This PR manage the memory used by window functions (buffered rows), also enable external spilling.
After this PR, we can run window functions on a partition with hundreds of millions of rows with only 1G.
Author: Davies Liu <davies@databricks.com>
Closes#10605 from davies/unsafe_window.
modify 'spark.memory.offHeap.enabled' default value to false
Author: zzcclp <xm_zzc@sina.com>
Closes#10633 from zzcclp/fix_spark.memory.offHeap.enabled_default_value.
MapPartitionsRDD was keeping a reference to `prev` after a call to
`clearDependencies` which could lead to memory leak.
Author: Guillaume Poulin <poulin.guillaume@gmail.com>
Closes#10623 from gpoulin/map_partition_deps.
This PR removes `spark.cleaner.ttl` and the associated TTL-based metadata cleaning code.
Now that we have the `ContextCleaner` and a timer to trigger periodic GCs, I don't think that `spark.cleaner.ttl` is necessary anymore. The TTL-based cleaning isn't enabled by default, isn't included in our end-to-end tests, and has been a source of user confusion when it is misconfigured. If the TTL is set too low, data which is still being used may be evicted / deleted, leading to hard to diagnose bugs.
For all of these reasons, I think that we should remove this functionality in Spark 2.0. Additional benefits of doing this include marginally reduced memory usage, since we no longer need to store timetsamps in hashmaps, and a handful fewer threads.
Author: Josh Rosen <joshrosen@databricks.com>
Closes#10534 from JoshRosen/remove-ttl-based-cleaning.
[SPARK-12640][SQL] Add simple benchmarking utility class and add Parquet scan benchmarks.
We've run benchmarks ad hoc to measure the scanner performance. We will continue to invest in this
and it makes sense to get these benchmarks into code. This adds a simple benchmarking utility to do
this.
Author: Nong Li <nong@databricks.com>
Author: Nong <nongli@gmail.com>
Closes#10589 from nongli/spark-12640.
Change Java countByKey, countApproxDistinctByKey return types to use Java Long, not Scala; update similar methods for consistency on java.long.Long.valueOf with no API change
Author: Sean Owen <sowen@cloudera.com>
Closes#10554 from srowen/SPARK-12604.
This PR adds bucket write support to Spark SQL. User can specify bucketing columns, numBuckets and sorting columns with or without partition columns. For example:
```
df.write.partitionBy("year").bucketBy(8, "country").sortBy("amount").saveAsTable("sales")
```
When bucketing is used, we will calculate bucket id for each record, and group the records by bucket id. For each group, we will create a file with bucket id in its name, and write data into it. For each bucket file, if sorting columns are specified, the data will be sorted before write.
Note that there may be multiply files for one bucket, as the data is distributed.
Currently we store the bucket metadata at hive metastore in a non-hive-compatible way. We use different bucketing hash function compared to hive, so we can't be compatible anyway.
Limitations:
* Can't write bucketed data without hive metastore.
* Can't insert bucketed data into existing hive tables.
Author: Wenchen Fan <wenchen@databricks.com>
Closes#10498 from cloud-fan/bucket-write.
To avoid to have a huge Java source (over 64K loc), that can't be compiled.
cc hvanhovell
Author: Davies Liu <davies@databricks.com>
Closes#10624 from davies/split_ident.
Move Py4jCallbackConnectionCleaner to Streaming because the callback server starts only in StreamingContext.
Author: Shixiong Zhu <shixiong@databricks.com>
Closes#10621 from zsxwing/SPARK-12617-2.
For the BinaryClassificationEvaluator, the scaladoc doesn't mention that "areaUnderPR" is supported, only that the default is "areadUnderROC".
Also, in the documentation, it is said that:
"The default metric used to choose the best ParamMap can be overriden by the setMetric method in each of these evaluators."
However, the method is called setMetricName.
This PR aims to fix both issues.
Author: BenFradet <benjamin.fradet@gmail.com>
Closes#10328 from BenFradet/SPARK-12368.
If initial model passed to GMM is not empty it causes `net.razorvine.pickle.PickleException`. It can be fixed by converting `initialModel.weights` to `list`.
Author: zero323 <matthew.szymkiewicz@gmail.com>
Closes#9986 from zero323/SPARK-12006.
