migration-guide.html"> |
| |
| SparkR (R on Spark) |
| |
| </a> |
| </li> |
| |
| |
| |
| </ul> |
| |
| </div> |
| </div> |
| |
| |
| <input id="nav-trigger" class="nav-trigger" checked type="checkbox"> |
| <label for="nav-trigger"></label> |
| <div class="content-with-sidebar" id="content"> |
| |
| <h1 class="title">Migration Guide: SQL, Datasets and DataFrame</h1> |
| |
| |
| <ul id="markdown-toc"> |
| <li><a href="#upgrading-from-spark-sql-24-to-30" id="markdown-toc-upgrading-from-spark-sql-24-to-30">Upgrading from Spark SQL 2.4 to 3.0</a></li> |
| <li><a href="#upgrading-from-spark-sql-24-to-241" id="markdown-toc-upgrading-from-spark-sql-24-to-241">Upgrading from Spark SQL 2.4 to 2.4.1</a></li> |
| <li><a href="#upgrading-from-spark-sql-23-to-24" id="markdown-toc-upgrading-from-spark-sql-23-to-24">Upgrading from Spark SQL 2.3 to 2.4</a></li> |
| <li><a href="#upgrading-from-spark-sql-22-to-23" id="markdown-toc-upgrading-from-spark-sql-22-to-23">Upgrading from Spark SQL 2.2 to 2.3</a></li> |
| <li><a href="#upgrading-from-spark-sql-21-to-22" id="markdown-toc-upgrading-from-spark-sql-21-to-22">Upgrading from Spark SQL 2.1 to 2.2</a></li> |
| <li><a href="#upgrading-from-spark-sql-20-to-21" id="markdown-toc-upgrading-from-spark-sql-20-to-21">Upgrading from Spark SQL 2.0 to 2.1</a></li> |
| <li><a href="#upgrading-from-spark-sql-16-to-20" id="markdown-toc-upgrading-from-spark-sql-16-to-20">Upgrading from Spark SQL 1.6 to 2.0</a></li> |
| <li><a href="#upgrading-from-spark-sql-15-to-16" id="markdown-toc-upgrading-from-spark-sql-15-to-16">Upgrading from Spark SQL 1.5 to 1.6</a></li> |
| <li><a href="#upgrading-from-spark-sql-14-to-15" id="markdown-toc-upgrading-from-spark-sql-14-to-15">Upgrading from Spark SQL 1.4 to 1.5</a></li> |
| <li><a href="#upgrading-from-spark-sql-13-to-14" id="markdown-toc-upgrading-from-spark-sql-13-to-14">Upgrading from Spark SQL 1.3 to 1.4</a></li> |
| <li><a href="#upgrading-from-spark-sql-10-12-to-13" id="markdown-toc-upgrading-from-spark-sql-10-12-to-13">Upgrading from Spark SQL 1.0-1.2 to 1.3</a></li> |
| <li><a href="#compatibility-with-apache-hive" id="markdown-toc-compatibility-with-apache-hive">Compatibility with Apache Hive</a></li> |
| </ul> |
| |
| <h2 id="upgrading-from-spark-sql-24-to-30">Upgrading from Spark SQL 2.4 to 3.0</h2> |
| <ul> |
| <li> |
| <p>Since Spark 3.0, when inserting a value into a table column with a different data type, the type coercion is performed as per ANSI SQL standard. Certain unreasonable type conversions such as converting <code class="highlighter-rouge">string</code> to <code class="highlighter-rouge">int</code> and <code class="highlighter-rouge">double</code> to <code class="highlighter-rouge">boolean</code> are disallowed. A runtime exception will be thrown if the value is out-of-range for the data type of the column. In Spark version 2.4 and earlier, type conversions during table insertion are allowed as long as they are valid <code class="highlighter-rouge">Cast</code>. When inserting an out-of-range value to a integral field, the low-order bits of the value is inserted(the same as Java/Scala numeric type casting). For example, if 257 is inserted to a field of byte type, the result is 1. The behavior is controlled by the option <code class="highlighter-rouge">spark.sql.storeAssignmentPolicy</code>, with a default value as “ANSI”. Setting the option as “Legacy” restores the previous behavior.</p> |
| </li> |
| <li> |
| <p>In Spark 3.0, the deprecated methods <code class="highlighter-rouge">SQLContext.createExternalTable</code> and <code class="highlighter-rouge">SparkSession.createExternalTable</code> have been removed in favor of its replacement, <code class="highlighter-rouge">createTable</code>.</p> |
| </li> |
| <li> |
| <p>In Spark 3.0, the deprecated <code class="highlighter-rouge">HiveContext</code> class has been removed. Use <code class="highlighter-rouge">SparkSession.builder.enableHiveSupport()</code> instead.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, configuration <code class="highlighter-rouge">spark.sql.crossJoin.enabled</code> become internal configuration, and is true by default, so by default spark won’t raise exception on sql with implicit cross join.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, we reversed argument order of the trim function from <code class="highlighter-rouge">TRIM(trimStr, str)</code> to <code class="highlighter-rouge">TRIM(str, trimStr)</code> to be compatible with other databases.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.4 and earlier, SQL queries such as <code class="highlighter-rouge">FROM <table></code> or <code class="highlighter-rouge">FROM <table> UNION ALL FROM <table></code> are supported by accident. In hive-style <code class="highlighter-rouge">FROM <table> SELECT <expr></code>, the <code class="highlighter-rouge">SELECT</code> clause is not negligible. Neither Hive nor Presto support this syntax. Therefore we will treat these queries as invalid since Spark 3.0.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, the Dataset and DataFrame API <code class="highlighter-rouge">unionAll</code> is not deprecated any more. It is an alias for <code class="highlighter-rouge">union</code>.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.4 and earlier, the parser of JSON data source treats empty strings as null for some data types such as <code class="highlighter-rouge">IntegerType</code>. For <code class="highlighter-rouge">FloatType</code> and <code class="highlighter-rouge">DoubleType</code>, it fails on empty strings and throws exceptions. Since Spark 3.0, we disallow empty strings and will throw exceptions for data types except for <code class="highlighter-rouge">StringType</code> and <code class="highlighter-rouge">BinaryType</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, the <code class="highlighter-rouge">from_json</code> functions supports two modes - <code class="highlighter-rouge">PERMISSIVE</code> and <code class="highlighter-rouge">FAILFAST</code>. The modes can be set via the <code class="highlighter-rouge">mode</code> option. The default mode became <code class="highlighter-rouge">PERMISSIVE</code>. In previous versions, behavior of <code class="highlighter-rouge">from_json</code> did not conform to either <code class="highlighter-rouge">PERMISSIVE</code> nor <code class="highlighter-rouge">FAILFAST</code>, especially in processing of malformed JSON records. For example, the JSON string <code class="highlighter-rouge">{"a" 1}</code> with the schema <code class="highlighter-rouge">a INT</code> is converted to <code class="highlighter-rouge">null</code> by previous versions but Spark 3.0 converts it to <code class="highlighter-rouge">Row(null)</code>.</p> |
| </li> |
| <li> |
| <p>The <code class="highlighter-rouge">ADD JAR</code> command previously returned a result set with the single value 0. It now returns an empty result set.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.4 and earlier, users can create map values with map type key via built-in function like <code class="highlighter-rouge">CreateMap</code>, <code class="highlighter-rouge">MapFromArrays</code>, etc. Since Spark 3.0, it’s not allowed to create map values with map type key with these built-in functions. Users can still read map values with map type key from data source or Java/Scala collections, though they are not very useful.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.4 and earlier, <code class="highlighter-rouge">Dataset.groupByKey</code> results to a grouped dataset with key attribute wrongly named as “value”, if the key is non-struct type, e.g. int, string, array, etc. This is counterintuitive and makes the schema of aggregation queries weird. For example, the schema of <code class="highlighter-rouge">ds.groupByKey(...).count()</code> is <code class="highlighter-rouge">(value, count)</code>. Since Spark 3.0, we name the grouping attribute to “key”. The old behaviour is preserved under a newly added configuration <code class="highlighter-rouge">spark.sql.legacy.dataset.nameNonStructGroupingKeyAsValue</code> with a default value of <code class="highlighter-rouge">false</code>.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.4 and earlier, float/double -0.0 is semantically equal to 0.0, but -0.0 and 0.0 are considered as different values when used in aggregate grouping keys, window partition keys and join keys. Since Spark 3.0, this bug is fixed. For example, <code class="highlighter-rouge">Seq(-0.0, 0.0).toDF("d").groupBy("d").count()</code> returns <code class="highlighter-rouge">[(0.0, 2)]</code> in Spark 3.0, and <code class="highlighter-rouge">[(0.0, 1), (-0.0, 1)]</code> in Spark 2.4 and earlier.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.4 and earlier, users can create a map with duplicated keys via built-in functions like <code class="highlighter-rouge">CreateMap</code>, <code class="highlighter-rouge">StringToMap</code>, etc. The behavior of map with duplicated keys is undefined, e.g. map look up respects the duplicated key appears first, <code class="highlighter-rouge">Dataset.collect</code> only keeps the duplicated key appears last, <code class="highlighter-rouge">MapKeys</code> returns duplicated keys, etc. Since Spark 3.0, these built-in functions will remove duplicated map keys with last wins policy. Users may still read map values with duplicated keys from data sources which do not enforce it (e.g. Parquet), the behavior will be undefined.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.4 and earlier, partition column value is converted as null if it can’t be casted to corresponding user provided schema. Since 3.0, partition column value is validated with user provided schema. An exception is thrown if the validation fails. You can disable such validation by setting <code class="highlighter-rouge">spark.sql.sources.validatePartitionColumns</code> to <code class="highlighter-rouge">false</code>.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.4 and earlier, the <code class="highlighter-rouge">SET</code> command works without any warnings even if the specified key is for <code class="highlighter-rouge">SparkConf</code> entries and it has no effect because the command does not update <code class="highlighter-rouge">SparkConf</code>, but the behavior might confuse users. Since 3.0, the command fails if a <code class="highlighter-rouge">SparkConf</code> key is used. You can disable such a check by setting <code class="highlighter-rouge">spark.sql.legacy.setCommandRejectsSparkCoreConfs</code> to <code class="highlighter-rouge">false</code>.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.4 and earlier, CSV datasource converts a malformed CSV string to a row with all <code class="highlighter-rouge">null</code>s in the PERMISSIVE mode. Since Spark 3.0, the returned row can contain non-<code class="highlighter-rouge">null</code> fields if some of CSV column values were parsed and converted to desired types successfully.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.4 and earlier, JSON datasource and JSON functions like <code class="highlighter-rouge">from_json</code> convert a bad JSON record to a row with all <code class="highlighter-rouge">null</code>s in the PERMISSIVE mode when specified schema is <code class="highlighter-rouge">StructType</code>. Since Spark 3.0, the returned row can contain non-<code class="highlighter-rouge">null</code> fields if some of JSON column values were parsed and converted to desired types successfully.</p> |
| </li> |
| <li> |
