This adds a new ShuffleManager based on sorting, as described in https://issues.apache.org/jira/browse/SPARK-2045. The bulk of the code is in an ExternalSorter class that is similar to ExternalAppendOnlyMap, but sorts key-value pairs by partition ID and can be used to create a single sorted file with a map task's output. (Longer-term I think this can take on the remaining functionality in ExternalAppendOnlyMap and replace it so we don't have code duplication.)
The main TODOs still left are:
- [x] enabling ExternalSorter to merge across spilled files
- [x] with an Ordering
- [x] without an Ordering, using the keys' hash codes
- [x] adding more tests (e.g. a version of our shuffle suite that runs on this)
- [x] rebasing on top of the size-tracking refactoring in #1165 when that is merged
- [x] disabling spilling if spark.shuffle.spill is set to false
Despite this though, this seems to work pretty well (running successfully in cases where the hash shuffle would OOM, such as 1000 reduce tasks on executors with only 1G memory), and it seems to be comparable in speed or faster than hash-based shuffle (it will create much fewer files for the OS to keep track of). So I'm posting it to get some early feedback.
After these TODOs are done, I'd also like to enable ExternalSorter to sort data within each partition by a key as well, which will allow us to use it to implement external spilling in reduce tasks in `sortByKey`.
Author: Matei Zaharia <matei@databricks.com>
Closes#1499 from mateiz/sort-based-shuffle and squashes the following commits:
bd841f9 [Matei Zaharia] Various review comments
d1c137fd [Matei Zaharia] Various review comments
a611159 [Matei Zaharia] Compile fixes due to rebase
62c56c8 [Matei Zaharia] Fix ShuffledRDD sometimes not returning Tuple2s.
f617432 [Matei Zaharia] Fix a failing test (seems to be due to change in SizeTracker logic)
9464d5f [Matei Zaharia] Simplify code and fix conflicts after latest rebase
0174149 [Matei Zaharia] Add cleanup behavior and cleanup tests for sort-based shuffle
eb4ee0d [Matei Zaharia] Remove customizable element type in ShuffledRDD
fa2e8db [Matei Zaharia] Allow nextBatchStream to be called after we're done looking at all streams
a34b352 [Matei Zaharia] Fix tracking of indices within a partition in SpillReader, and add test
03e1006 [Matei Zaharia] Add a SortShuffleSuite that runs ShuffleSuite with sort-based shuffle
3c7ff1f [Matei Zaharia] Obey the spark.shuffle.spill setting in ExternalSorter
ad65fbd [Matei Zaharia] Rebase on top of Aaron's Sorter change, and use Sorter in our buffer
44d2a93 [Matei Zaharia] Use estimateSize instead of atGrowThreshold to test collection sizes
5686f71 [Matei Zaharia] Optimize merging phase for in-memory only data:
5461cbb [Matei Zaharia] Review comments and more tests (e.g. tests with 1 element per partition)
e9ad356 [Matei Zaharia] Update ContextCleanerSuite to make sure shuffle cleanup tests use hash shuffle (since they were written for it)
c72362a [Matei Zaharia] Added bug fix and test for when iterators are empty
de1fb40 [Matei Zaharia] Make trait SizeTrackingCollection private[spark]
4988d16 [Matei Zaharia] tweak
c1b7572 [Matei Zaharia] Small optimization
ba7db7f [Matei Zaharia] Handle null keys in hash-based comparator, and add tests for collisions
ef4e397 [Matei Zaharia] Support for partial aggregation even without an Ordering
4b7a5ce [Matei Zaharia] More tests, and ability to sort data if a total ordering is given
e1f84be [Matei Zaharia] Fix disk block manager test
5a40a1c [Matei Zaharia] More tests
614f1b4 [Matei Zaharia] Add spill metrics to map tasks
cc52caf [Matei Zaharia] Add more error handling and tests for error cases
bbf359d [Matei Zaharia] More work
3a56341 [Matei Zaharia] More partial work towards sort-based shuffle
7a0895d [Matei Zaharia] Some more partial work towards sort-based shuffle
b615476 [Matei Zaharia] Scaffolding for sort-based shuffle
In a few places in MLlib, an expression of the form `log(1.0 + p)` is evaluated. When p is so small that `1.0 + p == 1.0`, the result is 0.0. However the correct answer is very near `p`. This is why `Math.log1p` exists.
Similarly for one instance of `exp(m) - 1` in GraphX; there's a special `Math.expm1` method.
While the errors occur only for very small arguments, given their use in machine learning algorithms, this is entirely possible.
Also note the related PR for Python: https://github.com/apache/spark/pull/1652
Author: Sean Owen <srowen@gmail.com>
Closes#1659 from srowen/SPARK-2748 and squashes the following commits:
c5926d4 [Sean Owen] Use log1p, expm1 for better precision for tiny arguments
JIRA issue: [SPARK-2410](https://issues.apache.org/jira/browse/SPARK-2410)
Another try for #1399 & #1600. Those two PR breaks Jenkins builds because we made a separate profile `hive-thriftserver` in sub-project `assembly`, but the `hive-thriftserver` module is defined outside the `hive-thriftserver` profile. Thus every time a pull request that doesn't touch SQL code will also execute test suites defined in `hive-thriftserver`, but tests fail because related .class files are not included in the assembly jar.