This PR moves a major part of the new SQL parser to Catalyst. This is a prelude to start using this parser for all of our SQL parsing. The following key changes have been made:
The ANTLR Parser & Supporting classes have been moved to the Catalyst project. They are now part of the ```org.apache.spark.sql.catalyst.parser``` package. These classes contained quite a bit of code that was originally from the Hive project, I have added aknowledgements whenever this applied. All Hive dependencies have been factored out. I have also taken this chance to clean-up the ```ASTNode``` class, and to improve the error handling.
The HiveQl object that provides the functionality to convert an AST into a LogicalPlan has been refactored into three different classes, one for every SQL sub-project:
- ```CatalystQl```: This implements Query and Expression parsing functionality.
- ```SparkQl```: This is a subclass of CatalystQL and provides SQL/Core only functionality such as Explain and Describe.
- ```HiveQl```: This is a subclass of ```SparkQl``` and this adds Hive-only functionality to the parser such as Analyze, Drop, Views, CTAS & Transforms. This class still depends on Hive.
cc rxin
Author: Herman van Hovell <hvanhovell@questtec.nl>
Closes#10583 from hvanhovell/SPARK-12575.
PySpark ```DecisionTreeClassifier``` & ```DecisionTreeRegressor``` should support ```setSeed``` like what we do at Scala side.
Author: Yanbo Liang <ybliang8@gmail.com>
Closes#9807 from yanboliang/spark-11815.
Add ```computeCost``` to ```KMeansModel``` as evaluator for PySpark spark.ml.
Author: Yanbo Liang <ybliang8@gmail.com>
Closes#9931 from yanboliang/SPARK-11945.
From JIRA:
Currently, PySpark wrappers for spark.ml Scala classes are brittle when accepting Param types. E.g., Normalizer's "p" param cannot be set to "2" (an integer); it must be set to "2.0" (a float). Fixing this is not trivial since there does not appear to be a natural place to insert the conversion before Python wrappers call Java's Params setter method.
A possible fix will be to include a method "_checkType" to PySpark's Param class which checks the type, prints an error if needed, and converts types when relevant (e.g., int to float, or scipy matrix to array). The Java wrapper method which copies params to Scala can call this method when available.
This fix instead checks the types at set time since I think failing sooner is better, but I can switch it around to check at copy time if that would be better. So far this only converts int to float and other conversions (like scipymatrix to array) are left for the future.
Author: Holden Karau <holden@us.ibm.com>
Closes#9581 from holdenk/SPARK-7675-PySpark-sparkml-Params-type-conversion.
For queries like :
select <> from table group by a distribute by a
we can eliminate distribute by ; since group by will anyways do a hash partitioning
Also applicable when user uses Dataframe API
Author: Yash Datta <Yash.Datta@guavus.com>
Closes#9858 from saucam/eliminatedistribute.
Whole code of Vector.scala, VectorSuite.scala and GraphKryoRegistrator.scala are no longer used so it's time to remove them in Spark 2.0.
Author: Kousuke Saruta <sarutak@oss.nttdata.co.jp>
Closes#10613 from sarutak/SPARK-12665.
JIRA: https://issues.apache.org/jira/browse/SPARK-12578
Slightly update to Hive parser. We should keep the distinct keyword when used in an aggregate function with OVER clause. So the CheckAnalysis will detect it and throw exception later.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#10557 from viirya/keep-distinct-hivesql.
Add ```read.text``` and ```write.text``` for SparkR.
cc sun-rui felixcheung shivaram
Author: Yanbo Liang <ybliang8@gmail.com>
Closes#10348 from yanboliang/spark-12393.
Cartesian product use UnsafeExternalSorter without comparator to do spilling, it will NPE if spilling happens.
This bug also hitted by #10605
cc JoshRosen
Author: Davies Liu <davies@databricks.com>
Closes#10606 from davies/fix_spilling.
This fix masks JDBC credentials in the explain output. URL patterns to specify credential seems to be vary between different databases. Added a new method to dialect to mask the credentials according to the database specific URL pattern.
While adding tests I noticed explain output includes array variable for partitions ([Lorg.apache.spark.Partition;3ff74546,). Modified the code to include the first, and last partition information.
Author: sureshthalamati <suresh.thalamati@gmail.com>
Closes#10452 from sureshthalamati/mask_jdbc_credentials_spark-12504.
Add `columnSimilarities` to IndexedRowMatrix for PySpark spark.mllib.linalg.
Author: Kai Jiang <jiangkai@gmail.com>
Closes#10158 from vectorijk/spark-12041.
