Hi all,
Jackson is incompatible with upstream versions, therefore bump the Jackson version to a more recent one. I bumped into some issues with Azure CosmosDB that is using a more recent version of Jackson. This can be fixed by adding exclusions and then it works without any issues. So no breaking changes in the API's.
I would also consider bumping the version of Jackson in Spark. I would suggest to keep up to date with the dependencies, since in the future this issue will pop up more frequently.
## What changes were proposed in this pull request?
Bump Jackson to 2.9.6
## How was this patch tested?
Compiled and tested it locally to see if anything broke.
Please review http://spark.apache.org/contributing.html before opening a pull request.
Closes#21596 from Fokko/fd-bump-jackson.
Authored-by: Fokko Driesprong <fokkodriesprong@godatadriven.com>
Signed-off-by: hyukjinkwon <gurwls223@apache.org>
While working on another PR, I noticed that there is quite some legacy Java in there that can be beautified. For example the use og features from Java8, such as:
- Collection libraries
- Try-with-resource blocks
No code has been changed
What are your thoughts on this?
This makes code easier to read, and using try-with-resource makes is less likely to forget to close something.
## What changes were proposed in this pull request?
(Please fill in changes proposed in this fix)
## How was this patch tested?
(Please explain how this patch was tested. E.g. unit tests, integration tests, manual tests)
(If this patch involves UI changes, please attach a screenshot; otherwise, remove this)
Please review http://spark.apache.org/contributing.html before opening a pull request.
Closes#22399 from Fokko/SPARK-25408.
Authored-by: Fokko Driesprong <fokkodriesprong@godatadriven.com>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
## What changes were proposed in this pull request?
Cause: Recently test_glr_summary failed for PR of SPARK-25118, which enables
spark-shell to run with default log level. It failed because this logdebug was
called for GeneralizedLinearRegressionTrainingSummary which invoked its toString
method, which started a Spark Job and ended up running into an infinite loop.
Fix: Remove logDebug statement for outer objects as closures aren't implemented
with outerclasses in Scala 2.12 and this debug statement looses its purpose
## How was this patch tested?
Ran python pyspark-ml tests on top of PR for SPARK-25118 and ClosureCleaner unit
tests
Closes#22616 from ankuriitg/ankur/SPARK-25586.
Authored-by: ankurgupta <ankur.gupta@cloudera.com>
Signed-off-by: Marcelo Vanzin <vanzin@cloudera.com>
## What changes were proposed in this pull request?
Rename method `benchmark` in `BenchmarkBase` as `runBenchmarkSuite `. Also add comments.
Currently the method name `benchmark` is a bit confusing. Also the name is the same as instances of `Benchmark`:
f246813afb/sql/hive/src/test/scala/org/apache/spark/sql/hive/orc/OrcReadBenchmark.scala (L330-L339)
## How was this patch tested?
Unit test.
Closes#22599 from gengliangwang/renameBenchmarkSuite.
Authored-by: Gengliang Wang <gengliang.wang@databricks.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
## What changes were proposed in this pull request?
This patch is to bump the master branch version to 3.0.0-SNAPSHOT.
## How was this patch tested?
N/A
Closes#22606 from gatorsmile/bump3.0.
Authored-by: gatorsmile <gatorsmile@gmail.com>
Signed-off-by: gatorsmile <gatorsmile@gmail.com>
## What changes were proposed in this pull request?
Currently, SQL tab in the WEBUI doesn't support hiding table. Other tabs in the web ui like, Jobs, stages etc supports hiding table (refer SPARK-23024 https://github.com/apache/spark/pull/20216).
In this PR, added the support for hide table in the sql tab also.
## How was this patch tested?
bin/spark-shell
```
sql("create table a (id int)")
for(i <- 1 to 100) sql(s"insert into a values ($i)")
```
Open SQL tab in the web UI
**Before fix:**
**After fix:** Consistent with the other tabs.
(Please explain how this patch was tested. E.g. unit tests, integration tests, manual tests)
(If this patch involves UI changes, please attach a screenshot; otherwise, remove this)
Please review http://spark.apache.org/contributing.html before opening a pull request.