In the most recent commit, module `hive-thriftserver` is moved into its own profile to fix this problem. All previous commits are squashed for clarity.
Author: Cheng Lian <lian.cs.zju@gmail.com>
Closes#1620 from liancheng/jdbc-with-maven-fix and squashes the following commits:
629988e [Cheng Lian] Moved hive-thriftserver module definition into its own profile
ec3c7a7 [Cheng Lian] Cherry picked the Hive Thrift server
(This is a replacement of #1399, trying to fix potential `HiveThriftServer2` port collision between parallel builds. Please refer to [these comments](https://github.com/apache/spark/pull/1399#issuecomment-50212572) for details.)
JIRA issue: [SPARK-2410](https://issues.apache.org/jira/browse/SPARK-2410)
Merging the Hive Thrift/JDBC server from [branch-1.0-jdbc](https://github.com/apache/spark/tree/branch-1.0-jdbc).
Thanks chenghao-intel for his initial contribution of the Spark SQL CLI.
Author: Cheng Lian <lian.cs.zju@gmail.com>
Closes#1600 from liancheng/jdbc and squashes the following commits:
ac4618b [Cheng Lian] Uses random port for HiveThriftServer2 to avoid collision with parallel builds
090beea [Cheng Lian] Revert changes related to SPARK-2678, decided to move them to another PR
21c6cf4 [Cheng Lian] Updated Spark SQL programming guide docs
fe0af31 [Cheng Lian] Reordered spark-submit options in spark-shell[.cmd]
199e3fb [Cheng Lian] Disabled MIMA for hive-thriftserver
1083e9d [Cheng Lian] Fixed failed test suites
7db82a1 [Cheng Lian] Fixed spark-submit application options handling logic
9cc0f06 [Cheng Lian] Starts beeline with spark-submit
cfcf461 [Cheng Lian] Updated documents and build scripts for the newly added hive-thriftserver profile
061880f [Cheng Lian] Addressed all comments by @pwendell
7755062 [Cheng Lian] Adapts test suites to spark-submit settings
40bafef [Cheng Lian] Fixed more license header issues
e214aab [Cheng Lian] Added missing license headers
b8905ba [Cheng Lian] Fixed minor issues in spark-sql and start-thriftserver.sh
f975d22 [Cheng Lian] Updated docs for Hive compatibility and Shark migration guide draft
3ad4e75 [Cheng Lian] Starts spark-sql shell with spark-submit
a5310d1 [Cheng Lian] Make HiveThriftServer2 play well with spark-submit
61f39f4 [Cheng Lian] Starts Hive Thrift server via spark-submit
2c4c539 [Cheng Lian] Cherry picked the Hive Thrift server
This reverts commit 06dc0d2c6b.
#1399 is making Jenkins fail. We should investigate and put this back after its passing tests.
Author: Michael Armbrust <michael@databricks.com>
Closes#1594 from marmbrus/revertJDBC and squashes the following commits:
59748da [Michael Armbrust] Revert "[SPARK-2410][SQL] Merging Hive Thrift/JDBC server"
JIRA issue:
- Main: [SPARK-2410](https://issues.apache.org/jira/browse/SPARK-2410)
- Related: [SPARK-2678](https://issues.apache.org/jira/browse/SPARK-2678)
Cherry picked the Hive Thrift/JDBC server from [branch-1.0-jdbc](https://github.com/apache/spark/tree/branch-1.0-jdbc).
(Thanks chenghao-intel for his initial contribution of the Spark SQL CLI.)
TODO
- [x] Use `spark-submit` to launch the server, the CLI and beeline
- [x] Migration guideline draft for Shark users
----
Hit by a bug in `SparkSubmitArguments` while working on this PR: all application options that are recognized by `SparkSubmitArguments` are stolen as `SparkSubmit` options. For example:
```bash
$ spark-submit --class org.apache.hive.beeline.BeeLine spark-internal --help
```
This actually shows usage information of `SparkSubmit` rather than `BeeLine`.
~~Fixed this bug here since the `spark-internal` related stuff also touches `SparkSubmitArguments` and I'd like to avoid conflict.~~
**UPDATE** The bug mentioned above is now tracked by [SPARK-2678](https://issues.apache.org/jira/browse/SPARK-2678). Decided to revert changes to this bug since it involves more subtle considerations and worth a separate PR.
Author: Cheng Lian <lian.cs.zju@gmail.com>
Closes#1399 from liancheng/thriftserver and squashes the following commits:
090beea [Cheng Lian] Revert changes related to SPARK-2678, decided to move them to another PR
21c6cf4 [Cheng Lian] Updated Spark SQL programming guide docs
fe0af31 [Cheng Lian] Reordered spark-submit options in spark-shell[.cmd]
199e3fb [Cheng Lian] Disabled MIMA for hive-thriftserver
1083e9d [Cheng Lian] Fixed failed test suites
7db82a1 [Cheng Lian] Fixed spark-submit application options handling logic
9cc0f06 [Cheng Lian] Starts beeline with spark-submit
cfcf461 [Cheng Lian] Updated documents and build scripts for the newly added hive-thriftserver profile
061880f [Cheng Lian] Addressed all comments by @pwendell
7755062 [Cheng Lian] Adapts test suites to spark-submit settings
40bafef [Cheng Lian] Fixed more license header issues
e214aab [Cheng Lian] Added missing license headers
b8905ba [Cheng Lian] Fixed minor issues in spark-sql and start-thriftserver.sh
f975d22 [Cheng Lian] Updated docs for Hive compatibility and Shark migration guide draft
3ad4e75 [Cheng Lian] Starts spark-sql shell with spark-submit
a5310d1 [Cheng Lian] Make HiveThriftServer2 play well with spark-submit
61f39f4 [Cheng Lian] Starts Hive Thrift server via spark-submit
2c4c539 [Cheng Lian] Cherry picked the Hive Thrift server
RoutingTableMessage was used to construct routing tables to enable
joining VertexRDDs with partitioned edges. It stored three elements: the
destination vertex ID, the source edge partition, and a byte specifying
the position in which the edge partition referenced the vertex to enable
join elimination.