| <p>Refreshing a cached table would trigger a table uncache operation and then a table cache (lazily) operation. In Spark version 2.4 and earlier, the cache name and storage level are not preserved before the uncache operation. Therefore, the cache name and storage level could be changed unexpectedly. Since Spark 3.0, cache name and storage level will be first preserved for cache recreation. It helps to maintain a consistent cache behavior upon table refreshing.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, JSON datasource and JSON function <code class="highlighter-rouge">schema_of_json</code> infer TimestampType from string values if they match to the pattern defined by the JSON option <code class="highlighter-rouge">timestampFormat</code>. Set JSON option <code class="highlighter-rouge">inferTimestamp</code> to <code class="highlighter-rouge">false</code> to disable such type inferring.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.4 and earlier, if <code class="highlighter-rouge">org.apache.spark.sql.functions.udf(Any, DataType)</code> gets a Scala closure with primitive-type argument, the returned UDF will return null if the input values is null. Since Spark 3.0, the UDF will return the default value of the Java type if the input value is null. For example, <code class="highlighter-rouge">val f = udf((x: Int) => x, IntegerType)</code>, <code class="highlighter-rouge">f($"x")</code> will return null in Spark 2.4 and earlier if column <code class="highlighter-rouge">x</code> is null, and return 0 in Spark 3.0. This behavior change is introduced because Spark 3.0 is built with Scala 2.12 by default.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, Proleptic Gregorian calendar is used in parsing, formatting, and converting dates and timestamps as well as in extracting sub-components like years, days and etc. Spark 3.0 uses Java 8 API classes from the java.time packages that based on ISO chronology (https://docs.oracle.com/javase/8/docs/api/java/time/chrono/IsoChronology.html). In Spark version 2.4 and earlier, those operations are performed by using the hybrid calendar (Julian + Gregorian, see https://docs.oracle.com/javase/7/docs/api/java/util/GregorianCalendar.html). The changes impact on the results for dates before October 15, 1582 (Gregorian) and affect on the following Spark 3.0 API:</p> |
| |
| <ul> |
| <li> |
| <p>CSV/JSON datasources use java.time API for parsing and generating CSV/JSON content. In Spark version 2.4 and earlier, java.text.SimpleDateFormat is used for the same purpose with fallbacks to the parsing mechanisms of Spark 2.0 and 1.x. For example, <code class="highlighter-rouge">2018-12-08 10:39:21.123</code> with the pattern <code class="highlighter-rouge">yyyy-MM-dd'T'HH:mm:ss.SSS</code> cannot be parsed since Spark 3.0 because the timestamp does not match to the pattern but it can be parsed by earlier Spark versions due to a fallback to <code class="highlighter-rouge">Timestamp.valueOf</code>. To parse the same timestamp since Spark 3.0, the pattern should be <code class="highlighter-rouge">yyyy-MM-dd HH:mm:ss.SSS</code>.</p> |
| </li> |
| <li> |
| <p>The <code class="highlighter-rouge">unix_timestamp</code>, <code class="highlighter-rouge">date_format</code>, <code class="highlighter-rouge">to_unix_timestamp</code>, <code class="highlighter-rouge">from_unixtime</code>, <code class="highlighter-rouge">to_date</code>, <code class="highlighter-rouge">to_timestamp</code> functions. New implementation supports pattern formats as described here https://docs.oracle.com/javase/8/docs/api/java/time/format/DateTimeFormatter.html and performs strict checking of its input. For example, the <code class="highlighter-rouge">2015-07-22 10:00:00</code> timestamp cannot be parse if pattern is <code class="highlighter-rouge">yyyy-MM-dd</code> because the parser does not consume whole input. Another example is the <code class="highlighter-rouge">31/01/2015 00:00</code> input cannot be parsed by the <code class="highlighter-rouge">dd/MM/yyyy hh:mm</code> pattern because <code class="highlighter-rouge">hh</code> supposes hours in the range <code class="highlighter-rouge">1-12</code>.</p> |
| </li> |
| <li> |
| <p>The <code class="highlighter-rouge">weekofyear</code>, <code class="highlighter-rouge">weekday</code>, <code class="highlighter-rouge">dayofweek</code>, <code class="highlighter-rouge">date_trunc</code>, <code class="highlighter-rouge">from_utc_timestamp</code>, <code class="highlighter-rouge">to_utc_timestamp</code>, and <code class="highlighter-rouge">unix_timestamp</code> functions use java.time API for calculation week number of year, day number of week as well for conversion from/to TimestampType values in UTC time zone.</p> |
| </li> |
| <li> |
| <p>the JDBC options <code class="highlighter-rouge">lowerBound</code> and <code class="highlighter-rouge">upperBound</code> are converted to TimestampType/DateType values in the same way as casting strings to TimestampType/DateType values. The conversion is based on Proleptic Gregorian calendar, and time zone defined by the SQL config <code class="highlighter-rouge">spark.sql.session.timeZone</code>. In Spark version 2.4 and earlier, the conversion is based on the hybrid calendar (Julian + Gregorian) and on default system time zone.</p> |
| </li> |
| <li> |
| <p>Formatting of <code class="highlighter-rouge">TIMESTAMP</code> and <code class="highlighter-rouge">DATE</code> literals.</p> |
| </li> |
| </ul> |
| </li> |
| <li> |
| <p>In Spark version 2.4 and earlier, invalid time zone ids are silently ignored and replaced by GMT time zone, for example, in the from_utc_timestamp function. Since Spark 3.0, such time zone ids are rejected, and Spark throws <code class="highlighter-rouge">java.time.DateTimeException</code>.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.4 and earlier, the <code class="highlighter-rouge">current_timestamp</code> function returns a timestamp with millisecond resolution only. Since Spark 3.0, the function can return the result with microsecond resolution if the underlying clock available on the system offers such resolution.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.4 and earlier, when reading a Hive Serde table with Spark native data sources(parquet/orc), Spark will infer the actual file schema and update the table schema in metastore. Since Spark 3.0, Spark doesn’t infer the schema anymore. This should not cause any problems to end users, but if it does, please set <code class="highlighter-rouge">spark.sql.hive.caseSensitiveInferenceMode</code> to <code class="highlighter-rouge">INFER_AND_SAVE</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, <code class="highlighter-rouge">TIMESTAMP</code> literals are converted to strings using the SQL config <code class="highlighter-rouge">spark.sql.session.timeZone</code>. In Spark version 2.4 and earlier, the conversion uses the default time zone of the Java virtual machine.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.4, when a spark session is created via <code class="highlighter-rouge">cloneSession()</code>, the newly created spark session inherits its configuration from its parent <code class="highlighter-rouge">SparkContext</code> even though the same configuration may exist with a different value in its parent spark session. Since Spark 3.0, the configurations of a parent <code class="highlighter-rouge">SparkSession</code> have a higher precedence over the parent <code class="highlighter-rouge">SparkContext</code>. The old behavior can be restored by setting <code class="highlighter-rouge">spark.sql.legacy.sessionInitWithConfigDefaults</code> to <code class="highlighter-rouge">true</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, parquet logical type <code class="highlighter-rouge">TIMESTAMP_MICROS</code> is used by default while saving <code class="highlighter-rouge">TIMESTAMP</code> columns. In Spark version 2.4 and earlier, <code class="highlighter-rouge">TIMESTAMP</code> columns are saved as <code class="highlighter-rouge">INT96</code> in parquet files. To set <code class="highlighter-rouge">INT96</code> to <code class="highlighter-rouge">spark.sql.parquet.outputTimestampType</code> restores the previous behavior.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, if <code class="highlighter-rouge">hive.default.fileformat</code> is not found in <code class="highlighter-rouge">Spark SQL configuration</code> then it will fallback to hive-site.xml present in the <code class="highlighter-rouge">Hadoop configuration</code> of <code class="highlighter-rouge">SparkContext</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, Spark will cast <code class="highlighter-rouge">String</code> to <code class="highlighter-rouge">Date/TimeStamp</code> in binary comparisons with dates/timestamps. The previous behaviour of casting <code class="highlighter-rouge">Date/Timestamp</code> to <code class="highlighter-rouge">String</code> can be restored by setting <code class="highlighter-rouge">spark.sql.legacy.typeCoercion.datetimeToString</code> to <code class="highlighter-rouge">true</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, when Avro files are written with user provided schema, the fields will be matched by field names between catalyst schema and avro schema instead of positions.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, when Avro files are written with user provided non-nullable schema, even the catalyst schema is nullable, Spark is still able to write the files. However, Spark will throw runtime NPE if any of the records contains null.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, we use a new protocol for fetching shuffle blocks, for external shuffle service users, we need to upgrade the server correspondingly. Otherwise, we’ll get the error message <code class="highlighter-rouge">UnsupportedOperationException: Unexpected message: FetchShuffleBlocks</code>. If it is hard to upgrade the shuffle service right now, you can still use the old protocol by setting <code class="highlighter-rouge">spark.shuffle.useOldFetchProtocol</code> to <code class="highlighter-rouge">true</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, a higher-order function <code class="highlighter-rouge">exists</code> follows the three-valued boolean logic, i.e., if the <code class="highlighter-rouge">predicate</code> returns any <code class="highlighter-rouge">null</code>s and no <code class="highlighter-rouge">true</code> is obtained, then <code class="highlighter-rouge">exists</code> will return <code class="highlighter-rouge">null</code> instead of <code class="highlighter-rouge">false</code>. For example, <code class="highlighter-rouge">exists(array(1, null, 3), x -> x % 2 == 0)</code> will be <code class="highlighter-rouge">null</code>. The previous behaviour can be restored by setting <code class="highlighter-rouge">spark.sql.legacy.arrayExistsFollowsThreeValuedLogic</code> to <code class="highlighter-rouge">false</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, if files or subdirectories disappear during recursive directory listing (i.e. they appear in an intermediate listing but then cannot be read or listed during later phases of the recursive directory listing, due to either concurrent file deletions or object store consistency issues) then the listing will fail with an exception unless <code class="highlighter-rouge">spark.sql.files.ignoreMissingFiles</code> is <code class="highlighter-rouge">true</code> (default <code class="highlighter-rouge">false</code>). In previous versions, these missing files or subdirectories would be ignored. Note that this change of behavior only applies during initial table file listing (or during <code class="highlighter-rouge">REFRESH TABLE</code>), not during query execution: the net change is that <code class="highlighter-rouge">spark.sql.files.ignoreMissingFiles</code> is now obeyed during table file listing / query planning, not only at query execution time.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, substitution order of nested WITH clauses is changed and an inner CTE definition takes precedence over an outer. In version 2.4 and earlier, <code class="highlighter-rouge">WITH t AS (SELECT 1), t2 AS (WITH t AS (SELECT 2) SELECT * FROM t) SELECT * FROM t2</code> returns <code class="highlighter-rouge">1</code> while in version 3.0 it returns <code class="highlighter-rouge">2</code>. The previous behaviour can be restored by setting <code class="highlighter-rouge">spark.sql.legacy.ctePrecedence.enabled</code> to <code class="highlighter-rouge">true</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, the <code class="highlighter-rouge">add_months</code> function does not adjust the resulting date to a last day of month if the original date is a last day of months. For example, <code class="highlighter-rouge">select add_months(DATE'2019-02-28', 1)</code> results <code class="highlighter-rouge">2019-03-28</code>. In Spark version 2.4 and earlier, the resulting date is adjusted when the original date is a last day of months. For example, adding a month to <code class="highlighter-rouge">2019-02-28</code> results in <code class="highlighter-rouge">2019-03-31</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, 0-argument Java UDF is executed in the executor side identically with other UDFs. In Spark version 2.4 and earlier, 0-argument Java UDF alone was executed in the driver side, and the result was propagated to executors, which might be more performant in some cases but caused inconsistency with a correctness issue in some cases.</p> |
