The bug description is a little misleading: the actual issue is that
.py files are not handled correctly when distributed by YARN. They're
added to "spark.submit.pyFiles", which, when processed by context.py,
explicitly whitelists certain extensions (see PACKAGE_EXTENSIONS),
and that does not include .py files.
On top of that, archives were not handled at all! They made it to the
driver's python path, but never made it to executors, since the mechanism
used to propagate their location (spark.submit.pyFiles) only works on
the driver side.
So, instead, ignore "spark.submit.pyFiles" and just build PYTHONPATH
correctly for both driver and executors. Individual .py files are
placed in a subdirectory of the container's local dir in the cluster,
which is then added to the python path. Archives are added directly.
The change, as a side effect, ends up solving the symptom described
in the bug. The issue was not that the files were not being distributed,
but that they were never made visible to the python application
running under Spark.
Also included is a proper unit test for running python on YARN, which
broke in several different ways with the previous code.
A short walk around of the changes:
- SparkSubmit does not try to be smart about how YARN handles python
files anymore. It just passes down the configs to the YARN client
code.
- The YARN client distributes python files and archives differently,
placing the files in a subdirectory.
- The YARN client now sets PYTHONPATH for the processes it launches;
to properly handle different locations, it uses YARN's support for
embedding env variables, so to avoid YARN expanding those at the
wrong time, SparkConf is now propagated to the AM using a conf file
instead of command line options.
- Because the Client initialization code is a maze of implicit
dependencies, some code needed to be moved around to make sure
all needed state was available when the code ran.
- The pyspark tests in YarnClusterSuite now actually distribute and try
to use both a python file and an archive containing a different python
module. Also added a yarn-client tests for completeness.
- I cleaned up some of the code around distributing files to YARN, to
avoid adding more copied & pasted code to handle the new files being
distributed.
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#6360 from vanzin/SPARK-5479 and squashes the following commits:
bcaf7e6 [Marcelo Vanzin] Feedback.
c47501f [Marcelo Vanzin] Fix yarn-client mode.
46b1d0c [Marcelo Vanzin] Merge branch 'master' into SPARK-5479
c743778 [Marcelo Vanzin] Only pyspark cares about python archives.
c8e5a82 [Marcelo Vanzin] Actually run pyspark in client mode.
705571d [Marcelo Vanzin] Move some code to the YARN module.
1dd4d0c [Marcelo Vanzin] Review feedback.
71ee736 [Marcelo Vanzin] Merge branch 'master' into SPARK-5479
220358b [Marcelo Vanzin] Scalastyle.
cdbb990 [Marcelo Vanzin] Merge branch 'master' into SPARK-5479
7fe3cd4 [Marcelo Vanzin] No need to distribute primary file to executors.
09045f1 [Marcelo Vanzin] Style.
943cbf4 [Marcelo Vanzin] [SPARK-5479] [yarn] Handle --py-files correctly in YARN.
Changed ```SparkEnv.actorSystem``` to be a function such that we can use the deprecated flag with it and added a deprecated message.
Author: Ilya Ganelin <ilya.ganelin@capitalone.com>
Closes#6731 from ilganeli/SPARK-7996 and squashes the following commits:
be43817 [Ilya Ganelin] Restored to val
9ed89e7 [Ilya Ganelin] Added a version info for deprecation
9610b08 [Ilya Ganelin] Converted actorSystem to function and added deprecated flag
Just as a safeguard against DOM rewriting.
Author: Andrew Or <andrew@databricks.com>
Closes#6732 from andrewor14/dag-viz-trim and squashes the following commits:
7e9bacb [Andrew Or] [MINOR] [UI] DAG visualization: trim whitespace from input
This was caused by this commit: f271347
This patch does not attempt to fix the root cause of why the `VisibleForTesting` annotation causes a NPE in the shell. We should find a way to fix that separately.
Author: Andrew Or <andrew@databricks.com>
Closes#6711 from andrewor14/fix-spark-shell and squashes the following commits:
bf62ecc [Andrew Or] Prevent NPE in spark-shell
Return null if catch EOFException, just like function "asKeyValueIterator" in this class
Author: Mingfei <mingfei.shi@intel.com>
Closes#6703 from shimingfei/returnNull and squashes the following commits:
205deec [Mingfei] return null if catch EOFException
Update build to use Java 7, and remove some comments and special-case support for Java 6.
Author: Sean Owen <sowen@cloudera.com>
Closes#6265 from srowen/SPARK-7733 and squashes the following commits:
59bda4e [Sean Owen] Update build to use Java 7, and remove some comments and special-case support for Java 6
…moved if dynamic allocation is enabled.
This is a work in progress. This patch ensures that an executor that has cached RDD blocks are not removed,
but makes no attempt to find another executor to remove. This is meant to get some feedback on the current
approach, and if it makes sense then I will look at choosing another executor to remove. No testing has been done either.
Author: Hari Shreedharan <hshreedharan@apache.org>
Closes#6508 from harishreedharan/dymanic-caching and squashes the following commits:
dddf1eb [Hari Shreedharan] Minor configuration description update.
10130e2 [Hari Shreedharan] Fix compile issue.
5417b53 [Hari Shreedharan] Add documentation for new config. Remove block from cachedBlocks when it is dropped.
875916a [Hari Shreedharan] Make some code more readable.
39940ca [Hari Shreedharan] Handle the case where the executor has not yet registered.
90ad711 [Hari Shreedharan] Remove unused imports and unused methods.
063985c [Hari Shreedharan] Send correct message instead of recursively calling same method.
ec2fd7e [Hari Shreedharan] Add file missed in last commit
5d10fad [Hari Shreedharan] Update cached blocks status using local info, rather than doing an RPC.
193af4c [Hari Shreedharan] WIP. Use local state rather than via RPC.
ae932ff [Hari Shreedharan] Fix config param name.
272969d [Hari Shreedharan] Fix seconds to millis bug.
5a1993f [Hari Shreedharan] Add timeout for cache executors. Ignore broadcast blocks while checking if there are cached blocks.
57fefc2 [Hari Shreedharan] [SPARK-7955][Core] Ensure executors with cached RDD blocks are not removed if dynamic allocation is enabled.
Author: Marcelo Vanzin <vanzin@cloudera.com>
Author: Jacek Lewandowski <lewandowski.jacek@gmail.com>
Closes#6560 from vanzin/SPARK-7169 and squashes the following commits:
737266f [Marcelo Vanzin] Feedback.
702d5a3 [Marcelo Vanzin] Scalastyle.
ce66e7e [Marcelo Vanzin] Remove metrics config handling from SparkConf.
439938a [Jacek Lewandowski] SPARK-7169: Metrics can be additionally configured from Spark configuration
Though totalStages = allStages - skippedStages is understandable. But consider the problem [SPARK-6973], I think totalStages = allStages is more reasonable. Like "2/1 (2 failed) (1 skipped)", this item also shows the skipped num, it also will be understandable.
Author: Xu Tingjun <xutingjun@huawei.com>
Author: Xutingjun <xutingjun@huawei.com>
Author: meiyoula <1039320815@qq.com>
Closes#5550 from XuTingjun/allJobsPage and squashes the following commits:
a742541 [Xu Tingjun] delete the loop
40ce94b [Xutingjun] remove stage id from completed set if it retries again
6459238 [meiyoula] delete space
9e23c71 [Xu Tingjun] recover numSkippedStages
b987ea7 [Xutingjun] delete skkiped stages from completed set
47525c6 [Xu Tingjun] modify total stages/tasks on the allJobsPage
This patch propose to lazy start the scheduler for dynamic allocation to avoid fast ramp down executor numbers is load is less.
This implementation will:
1. immediately start the scheduler is `numExecutorsTarget` is 0, this is the expected behavior.
2. if `numExecutorsTarget` is not zero, start the scheduler until the number is satisfied, if the load is less, this initial started executors will last for at least 60 seconds, user will have a window to submit a job, no need to revamp the executors.
3. if `numExecutorsTarget` is not satisfied until the timeout, this means resource is not enough, the scheduler will start until this timeout, will not wait infinitely.
Please help to review, thanks a lot.
Author: jerryshao <saisai.shao@intel.com>
Closes#6430 from jerryshao/SPARK-7699 and squashes the following commits:
02cac8e [jerryshao] Address the comments
7242450 [jerryshao] Remove the useless import
ecc0b00 [jerryshao] Address the comments
6f75f00 [jerryshao] Style changes
8b8decc [jerryshao] change the test name
fb822ca [jerryshao] Change the solution according to comments
1cc74e5 [jerryshao] Lazy start the scheduler for dynamic allocation
Author: Xutingjun <xutingjun@huawei.com>
Author: xutingjun <xutingjun@huawei.com>
Closes#6643 from XuTingjun/SPARK-8099 and squashes the following commits:
80b18cd [Xutingjun] change to STANDALONE | YARN
ce33148 [Xutingjun] set executor cores into system
e51cc9e [Xutingjun] set executor cores into system
0600861 [xutingjun] set executor cores into system
Reorganize code so that the launcher library handles most of the work
of printing usage messages, instead of having an awkward protocol between
the library and the scripts for that.
This mostly applies to SparkSubmit, since the launcher lib does not do
command line parsing for classes invoked in other ways, and thus cannot
handle failures for those. Most scripts end up going through SparkSubmit,
though, so it all works.
The change adds a new, internal command line switch, "--usage-error",
which prints the usage message and exits with a non-zero status. Scripts
can override the command printed in the usage message by setting an
environment variable - this avoids having to grep the output of
SparkSubmit to remove references to the "spark-submit" script.
The only sub-optimal part of the change is the special handling for the
spark-sql usage, which is now done in SparkSubmitArguments.
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#5841 from vanzin/SPARK-6324 and squashes the following commits:
2821481 [Marcelo Vanzin] Merge branch 'master' into SPARK-6324
bf139b5 [Marcelo Vanzin] Filter output of Spark SQL CLI help.
c6609bf [Marcelo Vanzin] Fix exit code never being used when printing usage messages.
6bc1b41 [Marcelo Vanzin] [SPARK-6324] [core] Centralize handling of script usage messages.
Completely trivial but I noticed this wrinkle in a log message today; `$sender` doesn't refer to anything and isn't interpolated here.
Author: Sean Owen <sowen@cloudera.com>
Closes#6650 from srowen/Interpolation and squashes the following commits:
518687a [Sean Owen] Actually interpolate log string
7edb866 [Sean Owen] Trivial: remove unused interpolation var in log message
The log page should only show desired length of bytes. Currently it shows bytes from the startIndex to the end of the file. The "Next" button on the page is always disabled.
Author: Carson Wang <carson.wang@intel.com>
Closes#6640 from carsonwang/logpage and squashes the following commits:
58cb3fd [Carson Wang] Show correct length of bytes on log page
In order to reduce the overhead of codegen, this PR switch to use Janino to compile SQL expressions into bytecode.