However, this was incompatible with sort-based shuffle (SPARK-2045). It
was also slightly wasteful, because partition IDs are usually much
smaller than 2^32, though this was mitigated by a custom serializer that
used variable-length encoding.
This commit replaces RoutingTableMessage with a pair of (VertexId, Int)
where the Int encodes both the source partition ID (in the lower 30
bits) and the position (in the top 2 bits).
Author: Ankur Dave <ankurdave@gmail.com>
Closes#1553 from ankurdave/remove-RoutingTableMessage and squashes the following commits:
697e17b [Ankur Dave] Replace RoutingTableMessage with pair
MessageToPartition was used in `Graph#partitionBy`. Unlike a Tuple2, it marked the key as transient to avoid sending it over the network. However, it was incompatible with sort-based shuffle (SPARK-2045) and represented only a minor optimization: for partitionBy, it improved performance by 6.3% (30.4 s to 28.5 s) and reduced communication by 5.6% (114.2 MB to 107.8 MB).
Author: Ankur Dave <ankurdave@gmail.com>
Closes#1537 from ankurdave/remove-MessageToPartition and squashes the following commits:
f9d0054 [Ankur Dave] Remove MessageToPartition
ab71364 [Ankur Dave] Remove unused VertexBroadcastMsg
fix examples
Author: CrazyJvm <crazyjvm@gmail.com>
Closes#1523 from CrazyJvm/graphx-example and squashes the following commits:
663457a [CrazyJvm] outDegrees does not take parameters
7cfff1d [CrazyJvm] fix example for joinVertices
VertexPartition and ShippableVertexPartition are contained in RDDs but are not marked Serializable, leading to NotSerializableExceptions when using Java serialization.
The fix is simply to mark them as Serializable. This PR does that and adds a test for serializing them using Java and Kryo serialization.
Author: Ankur Dave <ankurdave@gmail.com>
Closes#1376 from ankurdave/SPARK-2455 and squashes the following commits:
ed4a51b [Ankur Dave] Make (Shippable)VertexPartition serializable
1fd42c5 [Ankur Dave] Add failing tests for Java serialization
Patch introduces the new way of working also retaining the existing ways of doing things.
For example build instruction for yarn in maven is
`mvn -Pyarn -PHadoop2.2 clean package -DskipTests`
in sbt it can become
`MAVEN_PROFILES="yarn, hadoop-2.2" sbt/sbt clean assembly`
Also supports
`sbt/sbt -Pyarn -Phadoop-2.2 -Dhadoop.version=2.2.0 clean assembly`
Author: Prashant Sharma <prashant.s@imaginea.com>
Author: Patrick Wendell <pwendell@gmail.com>
Closes#772 from ScrapCodes/sbt-maven and squashes the following commits:
a8ac951 [Prashant Sharma] Updated sbt version.
62b09bb [Prashant Sharma] Improvements.
fa6221d [Prashant Sharma] Excluding sql from mima
4b8875e [Prashant Sharma] Sbt assembly no longer builds tools by default.
72651ca [Prashant Sharma] Addresses code reivew comments.
acab73d [Prashant Sharma] Revert "Small fix to run-examples script."
ac4312c [Prashant Sharma] Revert "minor fix"
6af91ac [Prashant Sharma] Ported oldDeps back. + fixes issues with prev commit.
65cf06c [Prashant Sharma] Servelet API jars mess up with the other servlet jars on the class path.
446768e [Prashant Sharma] minor fix
89b9777 [Prashant Sharma] Merge conflicts
d0a02f2 [Prashant Sharma] Bumped up pom versions, Since the build now depends on pom it is better updated there. + general cleanups.
dccc8ac [Prashant Sharma] updated mima to check against 1.0
a49c61b [Prashant Sharma] Fix for tools jar
a2f5ae1 [Prashant Sharma] Fixes a bug in dependencies.
cf88758 [Prashant Sharma] cleanup
9439ea3 [Prashant Sharma] Small fix to run-examples script.
96cea1f [Prashant Sharma] SPARK-1776 Have Spark's SBT build read dependencies from Maven.
36efa62 [Patrick Wendell] Set project name in pom files and added eclipse/intellij plugins.
4973dbd [Patrick Wendell] Example build using pom reader.
This PR is a sub-task of SPARK-2044 to move the execution of aggregation into shuffle implementations.