| </li> |
| <li> |
| <p>The result of <code class="highlighter-rouge">java.lang.Math</code>’s <code class="highlighter-rouge">log</code>, <code class="highlighter-rouge">log1p</code>, <code class="highlighter-rouge">exp</code>, <code class="highlighter-rouge">expm1</code>, and <code class="highlighter-rouge">pow</code> may vary across platforms. In Spark 3.0, the result of the equivalent SQL functions (including related SQL functions like <code class="highlighter-rouge">LOG10</code>) return values consistent with <code class="highlighter-rouge">java.lang.StrictMath</code>. In virtually all cases this makes no difference in the return value, and the difference is very small, but may not exactly match <code class="highlighter-rouge">java.lang.Math</code> on x86 platforms in cases like, for example, <code class="highlighter-rouge">log(3.0)</code>, whose value varies between <code class="highlighter-rouge">Math.log()</code> and <code class="highlighter-rouge">StrictMath.log()</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, Dataset query fails if it contains ambiguous column reference that is caused by self join. A typical example: <code class="highlighter-rouge">val df1 = ...; val df2 = df1.filter(...);</code>, then <code class="highlighter-rouge">df1.join(df2, df1("a") > df2("a"))</code> returns an empty result which is quite confusing. This is because Spark cannot resolve Dataset column references that point to tables being self joined, and <code class="highlighter-rouge">df1("a")</code> is exactly the same as <code class="highlighter-rouge">df2("a")</code> in Spark. To restore the behavior before Spark 3.0, you can set <code class="highlighter-rouge">spark.sql.analyzer.failAmbiguousSelfJoin.enabled</code> to <code class="highlighter-rouge">false</code>.</p> |
| </li> |
| <li>Since Spark 3.0, <code class="highlighter-rouge">Cast</code> function processes string literals such as ‘Infinity’, ‘+Infinity’, ‘-Infinity’, ‘NaN’, ‘Inf’, ‘+Inf’, ‘-Inf’ in case insensitive manner when casting the literals to <code class="highlighter-rouge">Double</code> or <code class="highlighter-rouge">Float</code> type to ensure greater compatibility with other database systems. This behaviour change is illustrated in the table below: |
| <table class="table"> |
| <tr> |
| <th> |
| <b>Operation</b> |
| </th> |
| <th> |
| <b>Result prior to Spark 3.0</b> |
| </th> |
| <th> |
| <b>Result starting Spark 3.0</b> |
| </th> |
| </tr> |
| <tr> |
| <td> |
| CAST('infinity' AS DOUBLE)<br /> |
| CAST('+infinity' AS DOUBLE)<br /> |
| CAST('inf' AS DOUBLE)<br /> |
| CAST('+inf' AS DOUBLE)<br /> |
| </td> |
| <td> |
| NULL |
| </td> |
| <td> |
| Double.PositiveInfinity |
| </td> |
| </tr> |
| <tr> |
| <td> |
| CAST('-infinity' AS DOUBLE)<br /> |
| CAST('-inf' AS DOUBLE)<br /> |
| </td> |
| <td> |
| NULL |
| </td> |
| <td> |
| Double.NegativeInfinity |
| </td> |
| </tr> |
| <tr> |
| <td> |
| CAST('infinity' AS FLOAT)<br /> |
| CAST('+infinity' AS FLOAT)<br /> |
| CAST('inf' AS FLOAT)<br /> |
| CAST('+inf' AS FLOAT)<br /> |
| </td> |
| <td> |
| NULL |
| </td> |
| <td> |
| Float.PositiveInfinity |
| </td> |
| </tr> |
| <tr> |
| <td> |
| CAST('-infinity' AS FLOAT)<br /> |
| CAST('-inf' AS FLOAT)<br /> |
| </td> |
| <td> |
| NULL |
| </td> |
| <td> |
| Float.NegativeInfinity |
| </td> |
| </tr> |
| <tr> |
| <td> |
| CAST('nan' AS DOUBLE) |
| </td> |
| <td> |
| NULL |
| </td> |
| <td> |
| Double.NaN |
| </td> |
| </tr> |
| <tr> |
| <td> |
| CAST('nan' AS FLOAT) |
| </td> |
| <td> |
| NULL |
| </td> |
| <td> |
| Float.NaN |
| </td> |
| </tr> |
| </table> |
| </li> |
| <li>Since Spark 3.0, special values are supported in conversion from strings to dates and timestamps. Those values are simply notational shorthands that will be converted to ordinary date or timestamp values when read. The following string values are supported for dates: |
| <ul> |
| <li><code class="highlighter-rouge">epoch [zoneId]</code> - 1970-01-01</li> |
| <li><code class="highlighter-rouge">today [zoneId]</code> - the current date in the time zone specified by <code class="highlighter-rouge">spark.sql.session.timeZone</code></li> |
| <li><code class="highlighter-rouge">yesterday [zoneId]</code> - the current date - 1</li> |
| <li><code class="highlighter-rouge">tomorrow [zoneId]</code> - the current date + 1</li> |
| <li><code class="highlighter-rouge">now</code> - the date of running the current query. It has the same notion as today |
| For example <code class="highlighter-rouge">SELECT date 'tomorrow' - date 'yesterday';</code> should output <code class="highlighter-rouge">2</code>. Here are special timestamp values:</li> |
| <li><code class="highlighter-rouge">epoch [zoneId]</code> - 1970-01-01 00:00:00+00 (Unix system time zero)</li> |
| <li><code class="highlighter-rouge">today [zoneId]</code> - midnight today</li> |
| <li><code class="highlighter-rouge">yesterday [zoneId]</code> - midnight yesterday</li> |
| <li><code class="highlighter-rouge">tomorrow [zoneId]</code> - midnight tomorrow</li> |
| <li><code class="highlighter-rouge">now</code> - current query start time |
| For example <code class="highlighter-rouge">SELECT timestamp 'tomorrow';</code>.</li> |
| </ul> |
| </li> |
| <li> |
| <p>Since Spark 3.0, the <code class="highlighter-rouge">size</code> function returns <code class="highlighter-rouge">NULL</code> for the <code class="highlighter-rouge">NULL</code> input. In Spark version 2.4 and earlier, this function gives <code class="highlighter-rouge">-1</code> for the same input. To restore the behavior before Spark 3.0, you can set <code class="highlighter-rouge">spark.sql.legacy.sizeOfNull</code> to <code class="highlighter-rouge">true</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, the interval literal syntax does not allow multiple from-to units anymore. For example, <code class="highlighter-rouge">SELECT INTERVAL '1-1' YEAR TO MONTH '2-2' YEAR TO MONTH'</code> throws parser exception.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, when casting interval values to string type, there is no “interval” prefix, e.g. <code class="highlighter-rouge">1 days 2 hours</code>. In Spark version 2.4 and earlier, the string contains the “interval” prefix like <code class="highlighter-rouge">interval 1 days 2 hours</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, when casting string value to integral types(tinyint, smallint, int and bigint), datetime types(date, timestamp and interval) and boolean type, the leading and trailing whitespaces(<= ACSII 32) will be trimmed before converted to these type values, e.g. <code class="highlighter-rouge">cast(' 1\t' as int)</code> results <code class="highlighter-rouge">1</code>, <code class="highlighter-rouge">cast(' 1\t' as boolean)</code> results <code class="highlighter-rouge">true</code>, <code class="highlighter-rouge">cast('2019-10-10\t as date)</code> results the date value <code class="highlighter-rouge">2019-10-10</code>. In Spark version 2.4 and earlier, while casting string to integrals and booleans, it will not trim the whitespaces from both ends, the foregoing results will be <code class="highlighter-rouge">null</code>, while to datetimes, only the trailing spaces(= ASCII 32) will be removed.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, numbers written in scientific notation(e.g. <code class="highlighter-rouge">1E2</code>) would be parsed as Double. In Spark version 2.4 and earlier, they’re parsed as Decimal. To restore the behavior before Spark 3.0, you can set <code class="highlighter-rouge">spark.sql.legacy.exponentLiteralAsDecimal.enabled</code> to <code class="highlighter-rouge">true</code>.</p> |
| </li> |
| <li>Since Spark 3.0, we pad decimal numbers with trailing zeros to the scale of the column for <code class="highlighter-rouge">spark-sql</code> interface, for example: |
| <table class="table"> |
| <tr> |
| <th> |
| <b>Query</b> |
| </th> |
| <th> |
| <b>Spark 2.4 or Prior</b> |
| </th> |
| <th> |
| <b>Spark 3.0</b> |
| </th> |
| </tr> |
| <tr> |
| <td> |
| <code>SELECT CAST(1 AS decimal(38, 18));</code> |
| </td> |
| <td> |
| <code>1</code> |
| </td> |
| <td> |
| <code>1.000000000000000000</code> |
| </td> |
| </tr> |
| </table> |
| </li> |
| <li> |
| <p>Since Spark 3.0, CREATE TABLE without a specific provider will use the value of <code class="highlighter-rouge">spark.sql.sources.default</code> as its provider. In Spark version 2.4 and earlier, it was hive. To restore the behavior before Spark 3.0, you can set <code class="highlighter-rouge">spark.sql.legacy.createHiveTableByDefault.enabled</code> to <code class="highlighter-rouge">true</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 3.0, the unary arithmetic operator plus(<code class="highlighter-rouge">+</code>) only accepts string, numeric and interval type values as inputs. Besides, <code class="highlighter-rouge">+</code> with a integral string representation will be coerced to double value, e.g. <code class="highlighter-rouge">+'1'</code> results <code class="highlighter-rouge">1.0</code>. In Spark version 2.4 and earlier, this operator is ignored. There is no type checking for it, thus, all type values with a <code class="highlighter-rouge">+</code> prefix are valid, e.g. <code class="highlighter-rouge">+ array(1, 2)</code> is valid and results <code class="highlighter-rouge">[1, 2]</code>. Besides, there is no type coercion for it at all, e.g. in Spark 2.4, the result of <code class="highlighter-rouge">+'1'</code> is string <code class="highlighter-rouge">1</code>.</p> |
| </li> |
| <li>Since Spark 3.0, day-time interval strings are converted to intervals with respect to the <code class="highlighter-rouge">from</code> and <code class="highlighter-rouge">to</code> bounds. If an input string does not match to the pattern defined by specified bounds, the <code class="highlighter-rouge">ParseException</code> exception is thrown. For example, <code class="highlighter-rouge">interval '2 10:20' hour to minute</code> raises the exception because the expected format is <code class="highlighter-rouge">[+|-]h[h]:[m]m</code>. In Spark version 2.4, the <code class="highlighter-rouge">from</code> bound was not taken into account, and the <code class="highlighter-rouge">to</code> bound was used to truncate the resulted interval. For instance, the day-time interval string from the showed example is converted to <code class="highlighter-rouge">interval 10 hours 20 minutes</code>. To restore the behavior before Spark 3.0, you can set <code class="highlighter-rouge">spark.sql.legacy.fromDayTimeString.enabled</code> to <code class="highlighter-rouge">true</code>.</li> |
| </ul> |
| |
| <h2 id="upgrading-from-spark-sql-24-to-241">Upgrading from Spark SQL 2.4 to 2.4.1</h2> |
| |
| <ul> |
| <li> |
| <p>The value of <code class="highlighter-rouge">spark.executor.heartbeatInterval</code>, when specified without units like “30” rather than “30s”, was |
| inconsistently interpreted as both seconds and milliseconds in Spark 2.4.0 in different parts of the code. |
| Unitless values are now consistently interpreted as milliseconds. Applications that set values like “30” |
| need to specify a value with units like “30s” now, to avoid being interpreted as milliseconds; otherwise, |