Closes#22592 from shahidki31/SPARK-25575.
Authored-by: Shahid <shahidki31@gmail.com>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
## What changes were proposed in this pull request?
Fix an obvious error.
## How was this patch tested?
Existing tests.
Closes#22577 from sadhen/minor_fix.
Authored-by: Darcy Shen <sadhen@zoho.com>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
## What changes were proposed in this pull request?
This PR adds a rule to force `.toLowerCase(Locale.ROOT)` or `toUpperCase(Locale.ROOT)`.
It produces an error as below:
```
[error] Are you sure that you want to use toUpperCase or toLowerCase without the root locale? In most cases, you
[error] should use toUpperCase(Locale.ROOT) or toLowerCase(Locale.ROOT) instead.
[error] If you must use toUpperCase or toLowerCase without the root locale, wrap the code block with
[error] // scalastyle:off caselocale
[error] .toUpperCase
[error] .toLowerCase
[error] // scalastyle:on caselocale
```
This PR excludes the cases above for SQL code path for external calls like table name, column name and etc.
For test suites, or when it's clear there's no locale problem like Turkish locale problem, it uses `Locale.ROOT`.
One minor problem is, `UTF8String` has both methods, `toLowerCase` and `toUpperCase`, and the new rule detects them as well. They are ignored.
## How was this patch tested?
Manually tested, and Jenkins tests.
Closes#22581 from HyukjinKwon/SPARK-25565.
Authored-by: hyukjinkwon <gurwls223@apache.org>
Signed-off-by: hyukjinkwon <gurwls223@apache.org>
## What changes were proposed in this pull request?
Since we don't fail a job when `AccumulatorV2.merge` fails, we should try to update the remaining accumulators so that they can still report correct values.
## How was this patch tested?
The new unit test.
Closes#22586 from zsxwing/SPARK-25568.
Authored-by: Shixiong Zhu <zsxwing@gmail.com>
Signed-off-by: gatorsmile <gatorsmile@gmail.com>
## What changes were proposed in this pull request?
Heartbeat shouldn't include accumulators for zero metrics.
Heartbeats sent from executors to the driver every 10 seconds contain metrics and are generally on the order of a few KBs. However, for large jobs with lots of tasks, heartbeats can be on the order of tens of MBs, causing tasks to die with heartbeat failures. We can mitigate this by not sending zero metrics to the driver.
## How was this patch tested?
(Please explain how this patch was tested. E.g. unit tests, integration tests, manual tests)
(If this patch involves UI changes, please attach a screenshot; otherwise, remove this)
Please review http://spark.apache.org/contributing.html before opening a pull request.
Closes#22473 from mukulmurthy/25449-heartbeat.
Authored-by: Mukul Murthy <mukul.murthy@gmail.com>
Signed-off-by: Shixiong Zhu <zsxwing@gmail.com>
## What changes were proposed in this pull request?
The specified test in OpenHashMapSuite to test large items is somehow flaky to throw OOM.
By considering the original work #6763 that added this test, the test can be against OpenHashSetSuite. And by doing this should be to save memory because OpenHashMap allocates two more arrays when growing the map/set.
## How was this patch tested?
Existing tests.
Closes#22569 from viirya/SPARK-25542.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
Caching the value of that config means different instances of SparkEnv
will always use whatever was the first value to be read. It also breaks
tests that use RDDInfo outside of the scope of a SparkContext.
Since this is not a performance sensitive area, there's no advantage
in caching the config value.
Closes#22558 from vanzin/SPARK-25546.
Authored-by: Marcelo Vanzin <vanzin@cloudera.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
## What changes were proposed in this pull request?
Special case the situation where we know the partioner and the number of requested partions output is the same as the current partioner to avoid a shuffle and instead compute distinct inside of each partion.