I leave `CoGoupedRDD` and `SubtractedRDD` unchanged because they have their implementations of aggregation. I'm not sure is it suitable to change these two RDDs.
Also I do not move sort related code of `OrderedRDDFunctions` into shuffle, this will be solved in another sub-task.
Author: jerryshao <saisai.shao@intel.com>
Closes#1064 from jerryshao/SPARK-2124 and squashes the following commits:
4a05a40 [jerryshao] Modify according to comments
1f7dcc8 [jerryshao] Style changes
50a2fd6 [jerryshao] Fix test suite issue after moving aggregator to Shuffle reader and writer
1a96190 [jerryshao] Code modification related to the ShuffledRDD
308f635 [jerryshao] initial works of move combiner to ShuffleManager's reader and writer
In GraphImpl, mapVertices and outerJoinVertices use a more efficient implementation when the map function conserves vertex attribute types. This is implemented by comparing the ClassTags of the old and new vertex attribute types. However, ClassTags store erased types, so the comparison will return a false positive for types with different type parameters, such as Option[Int] and Option[Double].
This PR resolves the problem by requesting that the compiler generate evidence of equality between the old and new vertex attribute types, and providing a default value for the evidence parameter if the two types are not equal. The methods can then check the value of the evidence parameter to see whether the types are equal.
It also adds a test called "mapVertices changing type with same erased type" that failed before the PR and succeeds now.
Callers of mapVertices and outerJoinVertices can no longer use a wildcard for a graph's VD type. To avoid "Error occurred in an application involving default arguments," they must bind VD to a type parameter, as this PR does for ShortestPaths and LabelPropagation.
Author: Ankur Dave <ankurdave@gmail.com>
Closes#967 from ankurdave/SPARK-1552 and squashes the following commits:
68a4fff [Ankur Dave] Undo conserve naming
7388705 [Ankur Dave] Remove unnecessary ClassTag for VD parameters
a704e5f [Ankur Dave] Use type equality constraint with default argument
29a5ab7 [Ankur Dave] Add failing test
f458c83 [Ankur Dave] Revert "[SPARK-1552] Fix type comparison bug in mapVertices and outerJoinVertices"
16d6af8 [Ankur Dave] [SPARK-1552] Fix type comparison bug in mapVertices and outerJoinVertices
Due to a bug introduced by apache/spark#497, Pregel does not unpersist replicated vertices from previous iterations. As a result, they stay cached until memory is full, wasting GC time.
This PR corrects the problem by unpersisting both the edges and the replicated vertices of previous iterations. This is safe because the edges and replicated vertices of the current iteration are cached by the call to `g.cache()` and then materialized by the call to `messages.count()`. Therefore no unmaterialized RDDs depend on `prevG.edges`. I verified that no recomputation occurs by running PageRank with a custom patch to Spark that warns when a partition is recomputed.
Thanks to Tim Weninger for reporting this bug.
Author: Ankur Dave <ankurdave@gmail.com>
Closes#972 from ankurdave/SPARK-2025 and squashes the following commits:
13d5b07 [Ankur Dave] Unpersist edges of previous graph in Pregel
Author: Takuya UESHIN <ueshin@happy-camper.st>
Closes#974 from ueshin/issues/SPARK-2029 and squashes the following commits:
e19e8f4 [Takuya UESHIN] Bump version number to 1.1.0-SNAPSHOT.
Author: Ankur Dave <ankurdave@gmail.com>
Closes#970 from ankurdave/SPARK-1991_docfix and squashes the following commits:
6d07343 [Ankur Dave] Minor: Fix documentation error from apache/spark#946
It is currently very difficult to repartition a graph over a different number of partitions. This PR adds an additional `partitionBy` function that takes the number of partitions.
Author: Joseph E. Gonzalez <joseph.e.gonzalez@gmail.com>
Closes#719 from jegonzal/graph_partitioning_options and squashes the following commits:
730b405 [Joseph E. Gonzalez] adding an additional number of partitions option to partitionBy
This PR adds support for specifying custom storage levels for the vertices and edges of a graph. This enables GraphX to handle graphs larger than memory size by specifying MEMORY_AND_DISK and then repartitioning the graph to use many small partitions, each of which does fit in memory. Spark will then automatically load partitions from disk as needed.
The user specifies the desired vertex and edge storage levels when building the graph by passing them to the graph constructor. These are then stored in the `targetStorageLevel` attribute of the VertexRDD and EdgeRDD respectively. Whenever GraphX needs to cache a VertexRDD or EdgeRDD (because it plans to use it more than once, for example), it uses the specified target storage level. Also, when the user calls `Graph#cache()`, the vertices and edges are persisted using their target storage levels.
In order to facilitate propagating the target storage levels across VertexRDD and EdgeRDD operations, we remove raw calls to the constructors and instead introduce the `withPartitionsRDD` and `withTargetStorageLevel` methods.
I tested this change by running PageRank and triangle count on a severely memory-constrained cluster (1 executor with 300 MB of memory, and a 1 GB graph). Before this PR, these algorithms used to fail with OutOfMemoryErrors. With this PR, and using the DISK_ONLY storage level, they succeed.