| the extremely short interval that results will likely cause applications to fail.</p> |
| </li> |
| <li> |
| <p>When turning a Dataset to another Dataset, Spark will up cast the fields in the original Dataset to the type of corresponding fields in the target DataSet. In version 2.4 and earlier, this up cast is not very strict, e.g. <code class="highlighter-rouge">Seq("str").toDS.as[Int]</code> fails, but <code class="highlighter-rouge">Seq("str").toDS.as[Boolean]</code> works and throw NPE during execution. In Spark 3.0, the up cast is stricter and turning String into something else is not allowed, i.e. <code class="highlighter-rouge">Seq("str").toDS.as[Boolean]</code> will fail during analysis.</p> |
| </li> |
| </ul> |
| |
| <h2 id="upgrading-from-spark-sql-23-to-24">Upgrading from Spark SQL 2.3 to 2.4</h2> |
| |
| <ul> |
| <li>In Spark version 2.3 and earlier, the second parameter to array_contains function is implicitly promoted to the element type of first array type parameter. This type promotion can be lossy and may cause <code class="highlighter-rouge">array_contains</code> function to return wrong result. This problem has been addressed in 2.4 by employing a safer type promotion mechanism. This can cause some change in behavior and are illustrated in the table below. |
| <table class="table"> |
| <tr> |
| <th> |
| <b>Query</b> |
| </th> |
| <th> |
| <b>Spark 2.3 or Prior</b> |
| </th> |
| <th> |
| <b>Spark 2.4</b> |
| </th> |
| <th> |
| <b>Remarks</b> |
| </th> |
| </tr> |
| <tr> |
| <td> |
| <code>SELECT array_contains(array(1), 1.34D);</code> |
| </td> |
| <td> |
| <code>true</code> |
| </td> |
| <td> |
| <code>false</code> |
| </td> |
| <td> |
| In Spark 2.4, left and right parameters are promoted to array type of double type and double type respectively. |
| </td> |
| </tr> |
| <tr> |
| <td> |
| <code>SELECT array_contains(array(1), '1');</code> |
| </td> |
| <td> |
| <code>true</code> |
| </td> |
| <td> |
| <code>AnalysisException</code> is thrown. |
| </td> |
| <td> |
| Explicit cast can be used in arguments to avoid the exception. In Spark 2.4, <code>AnalysisException</code> is thrown since integer type can not be promoted to string type in a loss-less manner. |
| </td> |
| </tr> |
| <tr> |
| <td> |
| <code>SELECT array_contains(array(1), 'anystring');</code> |
| </td> |
| <td> |
| <code>null</code> |
| </td> |
| <td> |
| <code>AnalysisException</code> is thrown. |
| </td> |
| <td> |
| Explicit cast can be used in arguments to avoid the exception. In Spark 2.4, <code>AnalysisException</code> is thrown since integer type can not be promoted to string type in a loss-less manner. |
| </td> |
| </tr> |
| </table> |
| </li> |
| <li> |
| <p>Since Spark 2.4, when there is a struct field in front of the IN operator before a subquery, the inner query must contain a struct field as well. In previous versions, instead, the fields of the struct were compared to the output of the inner query. Eg. if <code class="highlighter-rouge">a</code> is a <code class="highlighter-rouge">struct(a string, b int)</code>, in Spark 2.4 <code class="highlighter-rouge">a in (select (1 as a, 'a' as b) from range(1))</code> is a valid query, while <code class="highlighter-rouge">a in (select 1, 'a' from range(1))</code> is not. In previous version it was the opposite.</p> |
| </li> |
| <li> |
| <p>In versions 2.2.1+ and 2.3, if <code class="highlighter-rouge">spark.sql.caseSensitive</code> is set to true, then the <code class="highlighter-rouge">CURRENT_DATE</code> and <code class="highlighter-rouge">CURRENT_TIMESTAMP</code> functions incorrectly became case-sensitive and would resolve to columns (unless typed in lower case). In Spark 2.4 this has been fixed and the functions are no longer case-sensitive.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, Spark will evaluate the set operations referenced in a query by following a precedence rule as per the SQL standard. If the order is not specified by parentheses, set operations are performed from left to right with the exception that all INTERSECT operations are performed before any UNION, EXCEPT or MINUS operations. The old behaviour of giving equal precedence to all the set operations are preserved under a newly added configuration <code class="highlighter-rouge">spark.sql.legacy.setopsPrecedence.enabled</code> with a default value of <code class="highlighter-rouge">false</code>. When this property is set to <code class="highlighter-rouge">true</code>, spark will evaluate the set operators from left to right as they appear in the query given no explicit ordering is enforced by usage of parenthesis.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, Spark will display table description column Last Access value as UNKNOWN when the value was Jan 01 1970.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, Spark maximizes the usage of a vectorized ORC reader for ORC files by default. To do that, <code class="highlighter-rouge">spark.sql.orc.impl</code> and <code class="highlighter-rouge">spark.sql.orc.filterPushdown</code> change their default values to <code class="highlighter-rouge">native</code> and <code class="highlighter-rouge">true</code> respectively. ORC files created by native ORC writer cannot be read by some old Apache Hive releases. Use <code class="highlighter-rouge">spark.sql.orc.impl=hive</code> to create the files shared with Hive 2.1.1 and older.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, writing an empty dataframe to a directory launches at least one write task, even if physically the dataframe has no partition. This introduces a small behavior change that for self-describing file formats like Parquet and Orc, Spark creates a metadata-only file in the target directory when writing a 0-partition dataframe, so that schema inference can still work if users read that directory later. The new behavior is more reasonable and more consistent regarding writing empty dataframe.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, expression IDs in UDF arguments do not appear in column names. For example, a column name in Spark 2.4 is not <code class="highlighter-rouge">UDF:f(col0 AS colA#28)</code> but <code class="highlighter-rouge">UDF:f(col0 AS `colA`)</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, writing a dataframe with an empty or nested empty schema using any file formats (parquet, orc, json, text, csv etc.) is not allowed. An exception is thrown when attempting to write dataframes with empty schema.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, Spark compares a DATE type with a TIMESTAMP type after promotes both sides to TIMESTAMP. To set <code class="highlighter-rouge">false</code> to <code class="highlighter-rouge">spark.sql.legacy.compareDateTimestampInTimestamp</code> restores the previous behavior. This option will be removed in Spark 3.0.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, creating a managed table with nonempty location is not allowed. An exception is thrown when attempting to create a managed table with nonempty location. To set <code class="highlighter-rouge">true</code> to <code class="highlighter-rouge">spark.sql.legacy.allowCreatingManagedTableUsingNonemptyLocation</code> restores the previous behavior. This option will be removed in Spark 3.0.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, renaming a managed table to existing location is not allowed. An exception is thrown when attempting to rename a managed table to existing location.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, the type coercion rules can automatically promote the argument types of the variadic SQL functions (e.g., IN/COALESCE) to the widest common type, no matter how the input arguments order. In prior Spark versions, the promotion could fail in some specific orders (e.g., TimestampType, IntegerType and StringType) and throw an exception.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, Spark has enabled non-cascading SQL cache invalidation in addition to the traditional cache invalidation mechanism. The non-cascading cache invalidation mechanism allows users to remove a cache without impacting its dependent caches. This new cache invalidation mechanism is used in scenarios where the data of the cache to be removed is still valid, e.g., calling unpersist() on a Dataset, or dropping a temporary view. This allows users to free up memory and keep the desired caches valid at the same time.</p> |
| </li> |
| <li> |
| <p>In version 2.3 and earlier, Spark converts Parquet Hive tables by default but ignores table properties like <code class="highlighter-rouge">TBLPROPERTIES (parquet.compression 'NONE')</code>. This happens for ORC Hive table properties like <code class="highlighter-rouge">TBLPROPERTIES (orc.compress 'NONE')</code> in case of <code class="highlighter-rouge">spark.sql.hive.convertMetastoreOrc=true</code>, too. Since Spark 2.4, Spark respects Parquet/ORC specific table properties while converting Parquet/ORC Hive tables. As an example, <code class="highlighter-rouge">CREATE TABLE t(id int) STORED AS PARQUET TBLPROPERTIES (parquet.compression 'NONE')</code> would generate Snappy parquet files during insertion in Spark 2.3, and in Spark 2.4, the result would be uncompressed parquet files.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.0, Spark converts Parquet Hive tables by default for better performance. Since Spark 2.4, Spark converts ORC Hive tables by default, too. It means Spark uses its own ORC support by default instead of Hive SerDe. As an example, <code class="highlighter-rouge">CREATE TABLE t(id int) STORED AS ORC</code> would be handled with Hive SerDe in Spark 2.3, and in Spark 2.4, it would be converted into Spark’s ORC data source table and ORC vectorization would be applied. To set <code class="highlighter-rouge">false</code> to <code class="highlighter-rouge">spark.sql.hive.convertMetastoreOrc</code> restores the previous behavior.</p> |
| </li> |
| <li> |
| <p>In version 2.3 and earlier, CSV rows are considered as malformed if at least one column value in the row is malformed. CSV parser dropped such rows in the DROPMALFORMED mode or outputs an error in the FAILFAST mode. Since Spark 2.4, CSV row is considered as malformed only when it contains malformed column values requested from CSV datasource, other values can be ignored. As an example, CSV file contains the “id,name” header and one row “1234”. In Spark 2.4, selection of the id column consists of a row with one column value 1234 but in Spark 2.3 and earlier it is empty in the DROPMALFORMED mode. To restore the previous behavior, set <code class="highlighter-rouge">spark.sql.csv.parser.columnPruning.enabled</code> to <code class="highlighter-rouge">false</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, File listing for compute statistics is done in parallel by default. This can be disabled by setting <code class="highlighter-rouge">spark.sql.statistics.parallelFileListingInStatsComputation.enabled</code> to <code class="highlighter-rouge">False</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, Metadata files (e.g. Parquet summary files) and temporary files are not counted as data files when calculating table size during Statistics computation.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, empty strings are saved as quoted empty strings <code class="highlighter-rouge">""</code>. In version 2.3 and earlier, empty strings are equal to <code class="highlighter-rouge">null</code> values and do not reflect to any characters in saved CSV files. For example, the row of <code class="highlighter-rouge">"a", null, "", 1</code> was written as <code class="highlighter-rouge">a,,,1</code>. Since Spark 2.4, the same row is saved as <code class="highlighter-rouge">a,,"",1</code>. To restore the previous behavior, set the CSV option <code class="highlighter-rouge">emptyValue</code> to empty (not quoted) string.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.4, The LOAD DATA command supports wildcard <code class="highlighter-rouge">?</code> and <code class="highlighter-rouge">*</code>, which match any one character, and zero or more characters, respectively. Example: <code class="highlighter-rouge">LOAD DATA INPATH '/tmp/folder*/'</code> or <code class="highlighter-rouge">LOAD DATA INPATH '/tmp/part-?'</code>. Special Characters like <code class="highlighter-rouge">space</code> also now work in paths. Example: <code class="highlighter-rouge">LOAD DATA INPATH '/tmp/folder name/'</code>.</p> |