## How was this patch tested?
New unit test that verifies partitioner does not change if the partitioner is known and distinct is called with the same target # of partition.
Closes#22010 from holdenk/SPARK-21436-take-advantage-of-known-partioner-for-distinct-on-rdds.
Authored-by: Holden Karau <holden@pigscanfly.ca>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
changed metric value of METRIC_OUTPUT_RECORDS_WRITTEN from 'task.metrics.inputMetrics.recordsRead' to 'task.metrics.outputMetrics.recordsWritten'.
This bug was introduced in SPARK-22190. https://github.com/apache/spark/pull/19426
## How was this patch tested?
Existing tests
Closes#22555 from shahidki31/SPARK-25536.
Authored-by: Shahid <shahidki31@gmail.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
## What changes were proposed in this pull request?
SparkSubmit already logs in the user if a keytab is provided, the only issue is that it uses the existing configs which have "yarn" in their name. As such, the configs were changed to:
`spark.kerberos.keytab` and `spark.kerberos.principal`.
## How was this patch tested?
Will be tested with K8S tests, but needs to be tested with Yarn
- [x] K8S Secure HDFS tests
- [x] Yarn Secure HDFS tests vanzin
Closes#22362 from ifilonenko/SPARK-25372.
Authored-by: Ilan Filonenko <if56@cornell.edu>
Signed-off-by: Marcelo Vanzin <vanzin@cloudera.com>
## What changes were proposed in this pull request?
Previously SPARK-24519 created a modifiable config SHUFFLE_MIN_NUM_PARTS_TO_HIGHLY_COMPRESS. However, the config is being parsed for every creation of MapStatus, which could be very expensive. Another problem with the previous approach is that it created the illusion that this can be changed dynamically at runtime, which was not true. This PR changes it so the config is computed only once.
## How was this patch tested?
Removed a test case that's no longer valid.
Closes#22521 from rxin/SPARK-24519.
Authored-by: Reynold Xin <rxin@databricks.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
SPARK-4105 provided a solution to block corruption issue by retrying the fetch or the stage. In that solution there is a step that wraps the input stream with compression and/or encryption. This step is prone to exceptions, but in the current code there is no exception handling for this step and this has caused confusion for the user. The confusion was that after SPARK-4105 the user expects to see either a fetchFailed exception or a warning about a corrupted block. However an exception during wrapping can fail the job without any of those. This change adds exception handling for the wrapping step and also adds a fetch retry if we experience a corruption during the wrapping step. The reason for adding the retry is that usually user won't experience the same failure after rerunning the job and so it seems reasonable try to fetch and wrap one more time instead of failing.
Closes#22325 from rezasafi/localcorruption.
Authored-by: Reza Safi <rezasafi@cloudera.com>
Signed-off-by: Marcelo Vanzin <vanzin@cloudera.com>
## What changes were proposed in this pull request?
SHS V2 cannot enabled in Windows, because windows doesn't support POSIX permission.
## How was this patch tested?
test case fails in windows without this fix.
org.apache.spark.deploy.history.HistoryServerDiskManagerSuite test("leasing space")
SHS V2 cannot run successfully in Windows without this fix.
java.lang.UnsupportedOperationException: 'posix:permissions' not supported as initial attribute
at sun.nio.fs.WindowsSecurityDescriptor.fromAttribute(WindowsSecurityDescriptor.java:358)
Closes#22520 from jianjianjiao/FixWindowsPermssionsIssue.
Authored-by: Rong Tang <rotang@microsoft.com>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
After data has been streamed to disk, the buffers are inserted into the
memory store in some cases (eg., with broadcast blocks). But broadcast
code also disposes of those buffers when the data has been read, to
ensure that we don't leave mapped buffers using up memory, which then
leads to garbage data in the memory store.
## How was this patch tested?
Ran the old failing test in a loop. Full tests on jenkins
Closes#22546 from squito/SPARK-25422-master.