After this, the time used to compile a SQL expression is decreased from 100ms to 5ms, which is necessary to turn on codegen for general workload, also tests.
cc rxin
Author: Davies Liu <davies@databricks.com>
Closes#6479 from davies/janino and squashes the following commits:
cc689f5 [Davies Liu] remove globalLock
262d848 [Davies Liu] Merge branch 'master' of github.com:apache/spark into janino
eec3a33 [Davies Liu] address comments from Josh
f37c8c3 [Davies Liu] fix DecimalType and cast to String
202298b [Davies Liu] Merge branch 'master' of github.com:apache/spark into janino
a21e968 [Davies Liu] fix style
0ed3dc6 [Davies Liu] Merge branch 'master' of github.com:apache/spark into janino
551a851 [Davies Liu] fix tests
c3bdffa [Davies Liu] remove print
6089ce5 [Davies Liu] change logging level
7e46ac3 [Davies Liu] fix style
d8f0f6c [Davies Liu] Merge branch 'master' of github.com:apache/spark into janino
da4926a [Davies Liu] fix tests
03660f3 [Davies Liu] WIP: use Janino to compile Java source
f2629cd [Davies Liu] Merge branch 'master' of github.com:apache/spark into janino
f7d66cf [Davies Liu] use template based string for codegen
If maxTaskFailures is 1, the task set is aborted after 1 task failure. Other documentation and the code supports this reading, I think it's just this comment that was off. It's easy to make this mistake — can you please double-check if I'm correct? Thanks!
Author: Daniel Darabos <darabos.daniel@gmail.com>
Closes#6621 from darabos/patch-2 and squashes the following commits:
dfebdec [Daniel Darabos] Fix comment.
Author: Ryan Williams <ryan.blake.williams@gmail.com>
Closes#6624 from ryan-williams/execs and squashes the following commits:
b6f71d4 [Ryan Williams] don't attempt to lower number of executors by 0
Minor error in the monitoring docs. Also made indentation changes in `ApiRootResource`
Author: Hari Shreedharan <hshreedharan@apache.org>
Closes#6628 from harishreedharan/eventlog-formatting and squashes the following commits:
a12553d [Hari Shreedharan] Javadoc updates.
ca399b6 [Hari Shreedharan] [HOTFIX] History Server API docs error fix.
Replaced `fs.listFiles` with Hadoop-1 friendly `fs.listStatus` method.
Author: Hari Shreedharan <hshreedharan@apache.org>
Closes#6619 from harishreedharan/evetlog-hadoop-1-fix and squashes the following commits:
6192078 [Hari Shreedharan] [HOTFIX] Fix Hadoop-1 build caused by #5972.
The flaky tests in ExternalShuffleServiceSuite and SparkListenerWithClusterSuite will fail if there are not enough executors up before running the jobs.
This PR adds `JobProgressListener.waitUntilExecutorsUp`. The tests for the cluster mode can use it to wait until the expected executors are up.
Author: zsxwing <zsxwing@gmail.com>
Closes#6546 from zsxwing/SPARK-7989 and squashes the following commits:
5560e09 [zsxwing] Fix a typo
3b69840 [zsxwing] Fix flaky tests in ExternalShuffleServiceSuite and SparkListenerWithClusterSuite
Some places forget to call `assert` to check the return value of `AsynchronousListenerBus.waitUntilEmpty`. Instead of adding `assert` in these places, I think it's better to make `AsynchronousListenerBus.waitUntilEmpty` throw `TimeoutException`.
Author: zsxwing <zsxwing@gmail.com>
Closes#6550 from zsxwing/SPARK-8001 and squashes the following commits:
607674a [zsxwing] Make AsynchronousListenerBus.waitUntilEmpty throw TimeoutException if timeout
Author: Timothy Chen <tnachen@gmail.com>
Closes#6615 from tnachen/mesos_driver_path and squashes the following commits:
4f47b7c [Timothy Chen] Use the correct base path in mesos driver page.
...m History Server
This PR adds a new API that allows the user to download event logs for an application as a zip file. APIs have been added to download all logs for a given application or just for a specific attempt.
This also add an additional method to the ApplicationHistoryProvider to get the raw files, zipped.
Author: Hari Shreedharan <hshreedharan@apache.org>
Closes#5792 from harishreedharan/eventlog-download and squashes the following commits:
221cc26 [Hari Shreedharan] Update docs with new API information.
a131be6 [Hari Shreedharan] Fix style issues.
5528bd8 [Hari Shreedharan] Merge branch 'master' into eventlog-download
6e8156e [Hari Shreedharan] Simplify tests, use Guava stream copy methods.
d8ddede [Hari Shreedharan] Remove unnecessary case in EventLogDownloadResource.
ffffb53 [Hari Shreedharan] Changed interface to use zip stream. Added more tests.
1100b40 [Hari Shreedharan] Ensure that `Path` does not appear in interfaces, by rafactoring interfaces.
5a5f3e2 [Hari Shreedharan] Fix test ordering issue.
0b66948 [Hari Shreedharan] Minor formatting/import fixes.
4fc518c [Hari Shreedharan] Fix rat failures.
a48b91f [Hari Shreedharan] Refactor to make attemptId optional in the API. Also added tests.
0fc1424 [Hari Shreedharan] File download now works for individual attempts and the entire application.
350d7e8 [Hari Shreedharan] Merge remote-tracking branch 'asf/master' into eventlog-download
fd6ab00 [Hari Shreedharan] Fix style issues
32b7662 [Hari Shreedharan] Use UIRoot directly in ApiRootResource. Also, use `Response` class to set headers.
7b362b2 [Hari Shreedharan] Almost working.
3d18ebc [Hari Shreedharan] [WIP] Try getting the event log download to work.
Author: Patrick Wendell <patrick@databricks.com>
Closes#6328 from pwendell/spark-1.5-update and squashes the following commits:
2f42d02 [Patrick Wendell] A few more excludes
4bebcf0 [Patrick Wendell] Update to RC4
61aaf46 [Patrick Wendell] Using new release candidate
55f1610 [Patrick Wendell] Another exclude
04b4f04 [Patrick Wendell] More issues with transient 1.4 changes
36f549b [Patrick Wendell] [SPARK-7801] [BUILD] Updating versions to SPARK 1.5.0
This prevents the spark.jars from being cleared while using `--packages` or `--jars`
cc pwendell davies brkyvz
Author: Shivaram Venkataraman <shivaram@cs.berkeley.edu>
Closes#6568 from shivaram/SPARK-8028 and squashes the following commits:
3a9cf1f [Shivaram Venkataraman] Use addJar instead of setJars in SparkR This prevents the spark.jars from being cleared
Author: Sun Rui <rui.sun@intel.com>
Closes#6183 from sun-rui/SPARK-7227 and squashes the following commits:
dd6f5b3 [Sun Rui] Rename readEnv() back to readMap(). Add alias na.omit() for dropna().
41cf725 [Sun Rui] [SPARK-7227][SPARKR] Support fillna / dropna in R DataFrame.
Author: Reynold Xin <rxin@databricks.com>
Closes#6533 from rxin/whitespace-2 and squashes the following commits:
038314c [Reynold Xin] [SPARK-3850] Trim trailing spaces for core.
Author: Reynold Xin <rxin@databricks.com>
Closes#6528 from rxin/style-finalizer and squashes the following commits:
a2211ca [Reynold Xin] [SPARK-7976] Enable NoFinalizeChecker.
Spark's `ExternalSorter` writes shuffle output files during sort-based shuffle. Sort-shuffle contains a configuration, `spark.shuffle.sort.bypassMergeThreshold`, which causes ExternalSorter to skip sorting and merging and simply write separate files per partition, which are then concatenated together to form the final map output file.
The code paths used during this bypass are almost completely separate from ExternalSorter's other code paths, so refactoring them into a separate file can significantly simplify the code.
In addition to re-arranging code, this patch deletes a bunch of dead code. The main entry point into ExternalSorter is `insertAll()` and in SPARK-4479 / #3422 this method was modified to completely bypass in-memory buffering of records when `bypassMergeSort` takes effect. As a result, some of the spilling and merging code paths will no longer be called when `bypassMergeSort` is used, so we should be able to safely remove that code.
There's an open JIRA ([SPARK-6026](https://issues.apache.org/jira/browse/SPARK-6026)) for removing the `bypassMergeThreshold` parameter and code paths; I have not done that here, but the changes in this patch will make removing that parameter significantly easier if we ever decide to do that.
This patch also makes several improvements to shuffle-related tests and adds more defensive checks to certain shuffle classes:
- DiskBlockObjectWriter now throws an exception if `fileSegment()` is called before `commitAndClose()` has been called.
- DiskBlockObjectWriter's close methods are now idempotent, so calling any of the close methods twice in a row will no longer result in incorrect shuffle write metrics changes. Calling `revertPartialWritesAndClose()` on a closed DiskBlockObjectWriter now has no effect (before, it might mess up the metrics).
- The end-to-end shuffle record count metrics tests have been moved from InputOutputMetricsSuite to ShuffleSuite. This means that these tests will now be run against all shuffle implementations rather than just the default shuffle configuration.
- The end-to-end metrics tests now include a test of a job which performs aggregation in the shuffle.
- Our tests now check that `shuffleBytesWritten == totalShuffleBytesRead`.
- FileSegment now throws IllegalArgumentException if it is constructed with a negative length or offset.
Author: Josh Rosen <joshrosen@databricks.com>
Closes#6397 from JoshRosen/external-sorter-bypass-cleanup and squashes the following commits:
bf3f3f6 [Josh Rosen] Merge remote-tracking branch 'origin/master' into external-sorter-bypass-cleanup
8b216c4 [Josh Rosen] Guard against negative offsets and lengths in FileSegment
03f35a4 [Josh Rosen] Minor fix to cleanup logic.
b5cc35b [Josh Rosen] Move shuffle metrics tests to ShuffleSuite.
8b8fb9e [Josh Rosen] Add more tests + defensive programming to DiskBlockObjectWriter.
16564eb [Josh Rosen] Guard against calling fileSegment() before commitAndClose() has been called.
96811b4 [Josh Rosen] Remove confusing taskMetrics.shuffleWriteMetrics() optional call
8522b6a [Josh Rosen] Do not perform a map-side sort unless we're also doing map-side aggregation
08e40f3 [Josh Rosen] Remove excessively clever (and wrong) implementation of newBuffer()
d7f9938 [Josh Rosen] Add missing overrides; fix compilation
71d76ff [Josh Rosen] Update Javadoc
bf0d98f [Josh Rosen] Add comment to clarify confusing factory code
5197f73 [Josh Rosen] Add missing private[this]
30ef2c8 [Josh Rosen] Convert BypassMergeSortShuffleWriter to Java
bc1a820 [Josh Rosen] Fix bug when aggregator is used but map-side combine is disabled
0d3dcc0 [Josh Rosen] Remove unnecessary overloaded methods
25b964f [Josh Rosen] Rename SortShuffleSorter to SortShuffleFileWriter
0d9848c [Josh Rosen] Make it more clear that curWriteMetrics is now only used for spill metrics
7af7aea [Josh Rosen] Combine spill() and spillToMergeableFile()
6320112 [Josh Rosen] Add missing negation in deletion success check.
d267e0d [Josh Rosen] Fix style issue
7f15f7b [Josh Rosen] Back out extra cleanup-handling code, since this is already covered in stop()
25aa3bd [Josh Rosen] Make sure to delete outputFile after errors.