| </li> |
| <li> |
| <p>In Spark version 2.3 and earlier, HAVING without GROUP BY is treated as WHERE. This means, <code class="highlighter-rouge">SELECT 1 FROM range(10) HAVING true</code> is executed as <code class="highlighter-rouge">SELECT 1 FROM range(10) WHERE true</code> and returns 10 rows. This violates SQL standard, and has been fixed in Spark 2.4. Since Spark 2.4, HAVING without GROUP BY is treated as a global aggregate, which means <code class="highlighter-rouge">SELECT 1 FROM range(10) HAVING true</code> will return only one row. To restore the previous behavior, set <code class="highlighter-rouge">spark.sql.legacy.parser.havingWithoutGroupByAsWhere</code> to <code class="highlighter-rouge">true</code>.</p> |
| </li> |
| <li>In version 2.3 and earlier, when reading from a Parquet data source table, Spark always returns null for any column whose column names in Hive metastore schema and Parquet schema are in different letter cases, no matter whether <code class="highlighter-rouge">spark.sql.caseSensitive</code> is set to <code class="highlighter-rouge">true</code> or <code class="highlighter-rouge">false</code>. Since 2.4, when <code class="highlighter-rouge">spark.sql.caseSensitive</code> is set to <code class="highlighter-rouge">false</code>, Spark does case insensitive column name resolution between Hive metastore schema and Parquet schema, so even column names are in different letter cases, Spark returns corresponding column values. An exception is thrown if there is ambiguity, i.e. more than one Parquet column is matched. This change also applies to Parquet Hive tables when <code class="highlighter-rouge">spark.sql.hive.convertMetastoreParquet</code> is set to <code class="highlighter-rouge">true</code>.</li> |
| </ul> |
| |
| <h2 id="upgrading-from-spark-sql-22-to-23">Upgrading from Spark SQL 2.2 to 2.3</h2> |
| |
| <ul> |
| <li> |
| <p>Since Spark 2.3, the queries from raw JSON/CSV files are disallowed when the referenced columns only include the internal corrupt record column (named <code class="highlighter-rouge">_corrupt_record</code> by default). For example, <code class="highlighter-rouge">spark.read.schema(schema).json(file).filter($"_corrupt_record".isNotNull).count()</code> and <code class="highlighter-rouge">spark.read.schema(schema).json(file).select("_corrupt_record").show()</code>. Instead, you can cache or save the parsed results and then send the same query. For example, <code class="highlighter-rouge">val df = spark.read.schema(schema).json(file).cache()</code> and then <code class="highlighter-rouge">df.filter($"_corrupt_record".isNotNull).count()</code>.</p> |
| </li> |
| <li> |
| <p>The <code class="highlighter-rouge">percentile_approx</code> function previously accepted numeric type input and output double type results. Now it supports date type, timestamp type and numeric types as input types. The result type is also changed to be the same as the input type, which is more reasonable for percentiles.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.3, the Join/Filter’s deterministic predicates that are after the first non-deterministic predicates are also pushed down/through the child operators, if possible. In prior Spark versions, these filters are not eligible for predicate pushdown.</p> |
| </li> |
| <li>Partition column inference previously found incorrect common type for different inferred types, for example, previously it ended up with double type as the common type for double type and date type. Now it finds the correct common type for such conflicts. The conflict resolution follows the table below: |
| <table class="table"> |
| <tr> |
| <th> |
| <b>InputA \ InputB</b> |
| </th> |
| <th> |
| <b>NullType</b> |
| </th> |
| <th> |
| <b>IntegerType</b> |
| </th> |
| <th> |
| <b>LongType</b> |
| </th> |
| <th> |
| <b>DecimalType(38,0)*</b> |
| </th> |
| <th> |
| <b>DoubleType</b> |
| </th> |
| <th> |
| <b>DateType</b> |
| </th> |
| <th> |
| <b>TimestampType</b> |
| </th> |
| <th> |
| <b>StringType</b> |
| </th> |
| </tr> |
| <tr> |
| <td> |
| <b>NullType</b> |
| </td> |
| <td>NullType</td> |
| <td>IntegerType</td> |
| <td>LongType</td> |
| <td>DecimalType(38,0)</td> |
| <td>DoubleType</td> |
| <td>DateType</td> |
| <td>TimestampType</td> |
| <td>StringType</td> |
| </tr> |
| <tr> |
| <td> |
| <b>IntegerType</b> |
| </td> |
| <td>IntegerType</td> |
| <td>IntegerType</td> |
| <td>LongType</td> |
| <td>DecimalType(38,0)</td> |
| <td>DoubleType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| </tr> |
| <tr> |
| <td> |
| <b>LongType</b> |
| </td> |
| <td>LongType</td> |
| <td>LongType</td> |
| <td>LongType</td> |
| <td>DecimalType(38,0)</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| </tr> |
| <tr> |
| <td> |
| <b>DecimalType(38,0)*</b> |
| </td> |
| <td>DecimalType(38,0)</td> |
| <td>DecimalType(38,0)</td> |
| <td>DecimalType(38,0)</td> |
| <td>DecimalType(38,0)</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| </tr> |
| <tr> |
| <td> |
| <b>DoubleType</b> |
| </td> |
| <td>DoubleType</td> |
| <td>DoubleType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>DoubleType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| </tr> |
| <tr> |
| <td> |
| <b>DateType</b> |
| </td> |
| <td>DateType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>DateType</td> |
| <td>TimestampType</td> |
| <td>StringType</td> |
| </tr> |
| <tr> |
| <td> |
| <b>TimestampType</b> |
| </td> |
| <td>TimestampType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>TimestampType</td> |
| <td>TimestampType</td> |
| <td>StringType</td> |
| </tr> |
| <tr> |
| <td> |
| <b>StringType</b> |
| </td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| <td>StringType</td> |
| </tr> |
| </table> |
| |
| <p>Note that, for <b>DecimalType(38,0)*</b>, the table above intentionally does not cover all other combinations of scales and precisions because currently we only infer decimal type like <code class="highlighter-rouge">BigInteger</code>/<code class="highlighter-rouge">BigInt</code>. For example, 1.1 is inferred as double type.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.3, when either broadcast hash join or broadcast nested loop join is applicable, we prefer to broadcasting the table that is explicitly specified in a broadcast hint. For details, see the section <a href="sql-performance-tuning.html#broadcast-hint-for-sql-queries">Broadcast Hint</a> and <a href="https://issues.apache.org/jira/browse/SPARK-22489">SPARK-22489</a>.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.3, when all inputs are binary, <code class="highlighter-rouge">functions.concat()</code> returns an output as binary. Otherwise, it returns as a string. Until Spark 2.3, it always returns as a string despite of input types. To keep the old behavior, set <code class="highlighter-rouge">spark.sql.function.concatBinaryAsString</code> to <code class="highlighter-rouge">true</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.3, when all inputs are binary, SQL <code class="highlighter-rouge">elt()</code> returns an output as binary. Otherwise, it returns as a string. Until Spark 2.3, it always returns as a string despite of input types. To keep the old behavior, set <code class="highlighter-rouge">spark.sql.function.eltOutputAsString</code> to <code class="highlighter-rouge">true</code>.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.3, by default arithmetic operations between decimals return a rounded value if an exact representation is not possible (instead of returning NULL). This is compliant with SQL ANSI 2011 specification and Hive’s new behavior introduced in Hive 2.2 (HIVE-15331). This involves the following changes</p> |
| |
| <ul> |
| <li> |
| <p>The rules to determine the result type of an arithmetic operation have been updated. In particular, if the precision / scale needed are out of the range of available values, the scale is reduced up to 6, in order to prevent the truncation of the integer part of the decimals. All the arithmetic operations are affected by the change, ie. addition (<code class="highlighter-rouge">+</code>), subtraction (<code class="highlighter-rouge">-</code>), multiplication (<code class="highlighter-rouge">*</code>), division (<code class="highlighter-rouge">/</code>), remainder (<code class="highlighter-rouge">%</code>) and positive module (<code class="highlighter-rouge">pmod</code>).</p> |
| </li> |
| <li> |
| <p>Literal values used in SQL operations are converted to DECIMAL with the exact precision and scale needed by them.</p> |
| </li> |
| <li> |
| <p>The configuration <code class="highlighter-rouge">spark.sql.decimalOperations.allowPrecisionLoss</code> has been introduced. It defaults to <code class="highlighter-rouge">true</code>, which means the new behavior described here; if set to <code class="highlighter-rouge">false</code>, Spark uses previous rules, ie. it doesn’t adjust the needed scale to represent the values and it returns NULL if an exact representation of the value is not possible.</p> |
| </li> |
| </ul> |
| </li> |
| <li> |
| <p>Un-aliased subquery’s semantic has not been well defined with confusing behaviors. Since Spark 2.3, we invalidate such confusing cases, for example: <code class="highlighter-rouge">SELECT v.i from (SELECT i FROM v)</code>, Spark will throw an analysis exception in this case because users should not be able to use the qualifier inside a subquery. See <a href="https://issues.apache.org/jira/browse/SPARK-20690">SPARK-20690</a> and <a href="https://issues.apache.org/jira/browse/SPARK-21335">SPARK-21335</a> for more details.</p> |
| </li> |
| <li>When creating a <code class="highlighter-rouge">SparkSession</code> with <code class="highlighter-rouge">SparkSession.builder.getOrCreate()</code>, if there is an existing <code class="highlighter-rouge">SparkContext</code>, the builder was trying to update the <code class="highlighter-rouge">SparkConf</code> of the existing <code class="highlighter-rouge">SparkContext</code> with configurations specified to the builder, but the <code class="highlighter-rouge">SparkContext</code> is shared by all <code class="highlighter-rouge">SparkSession</code>s, so we should not update them. Since 2.3, the builder comes to not update the configurations. If you want to update them, you need to update them prior to creating a <code class="highlighter-rouge">SparkSession</code>.</li> |
| </ul> |
| |
| <h2 id="upgrading-from-spark-sql-21-to-22">Upgrading from Spark SQL 2.1 to 2.2</h2> |
| |
| <ul> |
| <li> |
| <p>Spark 2.1.1 introduced a new configuration key: <code class="highlighter-rouge">spark.sql.hive.caseSensitiveInferenceMode</code>. It had a default setting of <code class="highlighter-rouge">NEVER_INFER</code>, which kept behavior identical to 2.1.0. However, Spark 2.2.0 changes this setting’s default value to <code class="highlighter-rouge">INFER_AND_SAVE</code> to restore compatibility with reading Hive metastore tables whose underlying file schema have mixed-case column names. With the <code class="highlighter-rouge">INFER_AND_SAVE</code> configuration value, on first access Spark will perform schema inference on any Hive metastore table for which it has not already saved an inferred schema. Note that schema inference can be a very time-consuming operation for tables with thousands of partitions. If compatibility with mixed-case column names is not a concern, you can safely set <code class="highlighter-rouge">spark.sql.hive.caseSensitiveInferenceMode</code> to <code class="highlighter-rouge">NEVER_INFER</code> to avoid the initial overhead of schema inference. Note that with the new default <code class="highlighter-rouge">INFER_AND_SAVE</code> setting, the results of the schema inference are saved as a metastore key for future use. Therefore, the initial schema inference occurs only at a table’s first access.</p> |
