## What changes were proposed in this pull request?
In `PushDownOperatorsToDataSource`, we use `transformUp` to match `PhysicalOperation` and apply pushdown. This is problematic if we have multiple `Filter` and `Project` above the data source v2 relation.
e.g. for a query
```
Project
Filter
DataSourceV2Relation
```
The pattern match will be triggered twice and we will do operator pushdown twice. This is unnecessary, we can use `mapChildren` to only apply pushdown once.
## How was this patch tested?
existing test
Author: Wenchen Fan <wenchen@databricks.com>
Closes#21230 from cloud-fan/step2.
Instead of always throwing a generic exception when the AM fails,
print a generic error and throw the exception with the YARN
diagnostics containing the reason for the failure.
There was an issue with YARN sometimes providing a generic diagnostic
message, even though the AM provides a failure reason when
unregistering. That was happening because the AM was registering
too late, and if errors happened before the registration, YARN would
just create a generic "ExitCodeException" which wasn't very helpful.
Since most errors in this path are a result of not being able to
connect to the driver, this change modifies the AM registration
a bit so that the AM is registered before the connection to the
driver is established. That way, errors are properly propagated
through YARN back to the driver.
As part of that, I also removed the code that retried connections
to the driver from the client AM. At that point, the driver should
already be up and waiting for connections, so it's unlikely that
retrying would help - and in case it does, that means a flaky
network, which would mean problems would probably show up again.
The effect of that is that connection-related errors are reported
back to the driver much faster now (through the YARN report).
One thing to note is that there seems to be a race on the YARN
side that causes a report to be sent to the client without the
corresponding diagnostics string from the AM; the diagnostics are
available later from the RM web page. For that reason, the generic
error messages are kept in the Spark scheduler code, to help
guide users to a way of debugging their failure.
Also of note is that if YARN's max attempts configuration is lower
than Spark's, Spark will not unregister the AM with a proper
diagnostics message. Unfortunately there seems to be no way to
unregister the AM and still allow further re-attempts to happen.
Testing:
- existing unit tests
- some of our integration tests
- hardcoded an invalid driver address in the code and verified
the error in the shell. e.g.
```
scala> 18/05/04 15:09:34 ERROR cluster.YarnClientSchedulerBackend: YARN application has exited unexpectedly with state FAILED! Check the YARN application logs for more details.
18/05/04 15:09:34 ERROR cluster.YarnClientSchedulerBackend: Diagnostics message: Uncaught exception: org.apache.spark.SparkException: Exception thrown in awaitResult:
<AM stack trace>
Caused by: java.io.IOException: Failed to connect to localhost/127.0.0.1:1234
<More stack trace>
```
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#21243 from vanzin/SPARK-24182.
## What changes were proposed in this pull request?
The PR adds array_sort function to SparkR.
## How was this patch tested?
Tests added into R/pkg/tests/fulltests/test_sparkSQL.R
## Example
```
> df <- createDataFrame(list(list(list(2L, 1L, 3L, NA)), list(list(NA, 6L, 5L, NA, 4L))))
> head(collect(select(df, array_sort(df[[1]]))))
```
Result:
```
array_sort(_1)
1 1, 2, 3, NA
2 4, 5, 6, NA, NA
```
Author: Marek Novotny <mn.mikke@gmail.com>
Closes#21294 from mn-mikke/SPARK-24197.
## What changes were proposed in this pull request?
This is a followup of https://github.com/apache/spark/pull/20136 . #20136 didn't really work because in the test, we are using local backend, which shares the driver side `SparkEnv`, so `SparkEnv.get.executorId == SparkContext.DRIVER_IDENTIFIER` doesn't work.
This PR changes the check to `TaskContext.get != null`, and move the check to `SQLConf.get`, and fix all the places that violate this check:
* `InMemoryTableScanExec#createAndDecompressColumn` is executed inside `rdd.map`, we can't access `conf.offHeapColumnVectorEnabled` there. https://github.com/apache/spark/pull/21223 merged
* `DataType#sameType` may be executed in executor side, for things like json schema inference, so we can't call `conf.caseSensitiveAnalysis` there. This contributes to most of the code changes, as we need to add `caseSensitive` parameter to a lot of methods.