Authored-by: Imran Rashid <irashid@cloudera.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
Test steps :
1) bin/spark-shell --conf spark.ui.retainedTasks=200
```
val rdd = sc.parallelize(1 to 1000, 1000)
rdd.count
```
Stage tab in the UI will display 10 pages with 100 tasks per page. But number of retained tasks is only 200. So, from the 3rd page onwards will display nothing.
We have to calculate total pages based on the number of tasks need display in the UI.
**Before fix:**
**After fix:**
## How was this patch tested?
Manually tested
Closes#22526 from shahidki31/SPARK-25502.
Authored-by: Shahid <shahidki31@gmail.com>
Signed-off-by: Marcelo Vanzin <vanzin@cloudera.com>
## What changes were proposed in this pull request?
Currently there are two classes with the same naming BenchmarkBase:
1. `org.apache.spark.util.BenchmarkBase`
2. `org.apache.spark.sql.execution.benchmark.BenchmarkBase`
This is very confusing. And the benchmark object `org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark` is using the one in `org.apache.spark.util.BenchmarkBase`, while there is another class `BenchmarkBase` in the same package of it...
Here I propose:
1. the package `org.apache.spark.util.BenchmarkBase` should be in test package of core module. Move it to package `org.apache.spark.benchmark` .
2. Move `org.apache.spark.util.Benchmark` to test package of core module. Move it to package `org.apache.spark.benchmark` .
3. Rename the class `org.apache.spark.sql.execution.benchmark.BenchmarkBase` as `BenchmarkWithCodegen`
## How was this patch tested?
Unit test
Closes#22513 from gengliangwang/refactorBenchmarkBase.
Authored-by: Gengliang Wang <gengliang.wang@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
A continuation of squito's executor plugin task. By his request I took his code and added testing and moved the plugin initialization to a separate thread.
Executor plugins now run on one separate thread, so the executor does not wait on them. Added testing.
## How was this patch tested?
Added test cases that test using a sample plugin.
Closes#22192 from NiharS/executorPlugin.
Lead-authored-by: Nihar Sheth <niharrsheth@gmail.com>
Co-authored-by: NiharS <niharrsheth@gmail.com>
Signed-off-by: Marcelo Vanzin <vanzin@cloudera.com>
## What changes were proposed in this pull request?
This PR proposes to use add a helper in `PythonUtils` instead of direct accessing Scala package.
## How was this patch tested?
Jenkins tests.
Closes#22483 from HyukjinKwon/minor-refactoring.
Authored-by: hyukjinkwon <gurwls223@apache.org>
Signed-off-by: hyukjinkwon <gurwls223@apache.org>
## What changes were proposed in this pull request?
This goes to revert sequential PRs based on some discussion and comments at https://github.com/apache/spark/pull/16677#issuecomment-422650759.
#22344#22330#22239#16677
## How was this patch tested?
Existing tests.
Closes#22481 from viirya/revert-SPARK-19355-1.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
Refactor `FilterPushdownBenchmark` use `main` method. we can use 3 ways to run this test now:
1. bin/spark-submit --class org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark spark-sql_2.11-2.5.0-SNAPSHOT-tests.jar
2. build/sbt "sql/test:runMain org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark"
3. SPARK_GENERATE_BENCHMARK_FILES=1 build/sbt "sql/test:runMain org.apache.spark.sql.execution.benchmark.FilterPushdownBenchmark"
The method 2 and the method 3 do not need to compile the `spark-sql_*-tests.jar` package. So these two methods are mainly for developers to quickly do benchmark.
## How was this patch tested?
manual tests
Closes#22443 from wangyum/SPARK-25339.
Authored-by: Yuming Wang <yumwang@ebay.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
In the dev list, we can still discuss whether the next version is 2.5.0 or 3.0.0. Let us first bump the master branch version to `2.5.0-SNAPSHOT`.
## How was this patch tested?
N/A
Closes#22426 from gatorsmile/bumpVersionMaster.