| </li> |
| <li> |
| <p>Since Spark 2.2.1 and 2.3.0, the schema is always inferred at runtime when the data source tables have the columns that exist in both partition schema and data schema. The inferred schema does not have the partitioned columns. When reading the table, Spark respects the partition values of these overlapping columns instead of the values stored in the data source files. In 2.2.0 and 2.1.x release, the inferred schema is partitioned but the data of the table is invisible to users (i.e., the result set is empty).</p> |
| </li> |
| <li> |
| <p>Since Spark 2.2, view definitions are stored in a different way from prior versions. This may cause Spark unable to read views created by prior versions. In such cases, you need to recreate the views using <code class="highlighter-rouge">ALTER VIEW AS</code> or <code class="highlighter-rouge">CREATE OR REPLACE VIEW AS</code> with newer Spark versions.</p> |
| </li> |
| </ul> |
| |
| <h2 id="upgrading-from-spark-sql-20-to-21">Upgrading from Spark SQL 2.0 to 2.1</h2> |
| |
| <ul> |
| <li> |
| <p>Datasource tables now store partition metadata in the Hive metastore. This means that Hive DDLs such as <code class="highlighter-rouge">ALTER TABLE PARTITION ... SET LOCATION</code> are now available for tables created with the Datasource API.</p> |
| |
| <ul> |
| <li> |
| <p>Legacy datasource tables can be migrated to this format via the <code class="highlighter-rouge">MSCK REPAIR TABLE</code> command. Migrating legacy tables is recommended to take advantage of Hive DDL support and improved planning performance.</p> |
| </li> |
| <li> |
| <p>To determine if a table has been migrated, look for the <code class="highlighter-rouge">PartitionProvider: Catalog</code> attribute when issuing <code class="highlighter-rouge">DESCRIBE FORMATTED</code> on the table.</p> |
| </li> |
| </ul> |
| </li> |
| <li> |
| <p>Changes to <code class="highlighter-rouge">INSERT OVERWRITE TABLE ... PARTITION ...</code> behavior for Datasource tables.</p> |
| |
| <ul> |
| <li> |
| <p>In prior Spark versions <code class="highlighter-rouge">INSERT OVERWRITE</code> overwrote the entire Datasource table, even when given a partition specification. Now only partitions matching the specification are overwritten.</p> |
| </li> |
| <li> |
| <p>Note that this still differs from the behavior of Hive tables, which is to overwrite only partitions overlapping with newly inserted data.</p> |
| </li> |
| </ul> |
| </li> |
| </ul> |
| |
| <h2 id="upgrading-from-spark-sql-16-to-20">Upgrading from Spark SQL 1.6 to 2.0</h2> |
| |
| <ul> |
| <li> |
| <p><code class="highlighter-rouge">SparkSession</code> is now the new entry point of Spark that replaces the old <code class="highlighter-rouge">SQLContext</code> and</p> |
| |
| <p><code class="highlighter-rouge">HiveContext</code>. Note that the old SQLContext and HiveContext are kept for backward compatibility. A new <code class="highlighter-rouge">catalog</code> interface is accessible from <code class="highlighter-rouge">SparkSession</code> - existing API on databases and tables access such as <code class="highlighter-rouge">listTables</code>, <code class="highlighter-rouge">createExternalTable</code>, <code class="highlighter-rouge">dropTempView</code>, <code class="highlighter-rouge">cacheTable</code> are moved here.</p> |
| </li> |
| <li> |
| <p>Dataset API and DataFrame API are unified. In Scala, <code class="highlighter-rouge">DataFrame</code> becomes a type alias for |
| <code class="highlighter-rouge">Dataset[Row]</code>, while Java API users must replace <code class="highlighter-rouge">DataFrame</code> with <code class="highlighter-rouge">Dataset<Row></code>. Both the typed |
| transformations (e.g., <code class="highlighter-rouge">map</code>, <code class="highlighter-rouge">filter</code>, and <code class="highlighter-rouge">groupByKey</code>) and untyped transformations (e.g., |
| <code class="highlighter-rouge">select</code> and <code class="highlighter-rouge">groupBy</code>) are available on the Dataset class. Since compile-time type-safety in |
| Python and R is not a language feature, the concept of Dataset does not apply to these languages’ |
| APIs. Instead, <code class="highlighter-rouge">DataFrame</code> remains the primary programming abstraction, which is analogous to the |
| single-node data frame notion in these languages.</p> |
| </li> |
| <li> |
| <p>Dataset and DataFrame API <code class="highlighter-rouge">unionAll</code> has been deprecated and replaced by <code class="highlighter-rouge">union</code></p> |
| </li> |
| <li> |
| <p>Dataset and DataFrame API <code class="highlighter-rouge">explode</code> has been deprecated, alternatively, use <code class="highlighter-rouge">functions.explode()</code> with <code class="highlighter-rouge">select</code> or <code class="highlighter-rouge">flatMap</code></p> |
| </li> |
| <li> |
| <p>Dataset and DataFrame API <code class="highlighter-rouge">registerTempTable</code> has been deprecated and replaced by <code class="highlighter-rouge">createOrReplaceTempView</code></p> |
| </li> |
| <li> |
| <p>Changes to <code class="highlighter-rouge">CREATE TABLE ... LOCATION</code> behavior for Hive tables.</p> |
| |
| <ul> |
| <li> |
| <p>From Spark 2.0, <code class="highlighter-rouge">CREATE TABLE ... LOCATION</code> is equivalent to <code class="highlighter-rouge">CREATE EXTERNAL TABLE ... LOCATION</code> |
| in order to prevent accidental dropping the existing data in the user-provided locations. |
| That means, a Hive table created in Spark SQL with the user-specified location is always a Hive external table. |
| Dropping external tables will not remove the data. Users are not allowed to specify the location for Hive managed tables. |
| Note that this is different from the Hive behavior.</p> |
| </li> |
| <li> |
| <p>As a result, <code class="highlighter-rouge">DROP TABLE</code> statements on those tables will not remove the data.</p> |
| </li> |
| </ul> |
| </li> |
| <li> |
| <p><code class="highlighter-rouge">spark.sql.parquet.cacheMetadata</code> is no longer used. |
| See <a href="https://issues.apache.org/jira/browse/SPARK-13664">SPARK-13664</a> for details.</p> |
| </li> |
| </ul> |
| |
| <h2 id="upgrading-from-spark-sql-15-to-16">Upgrading from Spark SQL 1.5 to 1.6</h2> |
| |
| <ul> |
| <li>From Spark 1.6, by default, the Thrift server runs in multi-session mode. Which means each JDBC/ODBC |
| connection owns a copy of their own SQL configuration and temporary function registry. Cached |
| tables are still shared though. If you prefer to run the Thrift server in the old single-session |
| mode, please set option <code class="highlighter-rouge">spark.sql.hive.thriftServer.singleSession</code> to <code class="highlighter-rouge">true</code>. You may either add |
| this option to <code class="highlighter-rouge">spark-defaults.conf</code>, or pass it to <code class="highlighter-rouge">start-thriftserver.sh</code> via <code class="highlighter-rouge">--conf</code>:</li> |
| </ul> |
| |
| <figure class="highlight"><pre><code class="language-bash" data-lang="bash"> ./sbin/start-thriftserver.sh <span class="se">\</span> |
| <span class="nt">--conf</span> spark.sql.hive.thriftServer.singleSession<span class="o">=</span><span class="nb">true</span> <span class="se">\</span> |
| ... |
| </code></pre></figure> |
| |
| <ul> |
| <li>From Spark 1.6, LongType casts to TimestampType expect seconds instead of microseconds. This |
| change was made to match the behavior of Hive 1.2 for more consistent type casting to TimestampType |
| from numeric types. See <a href="https://issues.apache.org/jira/browse/SPARK-11724">SPARK-11724</a> for |
| details.</li> |
| </ul> |
| |
| <h2 id="upgrading-from-spark-sql-14-to-15">Upgrading from Spark SQL 1.4 to 1.5</h2> |
| |
| <ul> |
| <li> |
| <p>Optimized execution using manually managed memory (Tungsten) is now enabled by default, along with |
| code generation for expression evaluation. These features can both be disabled by setting |
| <code class="highlighter-rouge">spark.sql.tungsten.enabled</code> to <code class="highlighter-rouge">false</code>.</p> |
| </li> |
| <li> |
| <p>Parquet schema merging is no longer enabled by default. It can be re-enabled by setting |
| <code class="highlighter-rouge">spark.sql.parquet.mergeSchema</code> to <code class="highlighter-rouge">true</code>.</p> |
| </li> |
| <li> |
| <p>In-memory columnar storage partition pruning is on by default. It can be disabled by setting |
| <code class="highlighter-rouge">spark.sql.inMemoryColumnarStorage.partitionPruning</code> to <code class="highlighter-rouge">false</code>.</p> |
| </li> |
| <li> |
| <p>Unlimited precision decimal columns are no longer supported, instead Spark SQL enforces a maximum |
| precision of 38. When inferring schema from <code class="highlighter-rouge">BigDecimal</code> objects, a precision of (38, 18) is now |
| used. When no precision is specified in DDL then the default remains <code class="highlighter-rouge">Decimal(10, 0)</code>.</p> |
| </li> |
| <li> |
| <p>Timestamps are now stored at a precision of 1us, rather than 1ns</p> |
| </li> |
| <li> |
| <p>In the <code class="highlighter-rouge">sql</code> dialect, floating point numbers are now parsed as decimal. HiveQL parsing remains |
| unchanged.</p> |
| </li> |
| <li> |
| <p>The canonical name of SQL/DataFrame functions are now lower case (e.g., sum vs SUM).</p> |
| </li> |
| <li> |
| <p>JSON data source will not automatically load new files that are created by other applications |
| (i.e. files that are not inserted to the dataset through Spark SQL). |
| For a JSON persistent table (i.e. the metadata of the table is stored in Hive Metastore), |
| users can use <code class="highlighter-rouge">REFRESH TABLE</code> SQL command or <code class="highlighter-rouge">HiveContext</code>’s <code class="highlighter-rouge">refreshTable</code> method |
| to include those new files to the table. For a DataFrame representing a JSON dataset, users need to recreate |
| the DataFrame and the new DataFrame will include new files.</p> |
| </li> |
| </ul> |
| |
| <h2 id="upgrading-from-spark-sql-13-to-14">Upgrading from Spark SQL 1.3 to 1.4</h2> |
| |
| <h4 class="no_toc" id="dataframe-data-readerwriter-interface">DataFrame data reader/writer interface</h4> |
| |
| <p>Based on user feedback, we created a new, more fluid API for reading data in (<code class="highlighter-rouge">SQLContext.read</code>) |
| and writing data out (<code class="highlighter-rouge">DataFrame.write</code>), |
| and deprecated the old APIs (e.g., <code class="highlighter-rouge">SQLContext.parquetFile</code>, <code class="highlighter-rouge">SQLContext.jsonFile</code>).</p> |
| |
| <p>See the API docs for <code class="highlighter-rouge">SQLContext.read</code> ( |
| <a href="api/scala/index.html#org.apache.spark.sql.SQLContext@read:DataFrameReader">Scala</a>, |
| <a href="api/java/org/apache/spark/sql/SQLContext.html#read()">Java</a>, |
| <a href="api/python/pyspark.sql.html#pyspark.sql.SQLContext.read">Python</a> |
| ) and <code class="highlighter-rouge">DataFrame.write</code> ( |
| <a href="api/scala/index.html#org.apache.spark.sql.DataFrame@write:DataFrameWriter">Scala</a>, |
| <a href="api/java/org/apache/spark/sql/Dataset.html#write()">Java</a>, |
| <a href="api/python/pyspark.sql.html#pyspark.sql.DataFrame.write">Python</a> |
| ) more information.</p> |
| |
| <h4 class="no_toc" id="dataframegroupby-retains-grouping-columns">DataFrame.groupBy retains grouping columns</h4> |
| |
| <p>Based on user feedback, we changed the default behavior of <code class="highlighter-rouge">DataFrame.groupBy().agg()</code> to retain the |
| grouping columns in the resulting <code class="highlighter-rouge">DataFrame</code>. To keep the behavior in 1.3, set <code class="highlighter-rouge">spark.sql.retainGroupColumns</code> to <code class="highlighter-rouge">false</code>.</p> |
| |
| <div class="codetabs"> |
| <div data-lang="scala"> |
| |