Author: Ankur Dave <ankurdave@gmail.com>
Closes#946 from ankurdave/SPARK-1991 and squashes the following commits:
ce17d95 [Ankur Dave] Move pickStorageLevel to StorageLevel.fromString
ccaf06f [Ankur Dave] Shadow members in withXYZ() methods rather than using underscores
c34abc0 [Ankur Dave] Exclude all of GraphX from compatibility checks vs. 1.0.0
c5ca068 [Ankur Dave] Revert "Exclude all of GraphX from binary compatibility checks"
34bcefb [Ankur Dave] Exclude all of GraphX from binary compatibility checks
6fdd137 [Ankur Dave] [SPARK-1991] Support custom storage levels for vertices and edges
This PR accomplishes two things:
1. It introduces a Synthetic Benchmark application that generates an arbitrarily large log-normal graph and executes either PageRank or connected components on the graph. This can be used to profile GraphX system on arbitrary clusters without access to large graph datasets
2. This PR improves the implementation of the log-normal graph generator.
Author: Joseph E. Gonzalez <joseph.e.gonzalez@gmail.com>
Author: Ankur Dave <ankurdave@gmail.com>
Closes#720 from jegonzal/graphx_synth_benchmark and squashes the following commits:
e40812a [Ankur Dave] Exclude all of GraphX from compatibility checks vs. 1.0.0
bccccad [Ankur Dave] Fix long lines
374678a [Ankur Dave] Bugfix and style changes
1bdf39a [Joseph E. Gonzalez] updating options
d943972 [Joseph E. Gonzalez] moving the benchmark application into the examples folder.
f4f839a [Joseph E. Gonzalez] Creating a synthetic benchmark script.
stop resetting spark.driver.port in unit tests (scala, java and python).
Author: Syed Hashmi <shashmi@cloudera.com>
Author: CodingCat <zhunansjtu@gmail.com>
Closes#943 from syedhashmi/master and squashes the following commits:
885f210 [Syed Hashmi] Removing unnecessary file (created by mergetool)
b8bd4b5 [Syed Hashmi] Merge remote-tracking branch 'upstream/master'
b895e59 [Syed Hashmi] Revert "[SPARK-1784] Add a new partitioner"
57b6587 [Syed Hashmi] Revert "[SPARK-1784] Add a balanced partitioner"
1574769 [Syed Hashmi] [SPARK-1942] Stop clearing spark.driver.port in unit tests
4354836 [Syed Hashmi] Revert "SPARK-1686: keep schedule() calling in the main thread"
fd36542 [Syed Hashmi] [SPARK-1784] Add a balanced partitioner
6668015 [CodingCat] SPARK-1686: keep schedule() calling in the main thread
4ca94cc [Syed Hashmi] [SPARK-1784] Add a new partitioner
This is a modified version of apache/spark#10.
Author: Ankur Dave <ankurdave@gmail.com>
Author: Andres Perez <andres@tresata.com>
Closes#933 from ankurdave/shortestpaths and squashes the following commits:
03a103c [Ankur Dave] Style fixes
7a1ff48 [Ankur Dave] Improve ShortestPaths documentation
d75c8fc [Ankur Dave] Remove unnecessary VD type param, and pass through ED
d983fb4 [Ankur Dave] Fix style errors
60ed8e6 [Andres Perez] Add Shortest-path computations to graphx.lib with unit tests.
A straightforward implementation of LPA algorithm for detecting graph communities using the Pregel framework. Amongst the growing literature on community detection algorithms in networks, LPA is perhaps the most elementary, and despite its flaws it remains a nice and simple approach.
Author: Ankur Dave <ankurdave@gmail.com>
Author: haroldsultan <haroldsultan@gmail.com>
Author: Harold Sultan <haroldsultan@gmail.com>
Closes#905 from haroldsultan/master and squashes the following commits:
327aee0 [haroldsultan] Merge pull request #2 from ankurdave/label-propagation
227a4d0 [Ankur Dave] Untabify
0ac574c [haroldsultan] Merge pull request #1 from ankurdave/label-propagation
0e24303 [Ankur Dave] Add LabelPropagationSuite
84aa061 [Ankur Dave] LabelPropagation: Fix compile errors and style; rename from LPA
9830342 [Harold Sultan] initial version of LPA
905173df57 introduced a bug in partitionBy where, after repartitioning the edges, it reuses the VertexRDD without updating the routing tables to reflect the new edge layout. Subsequent accesses of the triplets contain nulls for many vertex properties.
This commit adds a test for this bug and fixes it by introducing `VertexRDD#withEdges` and calling it in `partitionBy`.
Author: Ankur Dave <ankurdave@gmail.com>
Closes#885 from ankurdave/SPARK-1931 and squashes the following commits:
3930cdd [Ankur Dave] Note how to set up VertexRDD for efficient joins
9bdbaa4 [Ankur Dave] [SPARK-1931] Reconstruct routing tables in Graph.partitionBy
Author: Zhen Peng <zhenpeng01@baidu.com>
Closes#769 from zhpengg/bugfix-graphx-edge-compare and squashes the following commits:
8a978ff [Zhen Peng] add ut for graphx Edge.lexicographicOrdering.compare
413c258 [Zhen Peng] there maybe a overflow for two Long's substraction
Fixes a misplaced comment from #785.