Successfully ran kinesis demo on a live, aws hosted kinesis stream against master and 1.6 branches. For reasons I don't entirely understand it required a manual merge to 1.5 which I did as shown here: 075c22e89b
The demo ran successfully on the 1.5 branch as well.
According to `mvn dependency:tree` it is still pulling a fairly old version of the aws-java-sdk (1.9.37), but this appears to have fixed the kinesis regression in 1.5.2.
Author: BrianLondon <brian@seatgeek.com>
Closes#10492 from BrianLondon/remove-only.
Update user guide doc for ```DecisionTreeRegressor``` providing variance of prediction.
cc jkbradley
Author: Yanbo Liang <ybliang8@gmail.com>
Closes#10594 from yanboliang/spark-12570.
There is an issue that Py4J's PythonProxyHandler.finalize blocks forever. (https://github.com/bartdag/py4j/pull/184)
Py4j will create a PythonProxyHandler in Java for "transformer_serializer" when calling "registerSerializer". If we call "registerSerializer" twice, the second PythonProxyHandler will override the first one, then the first one will be GCed and trigger "PythonProxyHandler.finalize". To avoid that, we should not call"registerSerializer" more than once, so that "PythonProxyHandler" in Java side won't be GCed.
Author: Shixiong Zhu <shixiong@databricks.com>
Closes#10514 from zsxwing/SPARK-12511.
As noted in the code, this change is to make this component easier to test in isolation.
Author: Nong <nongli@gmail.com>
Closes#10581 from nongli/spark-12636.
This patch added Py4jCallbackConnectionCleaner to clean the leak sockets of Py4J every 30 seconds. This is a workaround before Py4J fixes the leak issue https://github.com/bartdag/py4j/issues/187
Author: Shixiong Zhu <shixiong@databricks.com>
Closes#10579 from zsxwing/SPARK-12617.
JIRA: https://issues.apache.org/jira/browse/SPARK-12439
In toCatalystArray, we should look at the data type returned by dataTypeFor instead of silentSchemaFor, to determine if the element is native type. An obvious problem is when the element is Option[Int] class, catalsilentSchemaFor will return Int, then we will wrongly recognize the element is native type.
There is another problem when using Option as array element. When we encode data like Seq(Some(1), Some(2), None) with encoder, we will use MapObjects to construct an array for it later. But in MapObjects, we don't check if the return value of lambdaFunction is null or not. That causes a bug that the decoded data for Seq(Some(1), Some(2), None) would be Seq(1, 2, -1), instead of Seq(1, 2, null).
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#10391 from viirya/fix-catalystarray.
I looked at each case individually and it looks like they can all be removed. The only one that I had to think twice was toArray (I even thought about un-deprecating it, until I realized it was a problem in Java to have toArray returning java.util.List).
Author: Reynold Xin <rxin@databricks.com>
Closes#10569 from rxin/SPARK-12615.
address comments in #10435
This makes the API easier to use if user programmatically generate the call to hash, and they will get analysis exception if the arguments of hash is empty.
Author: Wenchen Fan <wenchen@databricks.com>
Closes#10588 from cloud-fan/hash.
JIRA: https://issues.apache.org/jira/browse/SPARK-12643
Without setting lib directory for antlr, the updates of imported grammar files can not be detected. So SparkSqlParser.g will not be rebuilt automatically.
Since it is a minor update, no JIRA ticket is opened. Let me know if it is needed. Thanks.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#10571 from viirya/antlr-build.
Modified the definition of R^2 for regression through origin. Added modified test for regression metrics.
Author: Imran Younus <iyounus@us.ibm.com>
Author: Imran Younus <imranyounus@gmail.com>
Closes#10384 from iyounus/SPARK_12331_R2_for_regression_through_origin.
Currently we don't support Hadoop 0.23 but there is a few code related to it so let's clean it up.
Author: Kousuke Saruta <sarutak@oss.nttdata.co.jp>
Closes#10590 from sarutak/SPARK-12641.
rxin davies shivaram
Took save mode from my PR #10480, and move everything to writer methods. This is related to PR #10559
- [x] it seems jsonRDD() is broken, need to investigate - this is not a public API though; will look into some more tonight. (fixed)
Author: felixcheung <felixcheung_m@hotmail.com>
Closes#10584 from felixcheung/rremovedeprecated.