| <figure class="highlight"><pre><code class="language-scala" data-lang="scala"><span class="c1">// In 1.3.x, in order for the grouping column "department" to show up,</span> |
| <span class="c1">// it must be included explicitly as part of the agg function call.</span> |
| <span class="nv">df</span><span class="o">.</span><span class="py">groupBy</span><span class="o">(</span><span class="s">"department"</span><span class="o">).</span><span class="py">agg</span><span class="o">(</span><span class="n">$</span><span class="s">"department"</span><span class="o">,</span> <span class="nf">max</span><span class="o">(</span><span class="s">"age"</span><span class="o">),</span> <span class="nf">sum</span><span class="o">(</span><span class="s">"expense"</span><span class="o">))</span> |
| |
| <span class="c1">// In 1.4+, grouping column "department" is included automatically.</span> |
| <span class="nv">df</span><span class="o">.</span><span class="py">groupBy</span><span class="o">(</span><span class="s">"department"</span><span class="o">).</span><span class="py">agg</span><span class="o">(</span><span class="nf">max</span><span class="o">(</span><span class="s">"age"</span><span class="o">),</span> <span class="nf">sum</span><span class="o">(</span><span class="s">"expense"</span><span class="o">))</span> |
| |
| <span class="c1">// Revert to 1.3 behavior (not retaining grouping column) by:</span> |
| <span class="nv">sqlContext</span><span class="o">.</span><span class="py">setConf</span><span class="o">(</span><span class="s">"spark.sql.retainGroupColumns"</span><span class="o">,</span> <span class="s">"false"</span><span class="o">)</span></code></pre></figure> |
| |
| </div> |
| |
| <div data-lang="java"> |
| |
| <figure class="highlight"><pre><code class="language-java" data-lang="java"><span class="c1">// In 1.3.x, in order for the grouping column "department" to show up,</span> |
| <span class="c1">// it must be included explicitly as part of the agg function call.</span> |
| <span class="n">df</span><span class="o">.</span><span class="na">groupBy</span><span class="o">(</span><span class="s">"department"</span><span class="o">).</span><span class="na">agg</span><span class="o">(</span><span class="n">col</span><span class="o">(</span><span class="s">"department"</span><span class="o">),</span> <span class="n">max</span><span class="o">(</span><span class="s">"age"</span><span class="o">),</span> <span class="n">sum</span><span class="o">(</span><span class="s">"expense"</span><span class="o">));</span> |
| |
| <span class="c1">// In 1.4+, grouping column "department" is included automatically.</span> |
| <span class="n">df</span><span class="o">.</span><span class="na">groupBy</span><span class="o">(</span><span class="s">"department"</span><span class="o">).</span><span class="na">agg</span><span class="o">(</span><span class="n">max</span><span class="o">(</span><span class="s">"age"</span><span class="o">),</span> <span class="n">sum</span><span class="o">(</span><span class="s">"expense"</span><span class="o">));</span> |
| |
| <span class="c1">// Revert to 1.3 behavior (not retaining grouping column) by:</span> |
| <span class="n">sqlContext</span><span class="o">.</span><span class="na">setConf</span><span class="o">(</span><span class="s">"spark.sql.retainGroupColumns"</span><span class="o">,</span> <span class="s">"false"</span><span class="o">);</span></code></pre></figure> |
| |
| </div> |
| |
| <div data-lang="python"> |
| |
| <figure class="highlight"><pre><code class="language-python" data-lang="python"><span class="kn">import</span> <span class="nn">pyspark.sql.functions</span> <span class="k">as</span> <span class="n">func</span> |
| |
| <span class="c1"># In 1.3.x, in order for the grouping column "department" to show up, |
| # it must be included explicitly as part of the agg function call. |
| </span><span class="n">df</span><span class="o">.</span><span class="n">groupBy</span><span class="p">(</span><span class="s">"department"</span><span class="p">)</span><span class="o">.</span><span class="n">agg</span><span class="p">(</span><span class="n">df</span><span class="p">[</span><span class="s">"department"</span><span class="p">],</span> <span class="n">func</span><span class="o">.</span><span class="nb">max</span><span class="p">(</span><span class="s">"age"</span><span class="p">),</span> <span class="n">func</span><span class="o">.</span><span class="nb">sum</span><span class="p">(</span><span class="s">"expense"</span><span class="p">))</span> |
| |
| <span class="c1"># In 1.4+, grouping column "department" is included automatically. |
| </span><span class="n">df</span><span class="o">.</span><span class="n">groupBy</span><span class="p">(</span><span class="s">"department"</span><span class="p">)</span><span class="o">.</span><span class="n">agg</span><span class="p">(</span><span class="n">func</span><span class="o">.</span><span class="nb">max</span><span class="p">(</span><span class="s">"age"</span><span class="p">),</span> <span class="n">func</span><span class="o">.</span><span class="nb">sum</span><span class="p">(</span><span class="s">"expense"</span><span class="p">))</span> |
| |
| <span class="c1"># Revert to 1.3.x behavior (not retaining grouping column) by: |
| </span><span class="n">sqlContext</span><span class="o">.</span><span class="n">setConf</span><span class="p">(</span><span class="s">"spark.sql.retainGroupColumns"</span><span class="p">,</span> <span class="s">"false"</span><span class="p">)</span></code></pre></figure> |
| |
| </div> |
| |
| </div> |
| |
| <h4 class="no_toc" id="behavior-change-on-dataframewithcolumn">Behavior change on DataFrame.withColumn</h4> |
| |
| <p>Prior to 1.4, DataFrame.withColumn() supports adding a column only. The column will always be added |
| as a new column with its specified name in the result DataFrame even if there may be any existing |
| columns of the same name. Since 1.4, DataFrame.withColumn() supports adding a column of a different |
| name from names of all existing columns or replacing existing columns of the same name.</p> |
| |
| <p>Note that this change is only for Scala API, not for PySpark and SparkR.</p> |
| |
| <h2 id="upgrading-from-spark-sql-10-12-to-13">Upgrading from Spark SQL 1.0-1.2 to 1.3</h2> |
| |
| <p>In Spark 1.3 we removed the “Alpha” label from Spark SQL and as part of this did a cleanup of the |
| available APIs. From Spark 1.3 onwards, Spark SQL will provide binary compatibility with other |
| releases in the 1.X series. This compatibility guarantee excludes APIs that are explicitly marked |
| as unstable (i.e., DeveloperAPI or Experimental).</p> |
| |
| <h4 class="no_toc" id="rename-of-schemardd-to-dataframe">Rename of SchemaRDD to DataFrame</h4> |
| |
| <p>The largest change that users will notice when upgrading to Spark SQL 1.3 is that <code class="highlighter-rouge">SchemaRDD</code> has |
| been renamed to <code class="highlighter-rouge">DataFrame</code>. This is primarily because DataFrames no longer inherit from RDD |
| directly, but instead provide most of the functionality that RDDs provide though their own |
| implementation. DataFrames can still be converted to RDDs by calling the <code class="highlighter-rouge">.rdd</code> method.</p> |
| |
| <p>In Scala, there is a type alias from <code class="highlighter-rouge">SchemaRDD</code> to <code class="highlighter-rouge">DataFrame</code> to provide source compatibility for |
| some use cases. It is still recommended that users update their code to use <code class="highlighter-rouge">DataFrame</code> instead. |
| Java and Python users will need to update their code.</p> |
| |
| <h4 class="no_toc" id="unification-of-the-java-and-scala-apis">Unification of the Java and Scala APIs</h4> |
| |
| <p>Prior to Spark 1.3 there were separate Java compatible classes (<code class="highlighter-rouge">JavaSQLContext</code> and <code class="highlighter-rouge">JavaSchemaRDD</code>) |
| that mirrored the Scala API. In Spark 1.3 the Java API and Scala API have been unified. Users |
| of either language should use <code class="highlighter-rouge">SQLContext</code> and <code class="highlighter-rouge">DataFrame</code>. In general these classes try to |
| use types that are usable from both languages (i.e. <code class="highlighter-rouge">Array</code> instead of language-specific collections). |
| In some cases where no common type exists (e.g., for passing in closures or Maps) function overloading |
| is used instead.</p> |
| |
| <p>Additionally, the Java specific types API has been removed. Users of both Scala and Java should |
| use the classes present in <code class="highlighter-rouge">org.apache.spark.sql.types</code> to describe schema programmatically.</p> |
| |
| <h4 class="no_toc" id="isolation-of-implicit-conversions-and-removal-of-dsl-package-scala-only">Isolation of Implicit Conversions and Removal of dsl Package (Scala-only)</h4> |
| |
| <p>Many of the code examples prior to Spark 1.3 started with <code class="highlighter-rouge">import sqlContext._</code>, which brought |
| all of the functions from sqlContext into scope. In Spark 1.3 we have isolated the implicit |
| conversions for converting <code class="highlighter-rouge">RDD</code>s into <code class="highlighter-rouge">DataFrame</code>s into an object inside of the <code class="highlighter-rouge">SQLContext</code>. |
| Users should now write <code class="highlighter-rouge">import sqlContext.implicits._</code>.</p> |
| |
| <p>Additionally, the implicit conversions now only augment RDDs that are composed of <code class="highlighter-rouge">Product</code>s (i.e., |
| case classes or tuples) with a method <code class="highlighter-rouge">toDF</code>, instead of applying automatically.</p> |
| |
| <p>When using function inside of the DSL (now replaced with the <code class="highlighter-rouge">DataFrame</code> API) users used to import |
| <code class="highlighter-rouge">org.apache.spark.sql.catalyst.dsl</code>. Instead the public dataframe functions API should be used: |
| <code class="highlighter-rouge">import org.apache.spark.sql.functions._</code>.</p> |
| |
| <h4 class="no_toc" id="removal-of-the-type-aliases-in-orgapachesparksql-for-datatype-scala-only">Removal of the type aliases in org.apache.spark.sql for DataType (Scala-only)</h4> |
| |
| <p>Spark 1.3 removes the type aliases that were present in the base sql package for <code class="highlighter-rouge">DataType</code>. Users |
| should instead import the classes in <code class="highlighter-rouge">org.apache.spark.sql.types</code></p> |
| |
| <h4 class="no_toc" id="udf-registration-moved-to-sqlcontextudf-java--scala">UDF Registration Moved to <code class="highlighter-rouge">sqlContext.udf</code> (Java & Scala)</h4> |
| |
| <p>Functions that are used to register UDFs, either for use in the DataFrame DSL or SQL, have been |
| moved into the udf object in <code class="highlighter-rouge">SQLContext</code>.</p> |
| |
| <div class="codetabs"> |
| <div data-lang="scala"> |
| |
| <figure class="highlight"><pre><code class="language-scala" data-lang="scala"><span class="nv">sqlContext</span><span class="o">.</span><span class="py">udf</span><span class="o">.</span><span class="py">register</span><span class="o">(</span><span class="s">"strLen"</span><span class="o">,</span> <span class="o">(</span><span class="n">s</span><span class="k">:</span> <span class="kt">String</span><span class="o">)</span> <span class="k">=></span> <span class="nv">s</span><span class="o">.</span><span class="py">length</span><span class="o">())</span></code></pre></figure> |
| |
| </div> |
| |
| <div data-lang="java"> |
| |
| <figure class="highlight"><pre><code class="language-java" data-lang="java"><span class="n">sqlContext</span><span class="o">.</span><span class="na">udf</span><span class="o">().</span><span class="na">register</span><span class="o">(</span><span class="s">"strLen"</span><span class="o">,</span> <span class="o">(</span><span class="nc">String</span> <span class="n">s</span><span class="o">)</span> <span class="o">-></span> <span class="n">s</span><span class="o">.</span><span class="na">length</span><span class="o">(),</span> <span class="nc">DataTypes</span><span class="o">.</span><span class="na">IntegerType</span><span class="o">);</span></code></pre></figure> |