Authored-by: gatorsmile <gatorsmile@gmail.com>
Signed-off-by: gatorsmile <gatorsmile@gmail.com>
We've seen some flakiness in jenkins in SchedulerIntegrationSuite which looks like it just needs a
longer timeout.
Closes#22385 from squito/SPARK-25400.
Authored-by: Imran Rashid <irashid@cloudera.com>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
## What changes were proposed in this pull request?
This PR ensures to call `super.afterAll()` in `override afterAll()` method for test suites.
* Some suites did not call `super.afterAll()`
* Some suites may call `super.afterAll()` only under certain condition
* Others never call `super.afterAll()`.
This PR also ensures to call `super.beforeAll()` in `override beforeAll()` for test suites.
## How was this patch tested?
Existing UTs
Closes#22337 from kiszk/SPARK-25338.
Authored-by: Kazuaki Ishizaki <ishizaki@jp.ibm.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
This is a rework of #21433 to address some concerns there.
Closes#22398 from michaelmior/long-callsite2.
Authored-by: Michael Mior <mmior@uwaterloo.ca>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
Correct some comparisons between unrelated types to what they seem to… have been trying to do
## How was this patch tested?
Existing tests.
Closes#22384 from srowen/SPARK-25398.
Authored-by: Sean Owen <sean.owen@databricks.com>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
## What changes were proposed in this pull request?
We will update block info coming from executors, at the timing like caching a RDD. However, when removing RDDs with unpersisting, we don't ask to update block info. So the block info is not updated.
We can fix this with few options:
1. Ask to update block info when unpersisting
This is simplest but changes driver-executor communication a bit.
2. Update block info when processing the event of unpersisting RDD
We send a `SparkListenerUnpersistRDD` event when unpersisting RDD. When processing this event, we can update block info of the RDD. This only changes event processing code so the risk seems to be lower.
Currently this patch takes option 2 for lower risk. If we agree first option has no risk, we can change to it.
## How was this patch tested?
Unit tests.
Closes#22341 from viirya/SPARK-24889.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: Marcelo Vanzin <vanzin@cloudera.com>
## What changes were proposed in this pull request?
Stop trimming values of properties loaded from a file
## How was this patch tested?
Added unit test demonstrating the issue hit in production.
Closes#22213 from gerashegalov/gera/SPARK-25221.
Authored-by: Gera Shegalov <gera@apache.org>
Signed-off-by: Marcelo Vanzin <vanzin@cloudera.com>
## What changes were proposed in this pull request?
When running TPC-DS benchmarks on 2.4 release, npoggi and winglungngai saw more than 10% performance regression on the following queries: q67, q24a and q24b. After we applying the PR https://github.com/apache/spark/pull/22338, the performance regression still exists. If we revert the changes in https://github.com/apache/spark/pull/19222, npoggi and winglungngai found the performance regression was resolved. Thus, this PR is to revert the related changes for unblocking the 2.4 release.
In the future release, we still can continue the investigation and find out the root cause of the regression.
## How was this patch tested?
The existing test cases
Closes#22361 from gatorsmile/revertMemoryBlock.
Authored-by: gatorsmile <gatorsmile@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
Add new executor level memory metrics (JVM used memory, on/off heap execution memory, on/off heap storage memory, on/off heap unified memory, direct memory, and mapped memory), and expose via the executors REST API. This information will help provide insight into how executor and driver JVM memory is used, and for the different memory regions. It can be used to help determine good values for spark.executor.memory, spark.driver.memory, spark.memory.fraction, and spark.memory.storageFraction.
## What changes were proposed in this pull request?
An ExecutorMetrics class is added, with jvmUsedHeapMemory, jvmUsedNonHeapMemory, onHeapExecutionMemory, offHeapExecutionMemory, onHeapStorageMemory, and offHeapStorageMemory, onHeapUnifiedMemory, offHeapUnifiedMemory, directMemory and mappedMemory. The new ExecutorMetrics is sent by executors to the driver as part of the Heartbeat. A heartbeat is added for the driver as well, to collect these metrics for the driver.