* `ParquetFilters` is used in the file scan function, which is executed in executor side, so we can't can't call `conf.parquetFilterPushDownDate` there. https://github.com/apache/spark/pull/21224 merged
* `WindowExec#createBoundOrdering` is called on executor side, so we can't use `conf.sessionLocalTimezone` there. https://github.com/apache/spark/pull/21225 merged
* `JsonToStructs` can be serialized to executors and evaluate, we should not call `SQLConf.get.getConf(SQLConf.FROM_JSON_FORCE_NULLABLE_SCHEMA)` in the body. https://github.com/apache/spark/pull/21226 merged
## How was this patch tested?
existing test
Author: Wenchen Fan <wenchen@databricks.com>
Closes#21190 from cloud-fan/minor.
## What changes were proposed in this pull request?
When multiple clients attempt to resolve artifacts via the `--packages` parameter, they could run into race condition when they each attempt to modify the dummy `org.apache.spark-spark-submit-parent-default.xml` file created in the default ivy cache dir.
This PR changes the behavior to encode UUID in the dummy module descriptor so each client will operate on a different resolution file in the ivy cache dir. In addition, this patch changes the behavior of when and which resolution files are cleaned to prevent accumulation of resolution files in the default ivy cache dir.
Since this PR is a successor of #18801, close#18801. Many codes were ported from #18801. **Many efforts were put here. I think this PR should credit to Victsm .**
## How was this patch tested?
added UT into `SparkSubmitUtilsSuite`
Author: Kazuaki Ishizaki <ishizaki@jp.ibm.com>
Closes#21251 from kiszk/SPARK-10878.
## What changes were proposed in this pull request?
Sometimes "SparkListenerSuite.local metrics" test fails because the average of executorDeserializeTime is too short. As squito suggested to avoid these situations in one of the task a reference introduced to an object implementing a custom Externalizable.readExternal which sleeps 1ms before returning.
## How was this patch tested?
With unit tests (and checking the effect of this change to the average with a much larger sleep time).
Author: “attilapiros” <piros.attila.zsolt@gmail.com>
Author: Attila Zsolt Piros <2017933+attilapiros@users.noreply.github.com>
Closes#21280 from attilapiros/SPARK-19181.
## What changes were proposed in this pull request?
Drastically improves performance and won't cause Spark applications to fail because they write too much data to the Docker image's specific file system. The file system's directories that back emptydir volumes are generally larger and more performant.
## How was this patch tested?
Has been in use via the prototype version of Kubernetes support, but lost in the transition to here.
Author: mcheah <mcheah@palantir.com>
Closes#21238 from mccheah/mount-local-dirs.
## What changes were proposed in this pull request?
I propose to add a clear statement for functions like `collect_list()` about non-deterministic behavior of such functions. The behavior must be taken into account by user while creating and running queries.
Author: Maxim Gekk <maxim.gekk@databricks.com>
Closes#21228 from MaxGekk/deterministic-comments.
## What changes were proposed in this pull request?
The PR introduces regr_slope, regr_intercept, regr_r2, regr_sxx, regr_syy, regr_sxy, regr_avgx, regr_avgy, regr_count.
The implementation of this functions mirrors Hive's one in HIVE-15978.
## How was this patch tested?
added UT (values compared with Hive)
Author: Marco Gaido <marcogaido91@gmail.com>
Closes#21054 from mgaido91/SPARK-23907.
## What changes were proposed in this pull request?
We reverted `spark.sql.hive.convertMetastoreOrc` at https://github.com/apache/spark/pull/20536 because we should not ignore the table-specific compression conf. Now, it's resolved via [SPARK-23355](8aa1d7b0ed).
## How was this patch tested?
Pass the Jenkins.