@pwendell
Author: Prashant Sharma <prashant.s@imaginea.com>
Closes#788 from ScrapCodes/patch-1 and squashes the following commits:
3ef6a69 [Prashant Sharma] Update package-info.java
67d9461 [Prashant Sharma] Update package-info.java
This is a few changes based on the original patch by @scrapcodes.
Author: Prashant Sharma <prashant.s@imaginea.com>
Author: Patrick Wendell <pwendell@gmail.com>
Closes#785 from pwendell/package-docs and squashes the following commits:
c32b731 [Patrick Wendell] Changes based on Prashant's patch
c0463d3 [Prashant Sharma] added eof new line
ce8bf73 [Prashant Sharma] Added eof new line to all files.
4c35f2e [Prashant Sharma] SPARK-1563 Add package-info.java and package.scala files for all packages that appear in docs
Three issues related to temp files that tests generate – these should be touched up for hygiene but are not urgent.
Modules have a log4j.properties which directs the unit-test.log output file to a directory like `[module]/target/unit-test.log`. But this ends up creating `[module]/[module]/target/unit-test.log` instead of former.
The `work/` directory is not deleted by "mvn clean", in the parent and in modules. Neither is the `checkpoint/` directory created under the various external modules.
Many tests create a temp directory, which is not usually deleted. This can be largely resolved by calling `deleteOnExit()` at creation and trying to call `Utils.deleteRecursively` consistently to clean up, sometimes in an `@After` method.
_If anyone seconds the motion, I can create a more significant change that introduces a new test trait along the lines of `LocalSparkContext`, which provides management of temp directories for subclasses to take advantage of._
Author: Sean Owen <sowen@cloudera.com>
Closes#732 from srowen/SPARK-1798 and squashes the following commits:
5af578e [Sean Owen] Try to consistently delete test temp dirs and files, and set deleteOnExit() for each
b21b356 [Sean Owen] Remove work/ and checkpoint/ dirs with mvn clean
bdd0f41 [Sean Owen] Remove duplicate module dir in log4j.properties output path for tests
Addressing issue in MimaBuild.scala.
Author: Ankur Dave <ankurdave@gmail.com>
Author: Joseph E. Gonzalez <joseph.e.gonzalez@gmail.com>
Closes#742 from jegonzal/edge_partition_serialization and squashes the following commits:
8ba6e0d [Ankur Dave] Add concatenation operators to MimaBuild.scala
cb2ed3a [Joseph E. Gonzalez] addressing missing exclusion in MimaBuild.scala
5d27824 [Ankur Dave] Disable reference tracking to fix serialization test
c0a9ae5 [Ankur Dave] Add failing test for EdgePartition Kryo serialization
a4a3faa [Joseph E. Gonzalez] Making EdgePartition serializable.
This appears to address the issue with edge partition serialization. The solution appears to be just registering the `PrimitiveKeyOpenHashMap`. However I noticed that we appear to have forked that code in GraphX but retained the same name (which is confusing). I also renamed our local copy to `GraphXPrimitiveKeyOpenHashMap`. We should consider dropping that and using the one in Spark if possible.
Author: Ankur Dave <ankurdave@gmail.com>
Author: Joseph E. Gonzalez <joseph.e.gonzalez@gmail.com>
Closes#724 from jegonzal/edge_partition_serialization and squashes the following commits:
b0a525a [Ankur Dave] Disable reference tracking to fix serialization test
bb7f548 [Ankur Dave] Add failing test for EdgePartition Kryo serialization
67dac22 [Joseph E. Gonzalez] Making EdgePartition serializable.
Their was a minor bug in which negative partition ids could be generated when constructing a 2D partitioning of a graph. This could lead to an inefficient 2D partition for large vertex id values.
Author: Joseph E. Gonzalez <joseph.e.gonzalez@gmail.com>
Closes#709 from jegonzal/fix_2d_partitioning and squashes the following commits:
937c562 [Joseph E. Gonzalez] fixing bug in 2d partitioning algorithm where negative partition ids could be generated.
This PR makes the following changes, primarily in e4fbd329aef85fe2c38b0167255d2a712893d683:
1. *Unify RDDs to avoid zipPartitions.* A graph used to be four RDDs: vertices, edges, routing table, and triplet view. This commit merges them down to two: vertices (with routing table), and edges (with replicated vertices).
2. *Avoid duplicate shuffle in graph building.* We used to do two shuffles when building a graph: one to extract routing information from the edges and move it to the vertices, and another to find nonexistent vertices referred to by edges. With this commit, the latter is done as a side effect of the former.
3. *Avoid no-op shuffle when joins are fully eliminated.* This is a side effect of unifying the edges and the triplet view.
4. *Join elimination for mapTriplets.*
5. *Ship only the needed vertex attributes when upgrading the triplet view.* If the triplet view already contains source attributes, and we now need both attributes, only ship destination attributes rather than re-shipping both. This is done in `ReplicatedVertexView#upgrade`.