The EventLoggingListener store information about the peak values for each metric, per active stage and executor. When a StageCompleted event is seen, a StageExecutorsMetrics event will be logged for each executor, with peak values for the stage.
The AppStatusListener records the peak values for each memory metric.
The new memory metrics are added to the executors REST API.
## How was this patch tested?
New unit tests have been added. This was also tested on our cluster.
Author: Edwina Lu <edlu@linkedin.com>
Author: Imran Rashid <irashid@cloudera.com>
Author: edwinalu <edwina.lu@gmail.com>
Closes#21221 from edwinalu/SPARK-23429.2.
## What changes were proposed in this pull request?
This adds a test following https://github.com/apache/spark/pull/21638
## How was this patch tested?
Existing tests and new test.
Closes#22356 from srowen/SPARK-22357.2.
Authored-by: Sean Owen <sean.owen@databricks.com>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
This is not a perfect solution. It is designed to minimize complexity on the basis of solving problems.
It is effective for English, Chinese characters, Japanese, Korean and so on.
```scala
before:
+---+---------------------------+-------------+
|id |中国 |s2 |
+---+---------------------------+-------------+
|1 |ab |[a] |
|2 |null |[中国, abc] |
|3 |ab1 |[hello world]|
|4 |か行 きゃ(kya) きゅ(kyu) きょ(kyo) |[“中国] |
|5 |中国（你好）a |[“中（国）, 312] |
|6 |中国山(东)服务区 |[“中(国）] |
|7 |中国山东服务区 |[中(国)] |
|8 | |[中国] |
+---+---------------------------+-------------+
after:
+---+-----------------------------------+----------------+
|id |中国 |s2 |
+---+-----------------------------------+----------------+
|1 |ab |[a] |
|2 |null |[中国, abc] |
|3 |ab1 |[hello world] |
|4 |か行 きゃ(kya) きゅ(kyu) きょ(kyo) |[“中国] |
|5 |中国（你好）a |[“中（国）, 312]|
|6 |中国山(东)服务区 |[“中(国）] |
|7 |中国山东服务区 |[中(国)] |
|8 | |[中国] |
+---+-----------------------------------+----------------+
```
## What changes were proposed in this pull request?
When there are wide characters such as Chinese characters or Japanese characters in the data, the show method has a alignment problem.
Try to fix this problem.
## How was this patch tested?
(Please explain how this patch was tested. E.g. unit tests, integration tests, manual tests)
Please review http://spark.apache.org/contributing.html before opening a pull request.
Closes#22048 from xuejianbest/master.
Authored-by: xuejianbest <384329882@qq.com>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
## What changes were proposed in this pull request?
Upgrade chill to 0.9.3, Kryo to 4.0.2, to get bug fixes and improvements.
The resolved tickets includes:
- SPARK-25258 Upgrade kryo package to version 4.0.2
- SPARK-23131 Kryo raises StackOverflow during serializing GLR model
- SPARK-25176 Kryo fails to serialize a parametrised type hierarchy
More details:
https://github.com/twitter/chill/releases/tag/v0.9.3cc3910d501
## How was this patch tested?
Existing tests.
Closes#22179 from wangyum/SPARK-23131.
Lead-authored-by: Yuming Wang <yumwang@ebay.com>
Co-authored-by: Dongjoon Hyun <dongjoon@apache.org>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
## What changes were proposed in this pull request?
An alternative fix for https://github.com/apache/spark/pull/21698
When Spark rerun tasks for an RDD, there are 3 different behaviors:
1. determinate. Always return the same result with same order when rerun.
2. unordered. Returns same data set in random order when rerun.
3. indeterminate. Returns different result when rerun.
Normally Spark doesn't need to care about it. Spark runs stages one by one, when a task is failed, just rerun it. Although the rerun task may return a different result, users will not be surprised.
However, Spark may rerun a finished stage when seeing fetch failures. When this happens, Spark needs to rerun all the tasks of all the succeeding stages if the RDD output is indeterminate, because the input of the succeeding stages has been changed.