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#21186 from dongjoon-hyun/SPARK-24112.
## What changes were proposed in this pull request?
Renames:
* `DataReaderFactory` to `InputPartition`
* `DataReader` to `InputPartitionReader`
* `createDataReaderFactories` to `planInputPartitions`
* `createUnsafeDataReaderFactories` to `planUnsafeInputPartitions`
* `createBatchDataReaderFactories` to `planBatchInputPartitions`
This fixes the changes in SPARK-23219, which renamed ReadTask to
DataReaderFactory. The intent of that change was to make the read and
write API match (write side uses DataWriterFactory), but the underlying
problem is that the two classes are not equivalent.
ReadTask/DataReader function as Iterable/Iterator. One InputPartition is
a specific partition of the data to be read, in contrast to
DataWriterFactory where the same factory instance is used in all write
tasks. InputPartition's purpose is to manage the lifecycle of the
associated reader, which is now called InputPartitionReader, with an
explicit create operation to mirror the close operation. This was no
longer clear from the API because DataReaderFactory appeared to be more
generic than it is and it isn't clear why a set of them is produced for
a read.
## How was this patch tested?
Existing tests, which have been updated to use the new name.
Author: Ryan Blue <blue@apache.org>
Closes#21145 from rdblue/SPARK-24073-revert-data-reader-factory-rename.
## What changes were proposed in this pull request?
Add a new test that triggers if PARQUET-1217 - a predicate pushdown bug - is not fixed in Spark's Parquet dependency.
## How was this patch tested?
New unit test passes.
Author: Henry Robinson <henry@apache.org>
Closes#21284 from henryr/spark-23852.
## What changes were proposed in this pull request?
In method *CoarseGrainedSchedulerBackend.killExecutors()*, `numPendingExecutors` should add
`executorsToKill.size` rather than `knownExecutors.size` if we do not adjust target number of executors.
## How was this patch tested?
N/A
Author: wuyi <ngone_5451@163.com>
Closes#21209 from Ngone51/SPARK-24141.
## What changes were proposed in this pull request?
We should overwrite "otherCopyArgs" to provide the SparkSession parameter otherwise TreeNode.toJSON cannot get the full constructor parameter list.
## How was this patch tested?
The new unit test.
Author: Shixiong Zhu <zsxwing@gmail.com>
Closes#21275 from zsxwing/SPARK-24214.
## What changes were proposed in this pull request?
Provide evaluateEachIteration method or equivalent for spark.ml GBTs.
## How was this patch tested?
UT.
Please review http://spark.apache.org/contributing.html before opening a pull request.
Author: WeichenXu <weichen.xu@databricks.com>
Closes#21097 from WeichenXu123/GBTeval.
## What changes were proposed in this pull request?
This copies the material from the spark.mllib user guide page for Naive Bayes to the spark.ml user guide page. I also improved the wording and organization slightly.
## How was this patch tested?
Built docs locally.
Author: Joseph K. Bradley <joseph@databricks.com>
Closes#21272 from jkbradley/nb-doc-update.
## What changes were proposed in this pull request?
The exception message should clearly distinguish sorting and bucketing in `save` and `jdbc` write.
When a user tries to write a sorted data using save or insertInto, it will throw an exception with message that `s"'$operation' does not support bucketing right now""`.
We should throw `s"'$operation' does not support sortBy right now""` instead.
## How was this patch tested?
More tests in `DataFrameReaderWriterSuite.scala`
Author: DB Tsai <d_tsai@apple.com>
Closes#21235 from dbtsai/fixException.
## What changes were proposed in this pull request?
This updates Parquet to 1.10.0 and updates the vectorized path for buffer management changes. Parquet 1.10.0 uses ByteBufferInputStream instead of byte arrays in encoders. This allows Parquet to break allocations into smaller chunks that are better for garbage collection.
## How was this patch tested?
Existing Parquet tests. Running in production at Netflix for about 3 months.