Author: Ankur Dave <ankurdave@gmail.com>
Closes#497 from ankurdave/unify-rdds and squashes the following commits:
332ab43 [Ankur Dave] Merge remote-tracking branch 'apache-spark/master' into unify-rdds
4933e2e [Ankur Dave] Exclude RoutingTable from binary compatibility check
5ba8789 [Ankur Dave] Add GraphX upgrade guide from Spark 0.9.1
13ac845 [Ankur Dave] Merge remote-tracking branch 'apache-spark/master' into unify-rdds
a04765c [Ankur Dave] Remove unnecessary toOps call
57202e8 [Ankur Dave] Replace case with pair parameter
75af062 [Ankur Dave] Add explicit return types
04d3ae5 [Ankur Dave] Convert implicit parameter to context bound
c88b269 [Ankur Dave] Revert upgradeIterator to if-in-a-loop
0d3584c [Ankur Dave] EdgePartition.size should be val
2a928b2 [Ankur Dave] Set locality wait
10b3596 [Ankur Dave] Clean up public API
ae36110 [Ankur Dave] Fix style errors
e4fbd32 [Ankur Dave] Unify GraphImpl RDDs + other graph load optimizations
d6d60e2 [Ankur Dave] In GraphLoader, coalesce to minEdgePartitions
62c7b78 [Ankur Dave] In Analytics, take PageRank numIter
d64e8d4 [Ankur Dave] Log current Pregel iteration
This pull request contains a rebased patch from @heathermiller (https://github.com/heathermiller/spark/pull/1) to add ClassTags on Serializer and types that depend on it (Broadcast and AccumulableCollection). Putting these in the public API signatures now will allow us to use Scala Pickling for serialization down the line without breaking binary compatibility.
One question remaining is whether we also want them on Accumulator -- Accumulator is passed as part of a bigger Task or TaskResult object via the closure serializer so it doesn't seem super useful to add the ClassTag there. Broadcast and AccumulableCollection in contrast were being serialized directly.
CC @rxin, @pwendell, @heathermiller
Author: Matei Zaharia <matei@databricks.com>
Closes#700 from mateiz/spark-1708 and squashes the following commits:
1a3d8b0 [Matei Zaharia] Use fake ClassTag in Java
3b449ed [Matei Zaharia] test fix
2209a27 [Matei Zaharia] Code style fixes
9d48830 [Matei Zaharia] Add a ClassTag on Serializer and things that depend on it
Commit for initial feedback, basically I am curious if we should prompt user for providing args esp. when its mandatory. And can we skip if they are not ?
Also few other things that did not work like
`bin/spark-submit examples/target/scala-2.10/spark-examples-1.0.0-SNAPSHOT-hadoop1.0.4.jar --class org.apache.spark.examples.SparkALS --arg 100 500 10 5 2`
Not all the args get passed properly, may be I have messed up something will try to sort it out hopefully.
Author: Prashant Sharma <prashant.s@imaginea.com>
Closes#552 from ScrapCodes/SPARK-1565/update-examples and squashes the following commits:
669dd23 [Prashant Sharma] Review comments
2727e70 [Prashant Sharma] SPARK-1565, update examples to be used with spark-submit script.
... that do not change schema
Author: Kan Zhang <kzhang@apache.org>
Closes#448 from kanzhang/SPARK-1460 and squashes the following commits:
111e388 [Kan Zhang] silence MiMa errors in EdgeRDD and VertexRDD
91dc787 [Kan Zhang] Taking into account newly added Ordering param
79ed52a [Kan Zhang] [SPARK-1460] Returning SchemaRDD on Set operations that do not change schema
These classes are only serializable to work around closure capture, so their fields should all be marked `@transient` to avoid wasteful serialization.
This PR supersedes apache/spark#519 and fixes the same bug.
Author: Ankur Dave <ankurdave@gmail.com>
Closes#520 from ankurdave/graphx-transient and squashes the following commits:
6431760 [Ankur Dave] Mark all fields of EdgePartition, Graph, and GraphOps `@transient`
Each vertex partition is co-located with a pid2vid array created in RoutingTable.scala. This array maps edge partition IDs to the list of vertices in the current vertex partition that are mentioned by edges in that partition. Therefore the pid2vid array should have one entry per edge partition.
GraphX currently creates one entry per *vertex* partition, which is a bug that leads to an ArrayIndexOutOfBoundsException when there are more edge partitions than vertex partitions. This commit fixes the bug and adds a test for this case.
Resolves SPARK-1329. Thanks to Daniel Darabos for reporting this bug.
Author: Ankur Dave <ankurdave@gmail.com>
Closes#368 from ankurdave/fix-pid2vid-size and squashes the following commits:
5a5c52a [Ankur Dave] SPARK-1329: Create pid2vid with correct number of partitions
GraphImpl.reverse used to reverse edges in each partition of the edge RDD but preserve the routing table and replicated vertex view, since reversing should not affect partitioning.
However, the old routing table would then have incorrect information for srcAttrOnly and dstAttrOnly. These RDDs should be switched.
A simple fix is for Graph.reverse to rebuild the routing table and replicated vertex view.
Thanks to Bogdan Ghidireac for reporting this issue on the [mailing list](http://apache-spark-user-list.1001560.n3.nabble.com/graph-reverse-amp-Pregel-API-td4338.html).
Author: Ankur Dave <ankurdave@gmail.com>
Closes#431 from ankurdave/fix-reverse-bug and squashes the following commits:
75d63cb [Ankur Dave] Rebuild routing table after Graph.reverse
The Graph.apply test in GraphSuite had some assertions in a closure in
a graph transformation. As a consequence, these assertions never
actually executed. Furthermore, these closures had a reference to
(non-serializable) test harness classes because they called assert(),
which could be a problem if we proactively check closure serializability
in the future.