931ca68 [Josh Rosen] Refactor tests.
6a35716 [Josh Rosen] Refactor logic for deciding when to bypass
4b03539 [Josh Rosen] Move conf prior to first use
1265b25 [Josh Rosen] Fix some style errors and comments.
02355ef [Josh Rosen] More simplification
d4cb536 [Josh Rosen] Delete more unused code
bb96678 [Josh Rosen] Add missing interface file
b6cc1eb [Josh Rosen] Realize that bypass never buffers; proceed to delete tons of code
6185ee2 [Josh Rosen] WIP towards moving bypass code into own file.
8d0678c [Josh Rosen] Move diskBytesSpilled getter next to variable
19bccd6 [Josh Rosen] Remove duplicated buffer creation code.
18959bb [Josh Rosen] Move comparator methods closer together.
Author: zhichao.li <zhichao.li@intel.com>
Closes#6317 from zhichao-li/workers and squashes the following commits:
d68bf11 [zhichao.li] change prefix
99b6768 [zhichao.li] remove extra space and add 'Alive' prefix
1e8eb06 [zhichao.li] only showing alive workers
Only parse standalone master url when master url starts with spark://
Author: Timothy Chen <tnachen@gmail.com>
Closes#6517 from tnachen/fix_mesos_client and squashes the following commits:
61a1198 [Timothy Chen] Fix master url parsing in rest submission client.
cc JoshRosen
Thanks for noticing this!
Author: Burak Yavuz <brkyvz@gmail.com>
Closes#6509 from brkyvz/sample-perf-reg and squashes the following commits:
497465d [Burak Yavuz] addressed code review
293f95f [Burak Yavuz] [SPARK-7957] Preserve partitioning when using randomSplit
Author: Holden Karau <holden@pigscanfly.ca>
Closes#6464 from holdenk/SPARK-7910-expose-partitioner-information-in-javardd and squashes the following commits:
de1e644 [Holden Karau] Fix the test to get the partitioner
bdb31cc [Holden Karau] Add Mima exclude for the new method
347ef4c [Holden Karau] Add a quick little test for the partitioner JavaAPI
f49dca9 [Holden Karau] Add partitoner information to JavaRDDLike and fix some whitespace
* As spark now supports long running service by updating tokens for namenode, but only accept parameters passed with "--k=v" format which is not very convinient. This patch add spark.* configs in properties file and system property.
* --principal and --keytabl options are passed to client but when we started thrift server or spark-shell these two are also passed into the Main class (org.apache.spark.sql.hive.thriftserver.HiveThriftServer2 and org.apache.spark.repl.Main).
In these two main class, arguments passed in will be processed with some 3rd libraries, which will lead to some error: "Invalid option: --principal" or "Unrecgnised option: --principal".
We should pass these command args in different forms, say system properties.
Author: WangTaoTheTonic <wangtao111@huawei.com>
Closes#6051 from WangTaoTheTonic/SPARK-7524 and squashes the following commits:
e65699a [WangTaoTheTonic] change logic to loadEnvironments
ebd9ea0 [WangTaoTheTonic] merge master
ecfe43a [WangTaoTheTonic] pass keytab and principal seperately in different mode
33a7f40 [WangTaoTheTonic] expand the use of the current configs
08bb4e8 [WangTaoTheTonic] fix wrong cite
73afa64 [WangTaoTheTonic] add configs for keytab and principal, move originals to internal
SimpleDateFormat is not thread-safe. This PR creates new `SimpleDateFormat` for each `SimpleDateParam` instance.
Author: zsxwing <zsxwing@gmail.com>
Closes#6406 from zsxwing/SPARK-7863 and squashes the following commits:
aeed4c1 [zsxwing] Rewrite SimpleDateParam
8cdd986 [zsxwing] Inline formats
9680a15 [zsxwing] Create SimpleDateFormat for each SimpleDateParam instance because it's not thread-safe
Shutdown hook for temp directories had priority 100 while SparkContext was 50. So the local root directory was deleted before SparkContext was shutdown. This leads to scary errors on running jobs, at the time of shutdown. This is especially a problem when running streaming examples, where Ctrl-C is the only way to shutdown.
The fix in this PR is to make the temp directory shutdown priority lower than SparkContext, so that the temp dirs are the last thing to get deleted, after the SparkContext has been shut down. Also, the DiskBlockManager shutdown priority is change from default 100 to temp_dir_prio + 1, so that it gets invoked just before all temp dirs are cleared.
Author: Tathagata Das <tathagata.das1565@gmail.com>
Closes#6482 from tdas/SPARK-7930 and squashes the following commits:
d7cbeb5 [Tathagata Das] Removed unnecessary line
1514d0b [Tathagata Das] Fixed shutdown hook priorities
The existing code rounds down to the nearest percent when computing the proportion
of a task's time that was spent on each phase of execution, and then computes
the scheduler delay proportion as 100 - sum(all other proportions). As a result,
a few extra percent can end up in the scheduler delay. This commit eliminates
the rounding so that the time visualizations correspond properly to the real times.
sarutak If you could take a look at this, that would be great! Not sure if there's a good
reason to round here that I missed.
cc shivaram
Author: Kay Ousterhout <kayousterhout@gmail.com>
Closes#6484 from kayousterhout/SPARK-7932 and squashes the following commits:
1723cc4 [Kay Ousterhout] [SPARK-7932] Fix misleading scheduler delay visualization
So we can enable a whitespace enforcement rule in the style checker to save code review time.
Author: Reynold Xin <rxin@databricks.com>
Closes#6473 from rxin/whitespace-core and squashes the following commits:
058195d [Reynold Xin] Fixed tests.
fce11e9 [Reynold Xin] [SPARK-7927] whitespace fixes for core.
There are too many extra call method `getCacheLocs` for `DAGScheduler`, which includes Akka communication.
To improve `DAGScheduler` performance, suppress extra calling the method.
In my application with over 1200 stages, the execution time became 3.8 min from 8.5 min with my patch.
Author: Takuya UESHIN <ueshin@happy-camper.st>
Closes#6352 from ueshin/issues/SPARK-7826 and squashes the following commits:
3d4d036 [Takuya UESHIN] Modify a test and the documentation.
10b1b22 [Takuya UESHIN] Simplify the unit test.
d858b59 [Takuya UESHIN] Move the storageLevel check inside the if (!cacheLocs.contains(rdd.id)) block.
6f3125c [Takuya UESHIN] Fix scalastyle.
b9c835c [Takuya UESHIN] Put the condition that checks if the RDD has uncached partition or not into variable for readability.
f87f2ec [Takuya UESHIN] Get cached locations from block manager only if the storage level of the RDD is not StorageLevel.NONE.
8248386 [Takuya UESHIN] Revert "Suppress extra calling getCacheLocs."
a4d944a [Takuya UESHIN] Add an unit test.
9a80fad [Takuya UESHIN] Suppress extra calling getCacheLocs.
See comments on https://github.com/apache/spark/pull/3913
Author: Reynold Xin <rxin@databricks.com>
Closes#6471 from rxin/sizeestimator and squashes the following commits:
c057095 [Reynold Xin] Fixed import.
2da478b [Reynold Xin] Remove SizeEstimator from o.a.spark package.
Author: Sandy Ryza <sandy@cloudera.com>
Closes#6440 from sryza/sandy-spark-7896 and squashes the following commits:
49d8a0d [Sandy Ryza] Fix bug introduced when reading over record boundaries
6006856 [Sandy Ryza] Fix overflow issues
006b4b2 [Sandy Ryza] Fix scalastyle by removing non ascii characters
8b000ca [Sandy Ryza] Add ascii art to describe layout of data in metaBuffer
f2053c0 [Sandy Ryza] Fix negative overflow issue
0368c78 [Sandy Ryza] Initialize size as 0
a5a4820 [Sandy Ryza] Use explicit types for all numbers in ChainedBuffer
b7e0213 [Sandy Ryza] SPARK-7896. Allow ChainedBuffer to store more than 2 GB
This is a somewhat obscure bug, but I think that it will seriously impact KryoSerializer users who use custom registrators which disabled auto-reset. When auto-reset is disabled, then this breaks things in some of our shuffle paths which actually end up creating multiple OutputStreams from the same shared SerializerInstance (which is unsafe).
This was introduced by a patch (SPARK-3386) which enables serializer re-use in some of the shuffle paths, since constructing new serializer instances is actually pretty costly for KryoSerializer. We had already fixed another corner-case (SPARK-7766) bug related to this, but missed this one.
I think that the root problem here is that KryoSerializerInstance can be used in a way which is unsafe even within a single thread, e.g. by creating multiple open OutputStreams from the same instance or by interleaving deserialize and deserializeStream calls. I considered a smaller patch which adds assertions to guard against this type of "misuse" but abandoned that approach after I realized how convoluted the Scaladoc became.
This patch fixes this bug by making it legal to create multiple streams from the same KryoSerializerInstance. Internally, KryoSerializerInstance now implements a `borrowKryo()` / `releaseKryo()` API that's backed by a "pool" of capacity 1. Each call to a KryoSerializerInstance method will borrow the Kryo, do its work, then release the serializer instance back to the pool. If the pool is empty and we need an instance, it will allocate a new Kryo on-demand. This makes it safe for multiple OutputStreams to be opened from the same serializer. If we try to release a Kryo back to the pool but the pool already contains a Kryo, then we'll just discard the new Kryo. I don't think there's a clear benefit to having a larger pool since our usages tend to fall into two cases, a) where we only create a single OutputStream and b) where we create a huge number of OutputStreams with the same lifecycle, then destroy the KryoSerializerInstance (this is what's happening in the bypassMergeSort code path that my regression test hits).
Author: Josh Rosen <joshrosen@databricks.com>
Closes#6415 from JoshRosen/SPARK-7873 and squashes the following commits:
00b402e [Josh Rosen] Initialize eagerly to fix a failing test
ba55d20 [Josh Rosen] Add explanatory comments
3f1da96 [Josh Rosen] Guard against duplicate close()
ab457ca [Josh Rosen] Sketch a loan/release based solution.
9816e8f [Josh Rosen] Add a failing test showing how deserialize() and deserializeStream() can interfere.
7350886 [Josh Rosen] Add failing regression test for SPARK-7873
This issue is related to #6419 .
Now AllJobPage doesn't have a "kill link" but I think fix the issue mentioned in #6419 just in case to avoid accidents in the future.
So, it's minor issue for now and I don't file this issue in JIRA.