Author: Ryan Blue <blue@apache.org>
Closes#21070 from rdblue/SPARK-23972-update-parquet-to-1.10.0.
## What changes were proposed in this pull request?
- Add OptionalInstrumentation as argument for getNumClasses in ml.classification.Classifier
- Change the function call for getNumClasses in train() in ml.classification.DecisionTreeClassifier, ml.classification.RandomForestClassifier, and ml.classification.NaiveBayes
- Modify the instrumentation creation in ml.classification.LinearSVC
- Change the log call in ml.classification.OneVsRest and ml.classification.LinearSVC
## How was this patch tested?
Manual.
Please review http://spark.apache.org/contributing.html before opening a pull request.
Author: Lu WANG <lu.wang@databricks.com>
Closes#21204 from ludatabricks/SPARK-23686.
## What changes were proposed in this pull request?
In Apache Spark 2.4, [SPARK-23355](https://issues.apache.org/jira/browse/SPARK-23355) fixes a bug which ignores table properties during convertMetastore for tables created by STORED AS ORC/PARQUET.
For some Parquet tables having table properties like TBLPROPERTIES (parquet.compression 'NONE'), it was ignored by default before Apache Spark 2.4. After upgrading cluster, Spark will write uncompressed file which is different from Apache Spark 2.3 and old.
This PR adds a migration note for that.
## How was this patch tested?
N/A
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#21269 from dongjoon-hyun/SPARK-23355-DOC.
## What changes were proposed in this pull request?
While reading CSV or JSON files, DataFrameReader's options are converted to Hadoop's parameters, for example there:
https://github.com/apache/spark/blob/branch-2.3/sql/core/src/main/scala/org/apache/spark/sql/execution/DataSourceScanExec.scala#L302
but the options are not propagated to Text datasource on schema inferring, for instance:
https://github.com/apache/spark/blob/branch-2.3/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/csv/CSVDataSource.scala#L184-L188
The PR proposes propagation of user's options to Text datasource on scheme inferring in similar way as user's options are converted to Hadoop parameters if schema is specified.
## How was this patch tested?
The changes were tested manually by using https://github.com/twitter/hadoop-lzo:
```
hadoop-lzo> mvn clean package
hadoop-lzo> ln -s ./target/hadoop-lzo-0.4.21-SNAPSHOT.jar ./hadoop-lzo.jar
```
Create 2 test files in JSON and CSV format and compress them:
```shell
$ cat test.csv
col1|col2
a|1
$ lzop test.csv
$ cat test.json
{"col1":"a","col2":1}
$ lzop test.json
```
Run `spark-shell` with hadoop-lzo:
```
bin/spark-shell --jars ~/hadoop-lzo/hadoop-lzo.jar
```
reading compressed CSV and JSON without schema:
```scala
spark.read.option("io.compression.codecs", "com.hadoop.compression.lzo.LzopCodec").option("inferSchema",true).option("header",true).option("sep","|").csv("test.csv.lzo").show()
+----+----+
|col1|col2|
+----+----+
| a| 1|
+----+----+
```
```scala
spark.read.option("io.compression.codecs", "com.hadoop.compression.lzo.LzopCodec").option("multiLine", true).json("test.json.lzo").printSchema
root
|-- col1: string (nullable = true)
|-- col2: long (nullable = true)
```
Author: Maxim Gekk <maxim.gekk@databricks.com>
Author: Maxim Gekk <max.gekk@gmail.com>
Closes#21182 from MaxGekk/text-options.
## What changes were proposed in this pull request?
This is to add a test case to check the behaviors when users write json in the specified UTF-16/UTF-32 encoding with multiline off.
## How was this patch tested?
N/A
Author: gatorsmile <gatorsmile@gmail.com>
Closes#21254 from gatorsmile/followupSPARK-23094.
## What changes were proposed in this pull request?
More close to Scala API behavior when can't parse input by throwing exception. Add tests.
## How was this patch tested?
Added tests.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#21211 from viirya/SPARK-24131-followup.