| |
| </div> |
| |
| </div> |
| |
| <p>Python UDF registration is unchanged.</p> |
| |
| <h2 id="compatibility-with-apache-hive">Compatibility with Apache Hive</h2> |
| |
| <p>Spark SQL is designed to be compatible with the Hive Metastore, SerDes and UDFs. |
| Currently, Hive SerDes and UDFs are based on Hive 1.2.1, |
| and Spark SQL can be connected to different versions of Hive Metastore |
| (from 0.12.0 to 2.3.6 and 3.0.0 to 3.1.2. Also see <a href="sql-data-sources-hive-tables.html#interacting-with-different-versions-of-hive-metastore">Interacting with Different Versions of Hive Metastore</a>).</p> |
| |
| <h4 class="no_toc" id="deploying-in-existing-hive-warehouses">Deploying in Existing Hive Warehouses</h4> |
| |
| <p>The Spark SQL Thrift JDBC server is designed to be “out of the box” compatible with existing Hive |
| installations. You do not need to modify your existing Hive Metastore or change the data placement |
| or partitioning of your tables.</p> |
| |
| <h3 class="no_toc" id="supported-hive-features">Supported Hive Features</h3> |
| |
| <p>Spark SQL supports the vast majority of Hive features, such as:</p> |
| |
| <ul> |
| <li>Hive query statements, including: |
| <ul> |
| <li><code class="highlighter-rouge">SELECT</code></li> |
| <li><code class="highlighter-rouge">GROUP BY</code></li> |
| <li><code class="highlighter-rouge">ORDER BY</code></li> |
| <li><code class="highlighter-rouge">CLUSTER BY</code></li> |
| <li><code class="highlighter-rouge">SORT BY</code></li> |
| </ul> |
| </li> |
| <li>All Hive operators, including: |
| <ul> |
| <li>Relational operators (<code class="highlighter-rouge">=</code>, <code class="highlighter-rouge"><=></code>, <code class="highlighter-rouge">==</code>, <code class="highlighter-rouge"><></code>, <code class="highlighter-rouge"><</code>, <code class="highlighter-rouge">></code>, <code class="highlighter-rouge">>=</code>, <code class="highlighter-rouge"><=</code>, etc)</li> |
| <li>Arithmetic operators (<code class="highlighter-rouge">+</code>, <code class="highlighter-rouge">-</code>, <code class="highlighter-rouge">*</code>, <code class="highlighter-rouge">/</code>, <code class="highlighter-rouge">%</code>, etc)</li> |
| <li>Logical operators (<code class="highlighter-rouge">AND</code>, <code class="highlighter-rouge">&&</code>, <code class="highlighter-rouge">OR</code>, <code class="highlighter-rouge">||</code>, etc)</li> |
| <li>Complex type constructors</li> |
| <li>Mathematical functions (<code class="highlighter-rouge">sign</code>, <code class="highlighter-rouge">ln</code>, <code class="highlighter-rouge">cos</code>, etc)</li> |
| <li>String functions (<code class="highlighter-rouge">instr</code>, <code class="highlighter-rouge">length</code>, <code class="highlighter-rouge">printf</code>, etc)</li> |
| </ul> |
| </li> |
| <li>User defined functions (UDF)</li> |
| <li>User defined aggregation functions (UDAF)</li> |
| <li>User defined serialization formats (SerDes)</li> |
| <li>Window functions</li> |
| <li>Joins |
| <ul> |
| <li><code class="highlighter-rouge">JOIN</code></li> |
| <li><code class="highlighter-rouge">{LEFT|RIGHT|FULL} OUTER JOIN</code></li> |
| <li><code class="highlighter-rouge">LEFT SEMI JOIN</code></li> |
| <li><code class="highlighter-rouge">CROSS JOIN</code></li> |
| </ul> |
| </li> |
| <li>Unions</li> |
| <li>Sub-queries |
| <ul> |
| <li><code class="highlighter-rouge">SELECT col FROM ( SELECT a + b AS col from t1) t2</code></li> |
| </ul> |
| </li> |
| <li>Sampling</li> |
| <li>Explain</li> |
| <li>Partitioned tables including dynamic partition insertion</li> |
| <li>View |
| <ul> |
| <li> |
| <p>If column aliases are not specified in view definition queries, both Spark and Hive will |
| generate alias names, but in different ways. In order for Spark to be able to read views created |
| by Hive, users should explicitly specify column aliases in view definition queries. As an |
| example, Spark cannot read <code class="highlighter-rouge">v1</code> created as below by Hive.</p> |
| |
| <div class="highlighter-rouge"><div class="highlight"><pre class="highlight"><code>CREATE VIEW v1 AS SELECT * FROM (SELECT c + 1 FROM (SELECT 1 c) t1) t2; |
| </code></pre></div> </div> |
| |
| <p>Instead, you should create <code class="highlighter-rouge">v1</code> as below with column aliases explicitly specified.</p> |
| |
| <div class="highlighter-rouge"><div class="highlight"><pre class="highlight"><code>CREATE VIEW v1 AS SELECT * FROM (SELECT c + 1 AS inc_c FROM (SELECT 1 c) t1) t2; |
| </code></pre></div> </div> |
| </li> |
| </ul> |
| </li> |
| <li>All Hive DDL Functions, including: |
| <ul> |
| <li><code class="highlighter-rouge">CREATE TABLE</code></li> |
| <li><code class="highlighter-rouge">CREATE TABLE AS SELECT</code></li> |
| <li><code class="highlighter-rouge">CREATE TABLE LIKE</code></li> |
| <li><code class="highlighter-rouge">ALTER TABLE</code></li> |
| </ul> |
| </li> |
| <li>Most Hive Data types, including: |
| <ul> |
| <li><code class="highlighter-rouge">TINYINT</code></li> |
| <li><code class="highlighter-rouge">SMALLINT</code></li> |
| <li><code class="highlighter-rouge">INT</code></li> |
| <li><code class="highlighter-rouge">BIGINT</code></li> |
| <li><code class="highlighter-rouge">BOOLEAN</code></li> |
| <li><code class="highlighter-rouge">FLOAT</code></li> |
| <li><code class="highlighter-rouge">DOUBLE</code></li> |
| <li><code class="highlighter-rouge">STRING</code></li> |
| <li><code class="highlighter-rouge">BINARY</code></li> |
| <li><code class="highlighter-rouge">TIMESTAMP</code></li> |
| <li><code class="highlighter-rouge">DATE</code></li> |
| <li><code class="highlighter-rouge">ARRAY<></code></li> |
| <li><code class="highlighter-rouge">MAP<></code></li> |
| <li><code class="highlighter-rouge">STRUCT<></code></li> |
| </ul> |
| </li> |
| </ul> |
| |
| <h3 class="no_toc" id="unsupported-hive-functionality">Unsupported Hive Functionality</h3> |
| |
| <p>Below is a list of Hive features that we don’t support yet. Most of these features are rarely used |
| in Hive deployments.</p> |
| |
| <p><strong>Major Hive Features</strong></p> |
| |
| <ul> |
| <li>Tables with buckets: bucket is the hash partitioning within a Hive table partition. Spark SQL |
| doesn’t support buckets yet.</li> |
| </ul> |
| |
| <p><strong>Esoteric Hive Features</strong></p> |
| |
| <ul> |
| <li><code class="highlighter-rouge">UNION</code> type</li> |
| <li>Unique join</li> |
| <li>Column statistics collecting: Spark SQL does not piggyback scans to collect column statistics at |
| the moment and only supports populating the sizeInBytes field of the hive metastore.</li> |
| </ul> |
| |
| <p><strong>Hive Input/Output Formats</strong></p> |
| |
| <ul> |
| <li>File format for CLI: For results showing back to the CLI, Spark SQL only supports TextOutputFormat.</li> |
| <li>Hadoop archive</li> |
| </ul> |
| |
| <p><strong>Hive Optimizations</strong></p> |
| |
| <p>A handful of Hive optimizations are not yet included in Spark. Some of these (such as indexes) are |
| less important due to Spark SQL’s in-memory computational model. Others are slotted for future |
| releases of Spark SQL.</p> |
| |
| <ul> |
| <li>Block-level bitmap indexes and virtual columns (used to build indexes)</li> |
| <li>Automatically determine the number of reducers for joins and groupbys: Currently, in Spark SQL, you |
| need to control the degree of parallelism post-shuffle using “<code class="highlighter-rouge">SET spark.sql.shuffle.partitions=[num_tasks];</code>”.</li> |
| <li>Meta-data only query: For queries that can be answered by using only metadata, Spark SQL still |
| launches tasks to compute the result.</li> |
| <li>Skew data flag: Spark SQL does not follow the skew data flags in Hive.</li> |
| <li><code class="highlighter-rouge">STREAMTABLE</code> hint in join: Spark SQL does not follow the <code class="highlighter-rouge">STREAMTABLE</code> hint.</li> |
| <li>Merge multiple small files for query results: if the result output contains multiple small files, |
| Hive can optionally merge the small files into fewer large files to avoid overflowing the HDFS |
| metadata. Spark SQL does not support that.</li> |
| </ul> |
| |
| <p><strong>Hive UDF/UDTF/UDAF</strong></p> |
| |
| <p>Not all the APIs of the Hive UDF/UDTF/UDAF are supported by Spark SQL. Below are the unsupported APIs:</p> |
| |
| <ul> |
| <li><code class="highlighter-rouge">getRequiredJars</code> and <code class="highlighter-rouge">getRequiredFiles</code> (<code class="highlighter-rouge">UDF</code> and <code class="highlighter-rouge">GenericUDF</code>) are functions to automatically |
| include additional resources required by this UDF.</li> |
| <li><code class="highlighter-rouge">initialize(StructObjectInspector)</code> in <code class="highlighter-rouge">GenericUDTF</code> is not supported yet. Spark SQL currently uses |
| a deprecated interface <code class="highlighter-rouge">initialize(ObjectInspector[])</code> only.</li> |
| <li><code class="highlighter-rouge">configure</code> (<code class="highlighter-rouge">GenericUDF</code>, <code class="highlighter-rouge">GenericUDTF</code>, and <code class="highlighter-rouge">GenericUDAFEvaluator</code>) is a function to initialize |
| functions with <code class="highlighter-rouge">MapredContext</code>, which is inapplicable to Spark.</li> |
| <li><code class="highlighter-rouge">close</code> (<code class="highlighter-rouge">GenericUDF</code> and <code class="highlighter-rouge">GenericUDAFEvaluator</code>) is a function to release associated resources. |
| Spark SQL does not call this function when tasks finish.</li> |
| <li><code class="highlighter-rouge">reset</code> (<code class="highlighter-rouge">GenericUDAFEvaluator</code>) is a function to re-initialize aggregation for reusing the same aggregation. |
| Spark SQL currently does not support the reuse of aggregation.</li> |
| <li><code class="highlighter-rouge">getWindowingEvaluator</code> (<code class="highlighter-rouge">GenericUDAFEvaluator</code>) is a function to optimize aggregation by evaluating |
| an aggregate over a fixed window.</li> |
| </ul> |
| |
| <h3 class="no_toc" id="incompatible-hive-udf">Incompatible Hive UDF</h3> |
| |
| <p>Below are the scenarios in which Hive and Spark generate different results:</p> |
| |
| <ul> |
| <li><code class="highlighter-rouge">SQRT(n)</code> If n < 0, Hive returns null, Spark SQL returns NaN.</li> |
| <li><code class="highlighter-rouge">ACOS(n)</code> If n < -1 or n > 1, Hive returns null, Spark SQL returns NaN.</li> |
| <li><code class="highlighter-rouge">ASIN(n)</code> If n < -1 or n > 1, Hive returns null, Spark SQL returns NaN.</li> |
| </ul> |
| |
| |
| </div> |
| |
| <!-- /container --> |
| </div> |
| |
| <script src="js/vendor/jquery-3.4.1.min.js"></script> |
| <script src="js/vendor/bootstrap.min.js"></script> |
| <script src="js/vendor/anchor.min.js"></script> |
| <script src="js/main.js"></script> |
| |
| <!-- MathJax Section --> |
| <script type="text/x-mathjax-config"> |
| MathJax.Hub.Config({ |
| TeX: { equationNumbers: { autoNumber: "AMS" } } |
| }); |
| </script> |
| <script> |
| // Note that we load MathJax this way to work with local file (file://), HTTP and HTTPS. |
| // We could use "//cdn.mathjax...", but that won't support "file://". |
| (function(d, script) { |
| script = d.createElement('script'); |
| script.type = 'text/javascript'; |
| script.async = true; |
| script.onload = function(){ |
| MathJax.Hub.Config({ |
| tex2jax: { |
| inlineMath: [ ["$", "$"], ["\\\\(","\\\\)"] ], |
| displayMath: [ ["$$","$$"], ["\\[", "\\]"] ], |
| processEscapes: true, |
| skipTags: ['script', 'noscript', 'style', 'textarea', 'pre'] |
| } |
| }); |
| }; |
| script.src = ('https:' == document.location.protocol ? 'https://' : 'http://') + |
| 'cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.1/MathJax.js' + |
| '?config=TeX-AMS-MML_HTMLorMML'; |
| d.getElementsByTagName('head')[0].appendChild(script); |
| }(document)); |
| </script> |
| </body> |
| </html> |