Author: Kousuke Saruta <sarutak@oss.nttdata.co.jp>
Closes#6432 from sarutak/remove-ambiguity-of-link and squashes the following commits:
cd1a503 [Kousuke Saruta] Fixed ambiguity link issue in AllJobPage
The previous name was confusing, because each stage can be associated with
many jobs, and jobId is just the ID of the first job that was associated
with the Stage. This commit also renames some of the method parameters in
DAGScheduler.scala to clarify when the jobId refers to the first job ID
associated with the stage (as opposed to the jobId associated with a job
that's currently being scheduled).
cc markhamstra JoshRosen (hopefully this will help prevent future bugs like SPARK-6880)
Author: Kay Ousterhout <kayousterhout@gmail.com>
Closes#6418 from kayousterhout/SPARK-7878 and squashes the following commits:
b71a9b8 [Kay Ousterhout] [SPARK-7878] Rename Stage.jobId to firstJobId
**Reproduction.** Run a long-running job, go to the job page, expand the DAG visualization, and click into a stage. Your stage is now killed. Why? This is because the visualization code just reaches into the stage table and grabs the first link it finds. In our case, this first link happens to be the kill link instead of the one to the stage page.
**Fix.** Use proper CSS selectors to avoid ambiguity.
This is an alternative to #6407. Thanks carsonwang for catching this.
Author: Andrew Or <andrew@databricks.com>
Closes#6419 from andrewor14/fix-ui-viz-kill and squashes the following commits:
25203bd [Andrew Or] Do not kill innocent stages
Author: zsxwing <zsxwing@gmail.com>
Closes#6333 from zsxwing/remove-actor-system-usage and squashes the following commits:
f125aa6 [zsxwing] Fix YarnAllocatorSuite
ceadcf6 [zsxwing] Change the "port" parameter type of "AkkaUtils.address" to "int"; update ApplicationMaster and YarnAllocator to get the driverUrl from RpcEnv
3239380 [zsxwing] Remove some places in core that calling SparkEnv.actorSystem
My experiments with scheduling very short tasks in standalone cluster mode indicated that a significant amount of time was being spent in scheduling the tasks (>500ms for 256 tasks). I found that most of the time was being spent in creating a new instance of serializer for each task. Changing this to just one serializer brought down the scheduling time to 8ms.
Author: Akshat Aranya <aaranya@quantcast.com>
Closes#6323 from coolfrood/master and squashes the following commits:
12d8c9e [Akshat Aranya] Reduce visibility of serializer
bd4a5dd [Akshat Aranya] Style fix
0b8ca93 [Akshat Aranya] Incorporate review comments
fe530cd [Akshat Aranya] Speed up task scheduling in standalone mode by reusing serializer instead of creating a new one for each task.
The previous PR for SPARK-7224 (#5790) broke JDK 6, because it used java.nio.Path, which was in jdk 7, and not in 6. This PR uses Guava's `Files` to handle directory creation, and etc...
The description from the previous PR:
> This patch contains an `IvyTestUtils` file, which dynamically generates jars and pom files to test the `--packages` feature without having to rely on the internet, and Maven Central.
cc pwendell
I also rand the flaky test about 20 times locally, it didn't fail a single time, but I think it may fail like once every 100 builds? I still haven't figured the cause yet, but the test before it, `--jars` was also failing after we turned off the `--packages` test in `SparkSubmitSuite`. It may be related to the launch of SparkSubmit.
Author: Burak Yavuz <brkyvz@gmail.com>
Closes#5892 from brkyvz/maven-utils and squashes the following commits:
e9b1903 [Burak Yavuz] fix merge conflict
68214e0 [Burak Yavuz] remove ignore for test(neglect spark dependencies)
e632381 [Burak Yavuz] fix ignore
9ef1408 [Burak Yavuz] re-enable --packages test
22eea62 [Burak Yavuz] Merge branch 'master' of github.com:apache/spark into maven-utils
05cd0de [Burak Yavuz] added mock repository generator
The default add time of 5s is still too slow for small jobs. Also, the current default remove time of 10 minutes seem rather high. This patch lowers both and rephrases a few log messages.
Author: Andrew Or <andrew@databricks.com>
Closes#6301 from andrewor14/da-minor and squashes the following commits:
6d614a6 [Andrew Or] Lower log level
2811492 [Andrew Or] Log information when requests are canceled
5fcd3eb [Andrew Or] Fix tests
3320710 [Andrew Or] Lower timeouts + rephrase a few log messages
Author: Imran Rashid <irashid@cloudera.com>
Closes#6284 from squito/SPARK-7760 and squashes the following commits:
5e02d8a [Imran Rashid] style; increase timeout
9987399 [Imran Rashid] comment
8c7ed63 [Imran Rashid] add /json back into master & worker pages; add test
https://issues.apache.org/jira/browse/SPARK-7758
When initializing `executionHive`, we only masks
`javax.jdo.option.ConnectionURL` to override metastore location. However,
other properties that relates to the actual Hive metastore data source are not
masked. For example, when using Spark SQL with a PostgreSQL backed Hive
metastore, `executionHive` actually tries to use settings read from
`hive-site.xml`, which talks about PostgreSQL, to connect to the temporary
Derby metastore, thus causes error.
To fix this, we need to mask all metastore data source properties.
Specifically, according to the code of [Hive `ObjectStore.getDataSourceProps()`
method] [1], all properties whose name mentions "jdo" and "datanucleus" must be
included.
[1]: https://github.com/apache/hive/blob/release-0.13.1/metastore/src/java/org/apache/hadoop/hive/metastore/ObjectStore.java#L288
Have tested using postgre sql as metastore, it worked fine.
Author: WangTaoTheTonic <wangtao111@huawei.com>
Closes#6314 from WangTaoTheTonic/SPARK-7758 and squashes the following commits:
ca7ae7c [WangTaoTheTonic] add comments
86caf2c [WangTaoTheTonic] delete unused import
e4f0feb [WangTaoTheTonic] block more data source related property
92a81fa [WangTaoTheTonic] fix style check
e3e683d [WangTaoTheTonic] override more configs to avoid failuer connecting to postgre sql
SPARK-3386 / #5606 modified the shuffle write path to re-use serializer instances across multiple calls to DiskBlockObjectWriter. It turns out that this introduced a very rare bug when using `KryoSerializer`: if auto-reset is disabled and reference-tracking is enabled, then we'll end up re-using the same serializer instance to write multiple output streams without calling `reset()` between write calls, which can lead to cases where objects in one file may contain references to objects that are in previous files, causing errors during deserialization.
This patch fixes this bug by calling `reset()` at the start of `serialize()` and `serializeStream()`. I also added a regression test which demonstrates that this problem only occurs when auto-reset is disabled and reference-tracking is enabled.
Author: Josh Rosen <joshrosen@databricks.com>
Closes#6293 from JoshRosen/kryo-instance-reuse-bug and squashes the following commits:
e19726d [Josh Rosen] Add fix for SPARK-7766.
71845e3 [Josh Rosen] Add failing regression test to trigger Kryo re-use bug
This PR adds the URLs to the driver logs to `SparkListenerApplicationStarted` event, which is later used by the `ExecutorsListener` to populate the URLs to the driver logs in its own state. This info is then used when the UI is rendered to display links to the logs.
Author: Hari Shreedharan <hshreedharan@apache.org>
Closes#6166 from harishreedharan/am-log-link and squashes the following commits:
943fc4f [Hari Shreedharan] Merge remote-tracking branch 'asf/master' into am-log-link
9e5c04b [Hari Shreedharan] Merge remote-tracking branch 'asf/master' into am-log-link
b3f9b9d [Hari Shreedharan] Updated comment based on feedback.
0840a95 [Hari Shreedharan] Move the result and sc.stop back to original location, minor import changes.
537a2f7 [Hari Shreedharan] Add test to ensure the log urls are populated and valid.
4033725 [Hari Shreedharan] Adding comments explaining how node reports are used to get the log urls.
6c5c285 [Hari Shreedharan] Import order.
346f4ea [Hari Shreedharan] Review feedback fixes.
629c1dc [Hari Shreedharan] Cleanup.
99fb1a3 [Hari Shreedharan] Send the log urls in App start event, to ensure that other listeners are not affected.
c0de336 [Hari Shreedharan] Ensure new unit test cleans up after itself.
50cdae3 [Hari Shreedharan] Added unit test, made the approach generic.
402e8e4 [Hari Shreedharan] Use `NodeReport` to get the URL for the logs. Also, make the environment variables generic so other cluster managers can use them as well.
1cf338f [Hari Shreedharan] [SPARK-7657][YARN] Add driver link in application UI, in cluster mode.
According to yhuai we spent 6-7 seconds cleaning closures in a partitioning job that takes 12 seconds. Since we provide these closures in Spark we know for sure they are serializable, so we can bypass the cleaning.
Author: Andrew Or <andrew@databricks.com>
Closes#6256 from andrewor14/sql-partition-speed-up and squashes the following commits:
a82b451 [Andrew Or] Fix style
10f7e3e [Andrew Or] Avoid getting call sites and cleaning closures
17e2943 [Andrew Or] Merge branch 'master' of github.com:apache/spark into sql-partition-speed-up
523f042 [Andrew Or] Skip unnecessary Utils.getCallSites too
f7fe143 [Andrew Or] Avoid unnecessary closure cleaning
Document current limitation of rdd.fold.
This does not resolve SPARK-6416 but just documents the issue.
CC JoshRosen
Author: Sean Owen <sowen@cloudera.com>
Closes#6231 from srowen/SPARK-6416 and squashes the following commits:
9fef39f [Sean Owen] Add comment to other languages; reword to highlight the difference from non-distributed collections and to not suggest it is a bug that is to be fixed
da40d84 [Sean Owen] Document current limitation of rdd.fold.
Two main changes:
Add two functions in ExternalBlockManager, which are putValues and getValues
because the implementation may not rely on the putBytes and getBytes
improve Tachyon integration.
Currently, when putting data into Tachyon, Spark first serialize all data in one partition into a ByteBuffer, and then write into Tachyon, this will uses much memory and increase GC overhead
when get data from Tachyon, getValues depends on getBytes, which also read all data into On heap byte arry, and result in much memory usage.
This PR changes the approach of the two functions, make them read / write data by stream to reduce memory usage.
In our testing, when data size is huge, this patch reduces about 30% GC time and 70% full GC time, and total execution time reduces about 10%
Author: Mingfei <mingfei.shi@intel.com>
Closes#5908 from shimingfei/Tachyon-integration-rebase and squashes the following commits:
033bc57 [Mingfei] modify accroding to comments
747c69a [Mingfei] modify according to comments - format changes
ce52c67 [Mingfei] put close() in a finally block
d2c60bb [Mingfei] modify according to comments, some code style change
4c11591 [Mingfei] modify according to comments split putIntoExternalBlockStore into two functions add default implementation for getValues and putValues
cc0a32e [Mingfei] Make getValues read data from Tachyon by stream Make putValues write data to Tachyon by stream
017593d [Mingfei] add getValues and putValues in ExternalBlockManager's Interface
…rther extension to non-json outputs too.