## What changes were proposed in this pull request?
HashAggregate uses the same hash algorithm and seed as ShuffleExchange, it may lead to bad hash conflict when shuffle.partitions=8192*n.
Considering below example:
```
SET spark.sql.shuffle.partitions=8192;
INSERT OVERWRITE TABLE target_xxx
SELECT
item_id,
auct_end_dt
FROM
from source_xxx
GROUP BY
item_id,
auct_end_dt;
```
In the shuffle stage, if user sets the shuffle.partition = 8192, all tuples in the same partition will meet the following relationship:
```
hash(tuple x) = hash(tuple y) + n * 8192
```
Then in the next HashAggregate stage, all tuples from the same partition need be put into a 16K BytesToBytesMap (unsafeRowAggBuffer).
Here, the HashAggregate uses the same hash algorithm on the same expression as shuffle, and uses the same seed, and 16K = 8192 * 2, so actually, all tuples in the same parititon will only be hashed to 2 different places in the BytesToBytesMap. It is bad hash conflict. With BytesToBytesMap growing, the conflict will always exist.
Before change:
<img width="334" alt="hash_conflict" src="https://user-images.githubusercontent.com/2989575/39250210-ed032d46-48d2-11e8-855a-c1afc2a0ceb5.png">
After change:
<img width="334" alt="no_hash_conflict" src="https://user-images.githubusercontent.com/2989575/39250218-f1cb89e0-48d2-11e8-9244-5a93c1e8b60d.png">
## How was this patch tested?
Unit tests and production cases.
Author: yucai <yyu1@ebay.com>
Closes#21149 from yucai/SPARK-24076.
It was missing the jax-rs annotation.
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#21245 from vanzin/SPARK-24188.
Change-Id: Ib338e34b363d7c729cc92202df020dc51033b719
## What changes were proposed in this pull request?
Mention `spark.sql.crossJoin.enabled` in error message when an implicit `CROSS JOIN` is detected.
## How was this patch tested?
`CartesianProductSuite` and `JoinSuite`.
Author: Henry Robinson <henry@apache.org>
Closes#21201 from henryr/spark-24128.
## What changes were proposed in this pull request?
Add support for all of the clustering methods
## How was this patch tested?
unit tests added
Please review http://spark.apache.org/contributing.html before opening a pull request.
Author: Lu WANG <lu.wang@databricks.com>
Closes#21195 from ludatabricks/SPARK-23975-1.
## What changes were proposed in this pull request?
PrefixSpan API for spark.ml. New implementation instead of #20810
## How was this patch tested?
TestSuite added.
Author: WeichenXu <weichen.xu@databricks.com>
Closes#20973 from WeichenXu123/prefixSpan2.
## What changes were proposed in this pull request?
ML test for StructuredStreaming: spark.ml.tuning
## How was this patch tested?
N/A
Author: WeichenXu <weichen.xu@databricks.com>
Closes#20261 from WeichenXu123/ml_stream_tuning_test.
## What changes were proposed in this pull request?
This PR fixes the migration note for SPARK-23291 since it's going to backport to 2.3.1. See the discussion in https://issues.apache.org/jira/browse/SPARK-23291
## How was this patch tested?
N/A
Author: hyukjinkwon <gurwls223@apache.org>
Closes#21249 from HyukjinKwon/SPARK-23291.
## What changes were proposed in this pull request?
Change FPGrowth from private to private[spark]. If no numPartitions is specified, then default value -1 is used. But -1 is only valid in the construction function of FPGrowth, but not in setNumPartitions. So I make this change and use the constructor directly rather than using set method.
## How was this patch tested?
Unit test is added
Author: Jeff Zhang <zjffdu@apache.org>
Closes#13493 from zjffdu/SPARK-15750.
## What changes were proposed in this pull request?
When creating an InterpretedPredicate instance, initialize any Nondeterministic expressions in the expression tree to avoid java.lang.IllegalArgumentException on later call to eval().