If the RDD output is determinate, we only need to rerun the failed tasks of the succeeding stages, because the input doesn't change.
If the RDD output is unordered, it's same as determinate, because shuffle partitioner is always deterministic(round-robin partitioner is not a shuffle partitioner that extends `org.apache.spark.Partitioner`), so the reducers will still get the same input data set.
This PR fixed the failure handling for `repartition`, to avoid correctness issues.
For `repartition`, it applies a stateful map function to generate a round-robin id, which is order sensitive and makes the RDD's output indeterminate. When the stage contains `repartition` reruns, we must also rerun all the tasks of all the succeeding stages.
**future improvement:**
1. Currently we can't rollback and rerun a shuffle map stage, and just fail. We should fix it later. https://issues.apache.org/jira/browse/SPARK-25341
2. Currently we can't rollback and rerun a result stage, and just fail. We should fix it later. https://issues.apache.org/jira/browse/SPARK-25342
3. We should provide public API to allow users to tag the random level of the RDD's computing function.
## How is this pull request tested?
a new test case
Closes#22112 from cloud-fan/repartition.
Lead-authored-by: Wenchen Fan <wenchen@databricks.com>
Co-authored-by: Xingbo Jiang <xingbo.jiang@databricks.com>
Signed-off-by: Xiao Li <gatorsmile@gmail.com>
Running a large Spark job with speculation turned on was causing executor heartbeats to time out on the driver end after sometime and eventually, after hitting the max number of executor failures, the job would fail.
## What changes were proposed in this pull request?
The main reason for the heartbeat timeouts was that the heartbeat-receiver-event-loop-thread was blocked waiting on the TaskSchedulerImpl object which was being held by one of the dispatcher-event-loop threads executing the method dequeueSpeculativeTasks() in TaskSetManager.scala. On further analysis of the heartbeat receiver method executorHeartbeatReceived() in TaskSchedulerImpl class, we found out that instead of waiting to acquire the lock on the TaskSchedulerImpl object, we can remove that lock and make the operations to the global variables inside the code block to be atomic. The block of code in that method only uses one global HashMap taskIdToTaskSetManager. Making that map a ConcurrentHashMap, we are ensuring atomicity of operations and speeding up the heartbeat receiver thread operation.
## How was this patch tested?
Screenshots of the thread dump have been attached below:
**heartbeat-receiver-event-loop-thread:**
<img width="1409" alt="screen shot 2018-08-24 at 9 19 57 am" src="https://user-images.githubusercontent.com/22228190/44593413-e25df780-a788-11e8-9520-176a18401a59.png">
**dispatcher-event-loop-thread:**
<img width="1409" alt="screen shot 2018-08-24 at 9 21 56 am" src="https://user-images.githubusercontent.com/22228190/44593484-13d6c300-a789-11e8-8d88-34b1d51d4541.png">
Closes#22221 from pgandhi999/SPARK-25231.
Authored-by: pgandhi <pgandhi@oath.com>
Signed-off-by: Thomas Graves <tgraves@apache.org>
The problem occurs because stage object is removed from liveStages in
AppStatusListener onStageCompletion. Because of this any onTaskEnd event
received after onStageCompletion event do not update stage metrics.
The fix is to retain stage objects in liveStages until all tasks are complete.
1. Fixed the reproducible example posted in the JIRA
2. Added unit test
Closes#22209 from ankuriitg/ankurgupta/SPARK-24415.
Authored-by: ankurgupta <ankur.gupta@cloudera.com>
Signed-off-by: Marcelo Vanzin <vanzin@cloudera.com>
## What changes were proposed in this pull request?
Add a new metric to measure the executor's process (JVM) CPU time.
## How was this patch tested?
Manually tested on a Spark cluster (see SPARK-25228 for an example screenshot).
Closes#22218 from LucaCanali/AddExecutrCPUTimeMetric.
Authored-by: LucaCanali <luca.canali@cern.ch>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