Author: Hari Shreedharan <hshreedharan@apache.org>
Closes#6273 from harishreedharan/json-to-api and squashes the following commits:
e14b73b [Hari Shreedharan] Rename `getJsonServlet` to `getServletHandler` i
42f8acb [Hari Shreedharan] Import order fixes.
2ef852f [Hari Shreedharan] [SPARK-7750][WebUI] Rename endpoints from `json` to `api` to allow further extension to non-json outputs too.
Currently, the background checkpointing thread fails silently if the checkpoint is not serializable. It is hard to debug and therefore its best to fail fast at `start()` when checkpointing is enabled and the checkpoint is not serializable.
Author: Tathagata Das <tathagata.das1565@gmail.com>
Closes#6292 from tdas/SPARK-7767 and squashes the following commits:
51304e6 [Tathagata Das] Addressed comments.
c35237b [Tathagata Das] Added test for checkpoint serialization in StreamingContext.start()
SPARK-7741 is the equivalent of SPARK-7237 in streaming. This is an alternative to #6268.
Author: Andrew Or <andrew@databricks.com>
Closes#6269 from andrewor14/clean-moar and squashes the following commits:
c51c9ab [Andrew Or] Add periods (trivial)
6c686ac [Andrew Or] Merge branch 'master' of github.com:apache/spark into clean-moar
79a435b [Andrew Or] Fix tests
d18c9f9 [Andrew Or] Merge branch 'master' of github.com:apache/spark into clean-moar
65ef07b [Andrew Or] Fix tests?
4b487a3 [Andrew Or] Add tests for closures passed to DStream operations
328139b [Andrew Or] Do not forget foreachRDD
5431f61 [Andrew Or] Clean streaming closures
72b7b73 [Andrew Or] Clean core closures
cc rxin, please take a quick look, I'm working on tests.
Author: Davies Liu <davies@databricks.com>
Closes#6238 from davies/readwrite and squashes the following commits:
c7200eb [Davies Liu] update tests
9cbf01b [Davies Liu] Merge branch 'master' of github.com:apache/spark into readwrite
f0c5a04 [Davies Liu] use sqlContext.read.load
5f68bc8 [Davies Liu] update tests
6437e9a [Davies Liu] Merge branch 'master' of github.com:apache/spark into readwrite
bcc6668 [Davies Liu] add reader amd writer API in Python
This PR is based on #6081, thanks adrian-wang.
Closes#6081
Author: Daoyuan Wang <daoyuan.wang@intel.com>
Author: Davies Liu <davies@databricks.com>
Closes#6230 from davies/range and squashes the following commits:
d3ce5fe [Davies Liu] add tests
789eda5 [Davies Liu] add range() in Python
4590208 [Davies Liu] Merge commit 'refs/pull/6081/head' of github.com:apache/spark into range
cbf5200 [Daoyuan Wang] let's add python support in a separate PR
f45e3b2 [Daoyuan Wang] remove redundant toLong
617da76 [Daoyuan Wang] fix safe marge for corner cases
867c417 [Daoyuan Wang] fix
13dbe84 [Daoyuan Wang] update
bd998ba [Daoyuan Wang] update comments
d3a0c1b [Daoyuan Wang] add range api()
This is similar to #5999, but for streaming. Roughly 200 lines are tests.
One thing to note here is that we already do some kind of scoping thing for call sites, so this patch adds the new RDD operation scoping logic in the same place. Also, this patch adds a `try finally` block to set the relevant variables in a safer way.
tdas zsxwing
------------------------
**Before**
<img src="https://cloud.githubusercontent.com/assets/2133137/7625996/d88211b8-f9b4-11e4-90b9-e11baa52d6d7.png" width="450px"/>
--------------------------
**After**
<img src="https://cloud.githubusercontent.com/assets/2133137/7625997/e0878f8c-f9b4-11e4-8df3-7dd611b13c87.png" width="650px"/>
Author: Andrew Or <andrew@databricks.com>
Closes#6034 from andrewor14/dag-viz-streaming and squashes the following commits:
932a64a [Andrew Or] Merge branch 'master' of github.com:apache/spark into dag-viz-streaming
e685df9 [Andrew Or] Rename createRDDWith
84d0656 [Andrew Or] Review feedback
697c086 [Andrew Or] Fix tests
53b9936 [Andrew Or] Set scopes for foreachRDD properly
1881802 [Andrew Or] Refactor DStream scope names again
af4ba8d [Andrew Or] Merge branch 'master' of github.com:apache/spark into dag-viz-streaming
fd07d22 [Andrew Or] Make MQTT lower case
f6de871 [Andrew Or] Merge branch 'master' of github.com:apache/spark into dag-viz-streaming
0ca1801 [Andrew Or] Remove a few unnecessary withScopes on aliases
fa4e5fb [Andrew Or] Pass in input stream name rather than defining it from within
1af0b0e [Andrew Or] Fix style
074c00b [Andrew Or] Review comments
d25a324 [Andrew Or] Merge branch 'master' of github.com:apache/spark into dag-viz-streaming
e4a93ac [Andrew Or] Fix tests?
25416dc [Andrew Or] Merge branch 'master' of github.com:apache/spark into dag-viz-streaming
9113183 [Andrew Or] Add tests for DStream scopes
b3806ab [Andrew Or] Fix test
bb80bbb [Andrew Or] Fix MIMA?
5c30360 [Andrew Or] Merge branch 'master' of github.com:apache/spark into dag-viz-streaming
5703939 [Andrew Or] Rename operations that create InputDStreams
7c4513d [Andrew Or] Group RDDs by DStream operations and batches
bf0ab6e [Andrew Or] Merge branch 'master' of github.com:apache/spark into dag-viz-streaming
05c2676 [Andrew Or] Wrap many more methods in withScope
c121047 [Andrew Or] Merge branch 'master' of github.com:apache/spark into dag-viz-streaming
65ef3e9 [Andrew Or] Fix NPE
a0d3263 [Andrew Or] Scope streaming operations instead of RDD operations
This PR revert #5404, change to pass the version of python in driver into JVM, check it in worker before deserializing closure, then it can works with different major version of Python.
Author: Davies Liu <davies@databricks.com>
Closes#6203 from davies/py_version and squashes the following commits:
b8fb76e [Davies Liu] fix test
6ce5096 [Davies Liu] use string for version
47c6278 [Davies Liu] check python version of worker with driver
This patch fixes two things:
**SPARK-7627.** Cached RDDs no longer light up on the job page. This is a simple fix.
**SPARK-7472.** Display skipped stages differently from normal stages.
The latter is a major UX issue. Because we link the job viz to the stage viz even for skipped stages, the user may inadvertently click into the stage page of a skipped stage, which is empty.
-------------------
<img src="https://cloud.githubusercontent.com/assets/2133137/7675241/de1a3da6-fcea-11e4-8101-88055cef78c5.png" width="300px" />
Author: Andrew Or <andrew@databricks.com>
Closes#6171 from andrewor14/dag-viz-skipped and squashes the following commits:
f261797 [Andrew Or] Merge branch 'master' of github.com:apache/spark into dag-viz-skipped
0eda358 [Andrew Or] Tweak skipped stage border color
c604150 [Andrew Or] Tweak grayscale colors
7010676 [Andrew Or] Merge branch 'master' of github.com:apache/spark into dag-viz-skipped
762b541 [Andrew Or] Use special prefix for stage clusters to avoid collisions
51c95b9 [Andrew Or] Merge branch 'master' of github.com:apache/spark into dag-viz-skipped
b928cd4 [Andrew Or] Fix potential leak + write tests for it
7c4c364 [Andrew Or] Show skipped stages differently
7cc34ce [Andrew Or] Merge branch 'master' of github.com:apache/spark into dag-viz-skipped
c121fa2 [Andrew Or] Fix cache color
Learnt a lesson from SPARK-7655: Spark should avoid to use `scala.concurrent.ExecutionContext.Implicits.global` because the user may submit blocking actions to `scala.concurrent.ExecutionContext.Implicits.global` and exhaust all threads in it. This could crash Spark. So Spark should always use its own thread pools for safety.
This PR removes all usages of `scala.concurrent.ExecutionContext.Implicits.global` and uses proper thread pools to replace them.
Author: zsxwing <zsxwing@gmail.com>
Closes#6223 from zsxwing/SPARK-7693 and squashes the following commits:
a33ff06 [zsxwing] Decrease the max thread number from 1024 to 128
cf4b3fc [zsxwing] Remove "import scala.concurrent.ExecutionContext.Implicits.global"
This patch wraps `SnappyOutputStream` to ensure that `close()` is idempotent and to guard against write-after-`close()` bugs. This is a workaround for https://github.com/xerial/snappy-java/issues/107, a bug where a non-idempotent `close()` method can lead to stream corruption. We can remove this workaround if we upgrade to a snappy-java version that contains my fix for this bug, but in the meantime this patch offers a backportable Spark fix.
Author: Josh Rosen <joshrosen@databricks.com>
Closes#6176 from JoshRosen/SPARK-7660-wrap-snappy and squashes the following commits:
8b77aae [Josh Rosen] Wrap SnappyOutputStream to fix SPARK-7660
We should not call `DirectTaskResult.value` when holding the `TaskSchedulerImpl` lock. It may cost dozens of seconds to deserialize a large object.
Author: zsxwing <zsxwing@gmail.com>
Closes#6195 from zsxwing/SPARK-7655 and squashes the following commits:
21f502e [zsxwing] Add more comments
e25fa88 [zsxwing] Add comments
15010b5 [zsxwing] Deserialize value should not hold the TaskSchedulerImpl lock
Because both `AkkaRpcEndpointRef.ask` and `BroadcastHashJoin` uses `scala.concurrent.ExecutionContext.Implicits.global`. However, because the tasks in `BroadcastHashJoin` are usually long-running tasks, which will occupy all threads in `global`. Then `ask` cannot get a chance to process the replies.
For `ask`, actually the tasks are very simple, so we can use `MoreExecutors.sameThreadExecutor()`. For `BroadcastHashJoin`, it's better to use `ThreadUtils.newDaemonCachedThreadPool`.