## How was this patch tested?
- sbt SQL tests
- python SQL tests
- new unit test
Author: Bruce Robbins <bersprockets@gmail.com>
Closes#21144 from bersprockets/interpretedpredicate.
## What changes were proposed in this pull request?
`LogicalPlan.resolve(...)` uses linear searches to find an attribute matching a name. This is fine in normal cases, but gets problematic when you try to resolve a large number of columns on a plan with a large number of attributes.
This PR adds an indexing structure to `resolve(...)` in order to find potential matches quicker. This PR improves the reference resolution time for the following code by 4x (11.8s -> 2.4s):
``` scala
val n = 4000
val values = (1 to n).map(_.toString).mkString(", ")
val columns = (1 to n).map("column" + _).mkString(", ")
val query =
s"""
|SELECT $columns
|FROM VALUES ($values) T($columns)
|WHERE 1=2 AND 1 IN ($columns)
|GROUP BY $columns
|ORDER BY $columns
|""".stripMargin
spark.time(sql(query))
```
## How was this patch tested?
Existing tests.
Author: Herman van Hovell <hvanhovell@databricks.com>
Closes#14083 from hvanhovell/SPARK-16406.
## What changes were proposed in this pull request?
The PR add the `slice` function. The behavior of the function is based on Presto's one.
The function slices an array according to the requested start index and length.
## How was this patch tested?
added UTs
Author: Marco Gaido <marcogaido91@gmail.com>
Closes#21040 from mgaido91/SPARK-23930.
## What changes were proposed in this pull request?
SPARK-24160 is causing a compilation failure (after SPARK-24143 was merged). This fixes the issue.
## How was this patch tested?
building successfully
Author: Marco Gaido <marcogaido91@gmail.com>
Closes#21256 from mgaido91/SPARK-24160_FOLLOWUP.
## What changes were proposed in this pull request?
DataFrameRangeSuite.test("Cancelling stage in a query with Range.") stays sometimes in an infinite loop and times out the build.
There were multiple issues with the test:
1. The first valid stageId is zero when the test started alone and not in a suite and the following code waits until timeout:
```
eventually(timeout(10.seconds), interval(1.millis)) {
assert(DataFrameRangeSuite.stageToKill > 0)
}
```
2. The `DataFrameRangeSuite.stageToKill` was overwritten by the task's thread after the reset which ended up in canceling the same stage 2 times. This caused the infinite wait.
This PR solves this mentioned flakyness by removing the shared `DataFrameRangeSuite.stageToKill` and using `onTaskStart` where stage ID is provided. In order to make sure cancelStage called for all stages `waitUntilEmpty` is called on `ListenerBus`.
In [PR20888](https://github.com/apache/spark/pull/20888) this tried to get solved by:
* Stopping the executor thread with `wait`
* Wait for all `cancelStage` called
* Kill the executor thread by setting `SparkContext.SPARK_JOB_INTERRUPT_ON_CANCEL`
but the thread killing left the shared `SparkContext` sometimes in a state where further jobs can't be submitted. As a result DataFrameRangeSuite.test("Cancelling stage in a query with Range.") test passed properly but the next test inside the suite was hanging.
## How was this patch tested?
Existing unit test executed 10k times.
Author: Gabor Somogyi <gabor.g.somogyi@gmail.com>
Closes#21214 from gaborgsomogyi/SPARK-23775_1.
## What changes were proposed in this pull request?
This patch modifies `ShuffleBlockFetcherIterator` so that the receipt of zero-size blocks is treated as an error. This is done as a preventative measure to guard against a potential source of data loss bugs.
In the shuffle layer, we guarantee that zero-size blocks will never be requested (a block containing zero records is always 0 bytes in size and is marked as empty such that it will never be legitimately requested by executors). However, the existing code does not fully take advantage of this invariant in the shuffle-read path: the existing code did not explicitly check whether blocks are non-zero-size.
Additionally, our decompression and deserialization streams treat zero-size inputs as empty streams rather than errors (EOF might actually be treated as "end-of-stream" in certain layers (longstanding behavior dating to earliest versions of Spark) and decompressors like Snappy may be tolerant to zero-size inputs).