Author: zsxwing <zsxwing@gmail.com>
Closes#6200 from zsxwing/SPARK-7655-2 and squashes the following commits:
cfdc605 [zsxwing] Remove redundant imort and minor doc fix
cf83153 [zsxwing] Add "sameThread" and "newDaemonCachedThreadPool with maxThreadNumber" to ThreadUtils
08ad0ee [zsxwing] Remove 'scala.concurrent.ExecutionContext.Implicits.global' in 'ask' and 'BroadcastHashJoin'
In translating spark.kryoserializer.buffer.mb to spark.kryoserializer.buffer, use of toDouble will lead to "Fractional values not supported" error even when spark.kryoserializer.buffer.mb is an integer.
ilganeli, andrewor14
Author: Nishkam Ravi <nravi@cloudera.com>
Author: nishkamravi2 <nishkamravi@gmail.com>
Author: nravi <nravi@c1704.halxg.cloudera.com>
Closes#6198 from nishkamravi2/master_nravi and squashes the following commits:
171a53c [nishkamravi2] Update SparkConfSuite.scala
5261bf6 [Nishkam Ravi] Add a test for deprecated config spark.kryoserializer.buffer.mb
5190f79 [Nishkam Ravi] In translating from deprecated spark.kryoserializer.buffer.mb to spark.kryoserializer.buffer use int conversion since fractions are not permissible
059ce82 [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
eaa13b5 [nishkamravi2] Update Client.scala
981afd2 [Nishkam Ravi] Check for read permission before initiating copy
1b81383 [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
0f1abd0 [nishkamravi2] Update Utils.scala
474e3bf [nishkamravi2] Update DiskBlockManager.scala
97c383e [nishkamravi2] Update Utils.scala
8691e0c [Nishkam Ravi] Add a try/catch block around Utils.removeShutdownHook
2be1e76 [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
1c13b79 [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
bad4349 [nishkamravi2] Update Main.java
36a6f87 [Nishkam Ravi] Minor changes and bug fixes
b7f4ae7 [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
4a45d6a [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
458af39 [Nishkam Ravi] Locate the jar using getLocation, obviates the need to pass assembly path as an argument
d9658d6 [Nishkam Ravi] Changes for SPARK-6406
ccdc334 [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
3faa7a4 [Nishkam Ravi] Launcher library changes (SPARK-6406)
345206a [Nishkam Ravi] spark-class merge Merge branch 'master_nravi' of https://github.com/nishkamravi2/spark into master_nravi
ac58975 [Nishkam Ravi] spark-class changes
06bfeb0 [nishkamravi2] Update spark-class
35af990 [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
32c3ab3 [nishkamravi2] Update AbstractCommandBuilder.java
4bd4489 [nishkamravi2] Update AbstractCommandBuilder.java
746f35b [Nishkam Ravi] "hadoop" string in the assembly name should not be mandatory (everywhere else in spark we mandate spark-assembly*hadoop*.jar)
bfe96e0 [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
ee902fa [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
d453197 [nishkamravi2] Update NewHadoopRDD.scala
6f41a1d [nishkamravi2] Update NewHadoopRDD.scala
0ce2c32 [nishkamravi2] Update HadoopRDD.scala
f7e33c2 [Nishkam Ravi] Merge branch 'master_nravi' of https://github.com/nishkamravi2/spark into master_nravi
ba1eb8b [Nishkam Ravi] Try-catch block around the two occurrences of removeShutDownHook. Deletion of semi-redundant occurrences of expensive operation inShutDown.
71d0e17 [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
494d8c0 [nishkamravi2] Update DiskBlockManager.scala
3c5ddba [nishkamravi2] Update DiskBlockManager.scala
f0d12de [Nishkam Ravi] Workaround for IllegalStateException caused by recent changes to BlockManager.stop
79ea8b4 [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
b446edc [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
5c9a4cb [nishkamravi2] Update TaskSetManagerSuite.scala
535295a [nishkamravi2] Update TaskSetManager.scala
3e1b616 [Nishkam Ravi] Modify test for maxResultSize
9f6583e [Nishkam Ravi] Changes to maxResultSize code (improve error message and add condition to check if maxResultSize > 0)
5f8f9ed [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
636a9ff [nishkamravi2] Update YarnAllocator.scala
8f76c8b [Nishkam Ravi] Doc change for yarn memory overhead
35daa64 [Nishkam Ravi] Slight change in the doc for yarn memory overhead
5ac2ec1 [Nishkam Ravi] Remove out
dac1047 [Nishkam Ravi] Additional documentation for yarn memory overhead issue
42c2c3d [Nishkam Ravi] Additional changes for yarn memory overhead issue
362da5e [Nishkam Ravi] Additional changes for yarn memory overhead
c726bd9 [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
f00fa31 [Nishkam Ravi] Improving logging for AM memoryOverhead
1cf2d1e [nishkamravi2] Update YarnAllocator.scala
ebcde10 [Nishkam Ravi] Modify default YARN memory_overhead-- from an additive constant to a multiplier (redone to resolve merge conflicts)
2e69f11 [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark into master_nravi
efd688a [Nishkam Ravi] Merge branch 'master' of https://github.com/apache/spark
2b630f9 [nravi] Accept memory input as "30g", "512M" instead of an int value, to be consistent with rest of Spark
3bf8fad [nravi] Merge branch 'master' of https://github.com/apache/spark
5423a03 [nravi] Merge branch 'master' of https://github.com/apache/spark
eb663ca [nravi] Merge branch 'master' of https://github.com/apache/spark
df2aeb1 [nravi] Improved fix for ConcurrentModificationIssue (Spark-1097, Hadoop-10456)
6b840f0 [nravi] Undo the fix for SPARK-1758 (the problem is fixed)
5108700 [nravi] Fix in Spark for the Concurrent thread modification issue (SPARK-1097, HADOOP-10456)
681b36f [nravi] Fix for SPARK-1758: failing test org.apache.spark.JavaAPISuite.wholeTextFiles
This fixes a bug where an executor that exits can cause the driver's OutputCommitCoordinator to stop. To fix this, we use an `isDriver` flag and check it in `stop()`.
See https://issues.apache.org/jira/browse/SPARK-7563 for more details.
Author: Josh Rosen <joshrosen@databricks.com>
Closes#6197 from JoshRosen/SPARK-7563 and squashes the following commits:
04b2cc5 [Josh Rosen] [SPARK-7563] OutputCommitCoordinator.stop() should only be executed on the driver
cc pwendell sarutak
This commit cleans up some unnecessary code, eliminates the feature where when you mouse-over a box in the timeline, the corresponding task is highlighted in the table (because that feature is only useful in the rare case when you have a very small number of tasks, in which case it's easy to figure out the mapping anyway), and fixes a bug where nothing shows up if you try to visualize a stage with only 1 task.
Author: Kay Ousterhout <kayousterhout@gmail.com>
Closes#6202 from kayousterhout/SPARK-7676 and squashes the following commits:
dfd29d4 [Kay Ousterhout] [SPARK-7676] Bug fix and cleanup of stage timeline view
This PR builds on #2342 by adding a timeline view for the Stage page,
showing how tasks spend their time.
With this timeline, we can understand following things of a Stage.
* When/where each task ran
* Total duration of each task
* Proportion of the time each task spends
Also, this timeline view can scrollable and zoomable.
Author: Kousuke Saruta <sarutak@oss.nttdata.co.jp>
Closes#5843 from sarutak/stage-page-timeline and squashes the following commits:
4ba9604 [Kousuke Saruta] Fixed the order of legends
16bb552 [Kousuke Saruta] Removed border of legend area
2e5d605 [Kousuke Saruta] Modified warning message
16cb2e6 [Kousuke Saruta] Merge branch 'master' of https://github.com/apache/spark into stage-page-timeline
7ae328f [Kousuke Saruta] Modified code style
d5f794a [Kousuke Saruta] Fixed performance issues more
64e6642 [Kousuke Saruta] Merge branch 'master' of https://github.com/apache/spark into stage-page-timeline
e4a3354 [Kousuke Saruta] minor code style change
878e3b8 [Kousuke Saruta] Fixed a bug that tooltip remains
b9d8f1b [Kousuke Saruta] Fixed performance issue
ac8842b [Kousuke Saruta] Fixed layout
2319739 [Kousuke Saruta] Modified appearances more
81903ab [Kousuke Saruta] Modified appearances
a79dcc3 [Kousuke Saruta] Modified appearance
55a390c [Kousuke Saruta] Ignored scalastyle for a line-comment
29eae3e [Kousuke Saruta] limited to longest 1000 tasks
2a9e376 [Kousuke Saruta] Minor cleanup
385b6d2 [Kousuke Saruta] Added link feature
ba1ac3e [Kousuke Saruta] Fixed style
2ae8520 [Kousuke Saruta] Updated bootstrap-tooltip.js from 2.2.2 to 2.3.2
af430f1 [Kousuke Saruta] Merge branch 'master' of git://git.apache.org/spark into stage-page-timeline
e694b8e [Kousuke Saruta] Added timeline view to StagePage
8f6610c [Kousuke Saruta] Fixed conflict
b587cf2 [Kousuke Saruta] initial commit
11fe67d [Kousuke Saruta] Fixed conflict
79ac03d [Kousuke Saruta] Merge branch 'master' of git://git.apache.org/spark into timeline-viewer-feature
a91abd3 [Kousuke Saruta] Merge branch 'master' of https://github.com/apache/spark into timeline-viewer-feature
ef34a5b [Kousuke Saruta] Implement tooltip using bootstrap
b09d0c5 [Kousuke Saruta] Move `stroke` and `fill` attribute of rect elements to css
d3c63c8 [Kousuke Saruta] Fixed a little bit bugs
a36291b [Kousuke Saruta] Merge branch 'master' of https://github.com/apache/spark into timeline-viewer-feature
28714b6 [Kousuke Saruta] Fixed highlight issue
0dc4278 [Kousuke Saruta] Addressed most of Patrics's feedbacks
8110acf [Kousuke Saruta] Added scroll limit to Job timeline
974a64a [Kousuke Saruta] Removed unused function
ee7a7f0 [Kousuke Saruta] Refactored
6a91872 [Kousuke Saruta] Temporary commit
6693f34 [Kousuke Saruta] Added link to job/stage box in the timeline in order to move to corresponding row when we click
8f88222 [Kousuke Saruta] Added job/stage description
aeed4b1 [Kousuke Saruta] Removed stage timeline
fc1696c [Kousuke Saruta] Merge branch 'timeline-viewer-feature' of github.com:sarutak/spark into timeline-viewer-feature
999ccd4 [Kousuke Saruta] Improved scalability
0fc6a31 [Kousuke Saruta] Merge branch 'master' of git://git.apache.org/spark into timeline-viewer-feature
19815ae [Kousuke Saruta] Merge branch 'master' of git://git.apache.org/spark into timeline-viewer-feature
68b7540 [Kousuke Saruta] Merge branch 'timeline-viewer-feature' of github.com:sarutak/spark into timeline-viewer-feature
52b5f0b [Kousuke Saruta] Merge branch 'master' of git://git.apache.org/spark into timeline-viewer-feature
dec85db [Kousuke Saruta] Merge branch 'master' of git://git.apache.org/spark into timeline-viewer-feature
fcdab7d [Kousuke Saruta] Merge branch 'master' of git://git.apache.org/spark into timeline-viewer-feature
dab7cc1 [Kousuke Saruta] Merge branch 'master' of git://git.apache.org/spark into timeline-viewer-feature
09cce97 [Kousuke Saruta] Cleanuped
16f82cf [Kousuke Saruta] Cleanuped
9fb522e [Kousuke Saruta] Cleanuped
d05f2c2 [Kousuke Saruta] Merge branch 'master' of git://git.apache.org/spark into timeline-viewer-feature
e85e9aa [Kousuke Saruta] Cleanup: Added TimelineViewUtils.scala
a76e569 [Kousuke Saruta] Removed unused setting in timeline-view.css
5ce1b21 [Kousuke Saruta] Added vis.min.js, vis.min.css and vis.map to .rat-exclude
082f709 [Kousuke Saruta] Added Timeline-View feature for Applications, Jobs and Stages
Added a simple checking for SparkContext.