As a result, if some other bug causes legitimate buffers to be replaced with zero-sized buffers (due to corruption on either the send or receive sides) then this would translate into silent data loss rather than an explicit fail-fast error.
This patch addresses this problem by adding a `buf.size != 0` check. See code comments for pointers to tests which guarantee the invariants relied on here.
## How was this patch tested?
Existing tests (which required modifications, since some were creating empty buffers in mocks). I also added a test to make sure we fail on zero-size blocks.
To test that the zero-size blocks are indeed a potential corruption source, I manually ran a workload in `spark-shell` with a modified build which replaces all buffers with zero-size buffers in the receive path.
Author: Josh Rosen <joshrosen@databricks.com>
Closes#21219 from JoshRosen/SPARK-24160.
## What changes were proposed in this pull request?
The PR adds the SQL function `array_sort`. The behavior of the function is based on Presto's one.
The function sorts the input array in ascending order. The elements of the input array must be orderable. Null elements will be placed at the end of the returned array.
## How was this patch tested?
Added UTs
Author: Kazuaki Ishizaki <ishizaki@jp.ibm.com>
Closes#21021 from kiszk/SPARK-23921.
## What changes were proposed in this pull request?
In current code(`MapOutputTracker.convertMapStatuses`), mapstatus are converted to (blockId, size) pair for all blocks – no matter the block is empty or not, which result in OOM when there are lots of consecutive empty blocks, especially when adaptive execution is enabled.
(blockId, size) pair is only used in `ShuffleBlockFetcherIterator` to control shuffle-read and only non-empty block request is sent. Can we just filter out the empty blocks in MapOutputTracker.convertMapStatuses and save memory?
## How was this patch tested?
not added yet.
Author: jinxing <jinxing6042@126.com>
Closes#21212 from jinxing64/SPARK-24143.
This avoids polluting and leaving garbage behind in /tmp, and allows the
usual build tools to clean up any leftover files.
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#21198 from vanzin/SPARK-24126.
## What changes were proposed in this pull request?
This refactors the external catalog to be an interface. It can be easier for the future work in the catalog federation. After the refactoring, `ExternalCatalog` is much cleaner without mixing the listener event generation logic.
## How was this patch tested?
The existing tests
Author: gatorsmile <gatorsmile@gmail.com>
Closes#21122 from gatorsmile/refactorExternalCatalog.
## What changes were proposed in this pull request?
add array flatten function to SparkR
## How was this patch tested?
Unit tests were added in R/pkg/tests/fulltests/test_sparkSQL.R
Author: Huaxin Gao <huaxing@us.ibm.com>
Closes#21244 from huaxingao/spark-24185.
## What changes were proposed in this pull request?
This PR enables the MicroBatchExecution to run no-data batches if some SparkPlan requires running another batch to output results based on updated watermark / processing time. In this PR, I have enabled streaming aggregations and streaming deduplicates to automatically run addition batch even if new data is available. See https://issues.apache.org/jira/browse/SPARK-24156 for more context.
Major changes/refactoring done in this PR.
- Refactoring MicroBatchExecution - A major point of confusion in MicroBatchExecution control flow was always (at least to me) was that `populateStartOffsets` internally called `constructNextBatch` which was not obvious from just the name "populateStartOffsets" and made the control flow from the main trigger execution loop very confusing (main loop in `runActivatedStream` called `constructNextBatch` but only if `populateStartOffsets` hadn't already called it). Instead, the refactoring makes it cleaner.
- `populateStartOffsets` only the updates `availableOffsets` and `committedOffsets`. Does not call `constructNextBatch`.
- Main loop in `runActivatedStream` calls `constructNextBatch` which returns true or false reflecting whether the next batch is ready for executing. This method is now idempotent; if a batch has already been constructed, then it will always return true until the batch has been executed.
- If next batch is ready then we call `runBatch` or sleep.
- That's it.
- Refactoring watermark management logic - This has been refactored out from `MicroBatchExecution` in a separate class to simplify `MicroBatchExecution`.
- New method `shouldRunAnotherBatch` in `IncrementalExecution` - This returns true if there is any stateful operation in the last execution plan that requires another batch for state cleanup, etc. This is used to decide whether to construct a batch or not in `constructNextBatch`.
- Changes to stream testing framework - Many tests used CheckLastBatch to validate answers. This assumed that there will be no more batches after the last set of input has been processed, so the last batch is the one that has output corresponding to the last input. This is not true anymore. To account for that, I made two changes.
- `CheckNewAnswer` is a new test action that verifies the new rows generated since the last time the answer was checked by `CheckAnswer`, `CheckNewAnswer` or `CheckLastBatch`. This is agnostic to how many batches occurred between the last check and now. To do make this easier, I added a common trait between MemorySink and MemorySinkV2 to abstract out some common methods.
- `assertNumStateRows` has been updated in the same way to be agnostic to batches while checking what the total rows and how many state rows were updated (sums up updates since the last check).
## How was this patch tested?
- Changes made to existing tests - Tests have been changed in one of the following patterns.
- Tests where the last input was given again to force another batch to be executed and state cleaned up / output generated, they were simplified by removing the extra input.
- Tests using aggregation+watermark where CheckLastBatch were replaced with CheckNewAnswer to make them batch agnostic.
- New tests added to check whether the flag works for streaming aggregation and deduplication
Author: Tathagata Das <tathagata.das1565@gmail.com>
Closes#21220 from tdas/SPARK-24157.
## What changes were proposed in this pull request?
Do continuous processing writes with multiple compute() calls.
The current strategy (before this PR) is hacky; we just call next() on an iterator which has already returned hasNext = false, knowing that all the nodes we whitelist handle this properly. This will have to be changed before we can support more complex query plans. (In particular, I have a WIP https://github.com/jose-torres/spark/pull/13 which should be able to support aggregates in a single partition with minimal additional work.)
Most of the changes here are just refactoring to accommodate the new model. The behavioral changes are:
* The writer now calls prev.compute(split, context) once per epoch within the epoch loop.
* ContinuousDataSourceRDD now spawns a ContinuousQueuedDataReader which is shared across multiple calls to compute() for the same partition.
## How was this patch tested?
existing unit tests
Author: Jose Torres <torres.joseph.f+github@gmail.com>
Closes#21200 from jose-torres/noAggr.
…path and set permissions properly
## What changes were proposed in this pull request?
Spark history server should create spark.history.store.path and set permissions properly. Note createdDirectories doesn't do anything if the directories are already created. This does not stomp on the permissions if the user had manually created the directory before the history server could.
## How was this patch tested?
Manually tested in a 100 node cluster. Ensured directories created with proper permissions. Ensured restarted worked apps/temp directories worked as apps were read.
Author: Thomas Graves <tgraves@thirteenroutine.corp.gq1.yahoo.com>
Closes#21234 from tgravescs/SPARK-24124.
## What changes were proposed in this pull request?
It's possible that Accumulators of Spark 1.x may no longer work with Spark 2.x. This is because `LegacyAccumulatorWrapper.isZero` may return wrong answer if `AccumulableParam` doesn't define equals/hashCode.
This PR fixes this by using reference equality check in `LegacyAccumulatorWrapper.isZero`.
## How was this patch tested?
a new test
Author: Wenchen Fan <wenchen@databricks.com>
Closes#21229 from cloud-fan/accumulator.
## What changes were proposed in this pull request?
Avoid unnecessary sleep (10 ms) in each invocation of MemoryStreamDataReader.next.
## How was this patch tested?
Ran ContinuousSuite from IDE.
Please review http://spark.apache.org/contributing.html before opening a pull request.
Author: Arun Mahadevan <arunm@apache.org>
Closes#21207 from arunmahadevan/memorystream.