Also added two rational checking against null at AM object.
Author: ehnalis <zoltan.zvara@gmail.com>
Closes#6083 from ehnalis/cluster and squashes the following commits:
926bd96 [ehnalis] Moved check to SparkContext.
7c89b6e [ehnalis] Remove false line.
ea2a5fe [ehnalis] [SPARK-7504] [YARN] NullPointerException when initializing SparkContext in YARN-cluster mode
4924e01 [ehnalis] [SPARK-7504] [YARN] NullPointerException when initializing SparkContext in YARN-cluster mode
39e4fa3 [ehnalis] SPARK-7504 [YARN] NullPointerException when initializing SparkContext in YARN-cluster mode
9f287c5 [ehnalis] [SPARK-7504] [YARN] NullPointerException when initializing SparkContext in YARN-cluster mode
<h3>Description</h3>
Detect REPL mode once per JVM lifespan.
Previous behavior was to check presence of interpreter mode every time a job was submitted. In the case of execution of multiple short-living jobs this was causing massive mutual blocks between submission threads.
For more details please refer to https://issues.apache.org/jira/browse/SPARK-7233.
<h3>Notes</h3>
* I inverted the return value in case of catching an exception from `true` to `false`. It seems more logical to assume that if the REPL class is not found, we aren't in the interpreter mode.
* I'd personally would call `classForName` with just a Spark classloader (`org.apache.spark.util.Utils#getSparkClassLoader`) but `org.apache.spark.util.Utils#getContextOrSparkClassLoader` is said to be preferable.
* I struggled to come up with a concise, readable and clear unit test. Suggestions are welcome if you feel necessary.
Author: Oleksii Kostyliev <etander@gmail.com>
Author: Oleksii Kostyliev <okostyliev@thunderhead.com>
Closes#5835 from preeze/SPARK-7233 and squashes the following commits:
69bb9e4 [Oleksii Kostyliev] SPARK-7527: fixed explanatory comment to meet style-checker requirements
26dcc24 [Oleksii Kostyliev] SPARK-7527: fixed explanatory comment to meet style-checker requirements
c6f9685 [Oleksii Kostyliev] Merge remote-tracking branch 'remotes/upstream/master' into SPARK-7233
b78a983 [Oleksii Kostyliev] SPARK-7527: revert the fix and let it be addressed separately at a later stage
b64d441 [Oleksii Kostyliev] SPARK-7233: inline inInterpreter parameter into instantiateClass
86e2606 [Oleksii Kostyliev] SPARK-7233, SPARK-7527: Handle interpreter mode properly.
c7ee69c [Oleksii Kostyliev] Merge remote-tracking branch 'upstream/master' into SPARK-7233
d6c07fc [Oleksii Kostyliev] SPARK-7233: properly handle the inverted meaning of isInInterpreter
c319039 [Oleksii Kostyliev] SPARK-7233: move inInterpreter to Utils and make it lazy
cc tdas
Author: zsxwing <zsxwing@gmail.com>
Closes#6160 from zsxwing/SPARK-7650 and squashes the following commits:
fe6ae15 [zsxwing] Fix the import order
a4ffd99 [zsxwing] Merge branch 'master' into SPARK-7650
dc402b6 [zsxwing] Move streaming css and js files to the streaming project
It doesn't look to me Heartbeat is sent to Worker from anyone.
Author: Kan Zhang <kzhang@apache.org>
Closes#6163 from kanzhang/deadwood and squashes the following commits:
56be118 [Kan Zhang] [core] Remove unreachable Heartbeat message from Worker
Use window.localStorage to store the status rather than the url so that the url won't be changed.
cc tdas
Author: zsxwing <zsxwing@gmail.com>
Closes#6158 from zsxwing/SPARK-7649 and squashes the following commits:
3c56fef [zsxwing] Use window.localStorage to store the status rather than the url
`dataDistribution` and `partitions` are `Option[Seq[_]]`. andrewor14 squito
Author: Xiangrui Meng <meng@databricks.com>
Closes#6157 from mengxr/SPARK-7643 and squashes the following commits:
99fe8a4 [Xiangrui Meng] use the correct size in RDDPage for storage info and partitions
In Spark Standalone setup, when some workers are DEAD, they will stay in master worker list for a while.
master.workers metrics for master is only showing the total number of workers, we need to monitor how many real ALIVE workers are there to ensure the cluster is healthy.
Author: Rex Xiong <pengx@microsoft.com>
Closes#6117 from twilightgod/add-aliveWorker-metrics and squashes the following commits:
6be69a5 [Rex Xiong] Fix comment for aliveWorkers metrics
a882f39 [Rex Xiong] Fix style for aliveWorkers metrics
38ce955 [Rex Xiong] Add aliveWorkers metrics in Master
zsxwing
Author: Andrew Or <andrew@databricks.com>
Closes#6138 from andrewor14/dag-viz-clean-properly and squashes the following commits:
19d4e98 [Andrew Or] Add synchronize
02542d6 [Andrew Or] Rename overloaded variable
d11bee1 [Andrew Or] Don't wait until all stages have started before rendering
This patch introduces a new shuffle manager that enhances the existing sort-based shuffle with a new cache-friendly sort algorithm that operates directly on binary data. The goals of this patch are to lower memory usage and Java object overheads during shuffle and to speed up sorting. It also lays groundwork for follow-up patches that will enable end-to-end processing of serialized records.
The new shuffle manager, `UnsafeShuffleManager`, can be enabled by setting `spark.shuffle.manager=tungsten-sort` in SparkConf.
The new shuffle manager uses directly-managed memory to implement several performance optimizations for certain types of shuffles. In cases where the new performance optimizations cannot be applied, the new shuffle manager delegates to SortShuffleManager to handle those shuffles.
UnsafeShuffleManager's optimizations will apply when _all_ of the following conditions hold:
- The shuffle dependency specifies no aggregation or output ordering.
- The shuffle serializer supports relocation of serialized values (this is currently supported
by KryoSerializer and Spark SQL's custom serializers).
- The shuffle produces fewer than 16777216 output partitions.
- No individual record is larger than 128 MB when serialized.
In addition, extra spill-merging optimizations are automatically applied when the shuffle compression codec supports concatenation of serialized streams. This is currently supported by Spark's LZF serializer.
At a high-level, UnsafeShuffleManager's design is similar to Spark's existing SortShuffleManager. In sort-based shuffle, incoming records are sorted according to their target partition ids, then written to a single map output file. Reducers fetch contiguous regions of this file in order to read their portion of the map output. In cases where the map output data is too large to fit in memory, sorted subsets of the output can are spilled to disk and those on-disk files are merged to produce the final output file.
UnsafeShuffleManager optimizes this process in several ways:
- Its sort operates on serialized binary data rather than Java objects, which reduces memory consumption and GC overheads. This optimization requires the record serializer to have certain properties to allow serialized records to be re-ordered without requiring deserialization. See SPARK-4550, where this optimization was first proposed and implemented, for more details.
- It uses a specialized cache-efficient sorter (UnsafeShuffleExternalSorter) that sorts arrays of compressed record pointers and partition ids. By using only 8 bytes of space per record in the sorting array, this fits more of the array into cache.
- The spill merging procedure operates on blocks of serialized records that belong to the same partition and does not need to deserialize records during the merge.
- When the spill compression codec supports concatenation of compressed data, the spill merge simply concatenates the serialized and compressed spill partitions to produce the final output partition. This allows efficient data copying methods, like NIO's `transferTo`, to be used and avoids the need to allocate decompression or copying buffers during the merge.
The shuffle read path is unchanged.
This patch is similar to [SPARK-4550](http://issues.apache.org/jira/browse/SPARK-4550) / #4450 but uses a slightly different implementation. The `unsafe`-based implementation featured in this patch lays the groundwork for followup patches that will enable sorting to operate on serialized data pages that will be prepared by Spark SQL's new `unsafe` operators (such as the new aggregation operator introduced in #5725).
### Future work
There are several tasks that build upon this patch, which will be left to future work:
- [SPARK-7271](https://issues.apache.org/jira/browse/SPARK-7271) Redesign / extend the shuffle interfaces to accept binary data as input. The goal here is to let us bypass serialization steps in cases where the sort input is produced by an operator that operates directly on binary data.
- Extension / redesign of the `Serializer` API. We can add new methods which allow serializers to determine the size requirements for serializing objects and for serializing objects directly to a specified memory address (similar to how `UnsafeRowConverter` works in Spark SQL).
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/apache/spark/5868)
<!-- Reviewable:end -->
Author: Josh Rosen <joshrosen@databricks.com>
Closes#5868 from JoshRosen/unsafe-sort and squashes the following commits:
ef0a86e [Josh Rosen] Fix scalastyle errors
7610f2f [Josh Rosen] Add tests for proper cleanup of shuffle data.
d494ffe [Josh Rosen] Fix deserialization of JavaSerializer instances.
52a9981 [Josh Rosen] Fix some bugs in the address packing code.
51812a7 [Josh Rosen] Change shuffle manager sort name to tungsten-sort
4023fa4 [Josh Rosen] Add @Private annotation to some Java classes.
de40b9d [Josh Rosen] More comments to try to explain metrics code
df07699 [Josh Rosen] Attempt to clarify confusing metrics update code
5e189c6 [Josh Rosen] Track time spend closing / flushing files; split TimeTrackingOutputStream into separate file.
d5779c6 [Josh Rosen] Merge remote-tracking branch 'origin/master' into unsafe-sort
c2ce78e [Josh Rosen] Fix a missed usage of MAX_PARTITION_ID
e3b8855 [Josh Rosen] Cleanup in UnsafeShuffleWriter
4a2c785 [Josh Rosen] rename 'sort buffer' to 'pointer array'
6276168 [Josh Rosen] Remove ability to disable spilling in UnsafeShuffleExternalSorter.
57312c9 [Josh Rosen] Clarify fileBufferSize units
2d4e4f4 [Josh Rosen] Address some minor comments in UnsafeShuffleExternalSorter.
fdcac08 [Josh Rosen] Guard against overflow when expanding sort buffer.
85da63f [Josh Rosen] Cleanup in UnsafeShuffleSorterIterator.
0ad34da [Josh Rosen] Fix off-by-one in nextInt() call
56781a1 [Josh Rosen] Rename UnsafeShuffleSorter to UnsafeShuffleInMemorySorter
e995d1a [Josh Rosen] Introduce MAX_SHUFFLE_OUTPUT_PARTITIONS.
e58a6b4 [Josh Rosen] Add more tests for PackedRecordPointer encoding.
4f0b770 [Josh Rosen] Attempt to implement proper shuffle write metrics.
d4e6d89 [Josh Rosen] Update to bit shifting constants
69d5899 [Josh Rosen] Remove some unnecessary override vals
8531286 [Josh Rosen] Add tests that automatically trigger spills.
7c953f9 [Josh Rosen] Add test that covers UnsafeShuffleSortDataFormat.swap().
e1855e5 [Josh Rosen] Fix a handful of misc. IntelliJ inspections
39434f9 [Josh Rosen] Avoid integer multiplication overflow in getMemoryUsage (thanks FindBugs!)
1e3ad52 [Josh Rosen] Delete unused ByteBufferOutputStream class.
ea4f85f [Josh Rosen] Roll back an unnecessary change in Spillable.
ae538dc [Josh Rosen] Document UnsafeShuffleManager.
ec6d626 [Josh Rosen] Add notes on maximum # of supported shuffle partitions.
0d4d199 [Josh Rosen] Bump up shuffle.memoryFraction to make tests pass.
b3b1924 [Josh Rosen] Properly implement close() and flush() in DummySerializerInstance.
1ef56c7 [Josh Rosen] Revise compression codec support in merger; test cross product of configurations.
b57c17f [Josh Rosen] Disable some overly-verbose logs that rendered DEBUG useless.
f780fb1 [Josh Rosen] Add test demonstrating which compression codecs support concatenation.
4a01c45 [Josh Rosen] Remove unnecessary log message
27b18b0 [Josh Rosen] That for inserting records AT the max record size.
fcd9a3c [Josh Rosen] Add notes + tests for maximum record / page sizes.
9d1ee7c [Josh Rosen] Fix MiMa excludes for ShuffleWriter change
fd4bb9e [Josh Rosen] Use own ByteBufferOutputStream rather than Kryo's
67d25ba [Josh Rosen] Update Exchange operator's copying logic to account for new shuffle manager
8f5061a [Josh Rosen] Strengthen assertion to check partitioning
01afc74 [Josh Rosen] Actually read data in UnsafeShuffleWriterSuite
1929a74 [Josh Rosen] Update to reflect upstream ShuffleBlockManager -> ShuffleBlockResolver rename.
e8718dd [Josh Rosen] Merge remote-tracking branch 'origin/master' into unsafe-sort
9b7ebed [Josh Rosen] More defensive programming RE: cleaning up spill files and memory after errors
7cd013b [Josh Rosen] Begin refactoring to enable proper tests for spilling.
722849b [Josh Rosen] Add workaround for transferTo() bug in merging code; refactor tests.
9883e30 [Josh Rosen] Merge remote-tracking branch 'origin/master' into unsafe-sort
b95e642 [Josh Rosen] Refactor and document logic that decides when to spill.
1ce1300 [Josh Rosen] More minor cleanup
5e8cf75 [Josh Rosen] More minor cleanup
e67f1ea [Josh Rosen] Remove upper type bound in ShuffleWriter interface.
cfe0ec4 [Josh Rosen] Address a number of minor review comments:
8a6fe52 [Josh Rosen] Rename UnsafeShuffleSpillWriter to UnsafeShuffleExternalSorter
11feeb6 [Josh Rosen] Update TODOs related to shuffle write metrics.
b674412 [Josh Rosen] Merge remote-tracking branch 'origin/master' into unsafe-sort
aaea17b [Josh Rosen] Add comments to UnsafeShuffleSpillWriter.
4f70141 [Josh Rosen] Fix merging; now passes UnsafeShuffleSuite tests.
133c8c9 [Josh Rosen] WIP towards testing UnsafeShuffleWriter.
f480fb2 [Josh Rosen] WIP in mega-refactoring towards shuffle-specific sort.
57f1ec0 [Josh Rosen] WIP towards packed record pointers for use in optimized shuffle sort.
69232fd [Josh Rosen] Enable compressible address encoding for off-heap mode.
7ee918e [Josh Rosen] Re-order imports in tests
3aeaff7 [Josh Rosen] More refactoring and cleanup; begin cleaning iterator interfaces
3490512 [Josh Rosen] Misc. cleanup
f156a8f [Josh Rosen] Hacky metrics integration; refactor some interfaces.
2776aca [Josh Rosen] First passing test for ExternalSorter.
5e100b2 [Josh Rosen] Super-messy WIP on external sort
595923a [Josh Rosen] Remove some unused variables.
8958584 [Josh Rosen] Fix bug in calculating free space in current page.
f17fa8f [Josh Rosen] Add missing newline
c2fca17 [Josh Rosen] Small refactoring of SerializerPropertiesSuite to enable test re-use:
b8a09fe [Josh Rosen] Back out accidental log4j.properties change
bfc12d3 [Josh Rosen] Add tests for serializer relocation property.
240864c [Josh Rosen] Remove PrefixComputer and require prefix to be specified as part of insert()
1433b42 [Josh Rosen] Store record length as int instead of long.
026b497 [Josh Rosen] Re-use a buffer in UnsafeShuffleWriter
0748458 [Josh Rosen] Port UnsafeShuffleWriter to Java.
87e721b [Josh Rosen] Renaming and comments
d3cc310 [Josh Rosen] Flag that SparkSqlSerializer2 supports relocation
e2d96ca [Josh Rosen] Expand serializer API and use new function to help control when new UnsafeShuffle path is used.
e267cee [Josh Rosen] Fix compilation of UnsafeSorterSuite
9c6cf58 [Josh Rosen] Refactor to use DiskBlockObjectWriter.
253f13e [Josh Rosen] More cleanup
8e3ec20 [Josh Rosen] Begin code cleanup.
4d2f5e1 [Josh Rosen] WIP
3db12de [Josh Rosen] Minor simplification and sanity checks in UnsafeSorter
767d3ca [Josh Rosen] Fix invalid range in UnsafeSorter.
e900152 [Josh Rosen] Add test for empty iterator in UnsafeSorter
57a4ea0 [Josh Rosen] Make initialSize configurable in UnsafeSorter
abf7bfe [Josh Rosen] Add basic test case.
81d52c5 [Josh Rosen] WIP on UnsafeSorter
This is pretty useful for MLlib.
<img src="https://cloud.githubusercontent.com/assets/2133137/7599650/c7d03dd8-f8b8-11e4-8c0a-0a89e786c90f.png" width="400px"/>
Author: Andrew Or <andrew@databricks.com>
Closes#6100 from andrewor14/dag-viz-hover and squashes the following commits:
fefe2af [Andrew Or] Link tooltips for nodes that belong to the same RDD
90c6a7e [Andrew Or] Assign classes to clusters and nodes, not IDs
Subsequent fix following #5966. I tried this out locally.
Author: Andrew Or <andrew@databricks.com>
Closes#6129 from andrewor14/211-compilation and squashes the following commits:
713868f [Andrew Or] Fix compilation issue for scala 2.11
This is necessary for streaming and long-running Spark applications. zsxwing tdas
Author: Andrew Or <andrew@databricks.com>
Closes#6125 from andrewor14/viz-listener-leak and squashes the following commits:
8660949 [Andrew Or] Fix thing + add tests
33c0843 [Andrew Or] Clean up old job state
…arch.
- zSeries 64-bit Java reports its architecture as s390x, so enhance the 64-bit check to accommodate that value.
- SizeEstimator can detect whether IBM Java is using compressed object pointers using info in the "java.vm.info" property, so will do a better job than failing on the HotSpot MBean and guessing.
Author: Tim Ellison <t.p.ellison@gmail.com>
Closes#6085 from tellison/SizeEstimator and squashes the following commits:
1b6ff6a [Tim Ellison] Merge branch 'master' of https://github.com/apache/spark into SizeEstimator
0968989 [Tim Ellison] [MINOR] Enhance SizeEstimator to detect IBM compressed refs and s390 arch.
escape spaces in the arguments.
Author: Masayoshi TSUZUKI <tsudukim@oss.nttdata.co.jp>
Author: Kousuke Saruta <sarutak@oss.nttdata.co.jp>
Closes#5447 from tsudukim/feature/SPARK-6568-2 and squashes the following commits:
3f9a188 [Masayoshi TSUZUKI] modified some errors.
ed46047 [Masayoshi TSUZUKI] avoid scalastyle errors.
1784239 [Masayoshi TSUZUKI] removed Utils.formatPath.
e03f289 [Masayoshi TSUZUKI] removed testWindows from Utils.resolveURI and Utils.resolveURIs. replaced SystemUtils.IS_OS_WINDOWS to Utils.isWindows. removed Utils.formatPath from PythonRunner.scala.
84c33d0 [Masayoshi TSUZUKI] - use resolveURI in nonLocalPaths - run tests for Windows path only on Windows
016128d [Masayoshi TSUZUKI] fixed to use File.toURI()
2c62e3b [Masayoshi TSUZUKI] Merge pull request #1 from sarutak/SPARK-6568-2
7019a8a [Masayoshi TSUZUKI] Merge branch 'master' of https://github.com/apache/spark into feature/SPARK-6568-2
45946ee [Kousuke Saruta] Merge branch 'master' of git://git.apache.org/spark into SPARK-6568-2
10f1c73 [Kousuke Saruta] Added a comment
93c3c40 [Kousuke Saruta] Merge branch 'classpath-handling-fix' of github.com:sarutak/spark into SPARK-6568-2
649da82 [Kousuke Saruta] Fix classpath handling
c7ba6a7 [Masayoshi TSUZUKI] [SPARK-6568] spark-shell.cmd --jars option does not accept the jar that has space in its path
These R process only used to communicate with JVM process on local, so binding to localhost is more reasonable then wildcard ip.
Author: linweizhong <linweizhong@huawei.com>
Closes#6053 from Sephiroth-Lin/spark-7526 and squashes the following commits:
5303af7 [linweizhong] bind to localhost rather than wildcard ip
This fixes the label bleeding issue described in the JIRA and pictured in the screenshots below. I also took the opportunity to move some code to the places that they belong more to. In particular:
(1) Drawing cluster labels is now implemented in my branch of dagre-d3 instead of in Spark
(2) All graph styling is now moved from Scala to JS
Note that these changes are related because our existing mechanism of "tacking on cluster labels" afterwards isn't flexible enough for us to fix issues like this one easily. For the other half of the changes, visit http://github.com/andrewor14/dagre-d3.
-------------------
**Before.**
<img src="https://cloud.githubusercontent.com/assets/2133137/7582769/b1423440-f845-11e4-8248-b3446a01bf79.png" width="300px"/>
-------------------
**After.**
<img src="https://cloud.githubusercontent.com/assets/2133137/7582742/74891ae6-f845-11e4-96c4-41c7b8aedbdf.png" width="400px"/>
Author: Andrew Or <andrew@databricks.com>
Closes#6076 from andrewor14/dag-viz-bleed and squashes the following commits:
5858d7a [Andrew Or] Merge branch 'master' of github.com:apache/spark into dag-viz-bleed
c686dc4 [Andrew Or] Fix tooltip placement
d908c36 [Andrew Or] Add link to dagre-d3 changes (minor)
4a4fb58 [Andrew Or] Fix bleeding + move all styling to JS