This commit simply changes the Graph.apply test to collect the graph
triplets so it can assert about each triplet from a map method.
Author: William Benton <willb@redhat.com>
Closes#415 from willb/graphsuite-nop-fix and squashes the following commits:
0b63658 [William Benton] Ensure assertions in Graph.apply are asserted.
For your consideration: scalac currently notes a number of feature warnings during compilation:
```
[warn] there were 65 feature warning(s); re-run with -feature for details
```
Warnings are like:
```
[warn] /Users/srowen/Documents/spark/core/src/main/scala/org/apache/spark/SparkContext.scala:1261: implicit conversion method rddToPairRDDFunctions should be enabled
[warn] by making the implicit value scala.language.implicitConversions visible.
[warn] This can be achieved by adding the import clause 'import scala.language.implicitConversions'
[warn] or by setting the compiler option -language:implicitConversions.
[warn] See the Scala docs for value scala.language.implicitConversions for a discussion
[warn] why the feature should be explicitly enabled.
[warn] implicit def rddToPairRDDFunctions[K: ClassTag, V: ClassTag](rdd: RDD[(K, V)]) =
[warn] ^
```
scalac is suggesting that it's just best practice to explicitly enable certain language features by importing them where used.
This PR simply adds the imports it suggests (and squashes one other Java warning along the way). This leaves just deprecation warnings in the build.
Author: Sean Owen <sowen@cloudera.com>
Closes#404 from srowen/SPARK-1488 and squashes the following commits:
8598980 [Sean Owen] Quiet scalac warnings about language features by explicitly importing language features.
39bc831 [Sean Owen] Enable -feature in scalac to emit language feature warnings
stack these together in a commit else they show up chunk by chunk in different commits.
Author: Sandeep <sandeep@techaddict.me>
Closes#380 from techaddict/white_space and squashes the following commits:
b58f294 [Sandeep] Remove Unnecessary Whitespace's
[SPARK-729](https://spark-project.atlassian.net/browse/SPARK-729) concerns when free variables in closure arguments to transformations are captured. Currently, it is possible for closures to get the environment in which they are serialized (not the environment in which they are created). There are a few possible approaches to solving this problem and this PR will discuss some of them. The approach I took has the advantage of being simple, obviously correct, and minimally-invasive, but it preserves something that has been bothering me about Spark's closure handling, so I'd like to discuss an alternative and get some feedback on whether or not it is worth pursuing.
## What I did
The basic approach I took depends on the work I did for #143, and so this PR is based atop that. Specifically: #143 modifies `ClosureCleaner.clean` to preemptively determine whether or not closures are serializable immediately upon closure cleaning (rather than waiting for an job involving that closure to be scheduled). Thus non-serializable closure exceptions will be triggered by the line defining the closure rather than triggered where the closure is used.
Since the easiest way to determine whether or not a closure is serializable is to attempt to serialize it, the code in #143 is creating a serialized closure as part of `ClosureCleaner.clean`. `clean` currently modifies its argument, but the method in `SparkContext` that wraps it to return a value (a reference to the modified-in-place argument). This branch modifies `ClosureCleaner.clean` so that it returns a value: if it is cleaning a serializable closure, it returns the result of deserializing its serialized argument; therefore it is returning a closure with an environment captured at cleaning time. `SparkContext.clean` then returns the result of `ClosureCleaner.clean`, rather than a reference to its modified-in-place argument.
I've added tests for this behavior (777a1bc). The pull request as it stands, given the changes in #143, is nearly trivial. There is some overhead from deserializing the closure, but it is minimal and the benefit of obvious operational correctness (vs. a more sophisticated but harder-to-validate transformation in `ClosureCleaner`) seems pretty important. I think this is a fine way to solve this problem, but it's not perfect.
## What we might want to do
The thing that has been bothering me about Spark's handling of closures is that it seems like we should be able to statically ensure that cleaning and serialization happen exactly once for a given closure. If we serialize a closure in order to determine whether or not it is serializable, we should be able to hang on to the generated byte buffer and use it instead of re-serializing the closure later. By replacing closures with instances of a sum type that encodes whether or not a closure has been cleaned or serialized, we could handle clean, to-be-cleaned, and serialized closures separately with case matches. Here's a somewhat-concrete sketch (taken from my git stash) of what this might look like:
```scala
package org.apache.spark.util
import java.nio.ByteBuffer
import scala.reflect.ClassManifest
sealed abstract class ClosureBox[T] { def func: T }
final case class RawClosure[T](func: T) extends ClosureBox[T] {}
final case class CleanedClosure[T](func: T) extends ClosureBox[T] {}
final case class SerializedClosure[T](func: T, bytebuf: ByteBuffer) extends ClosureBox[T] {}
object ClosureBoxImplicits {
implicit def closureBoxFromFunc[T <: AnyRef](fun: T) = new RawClosure[T](fun)
}
```
With these types declared, we'd be able to change `ClosureCleaner.clean` to take a `ClosureBox[T=>U]` (possibly generated by implicit conversion) and return a `ClosureBox[T=>U]` (either a `CleanedClosure[T=>U]` or a `SerializedClosure[T=>U]`, depending on whether or not serializability-checking was enabled) instead of a `T=>U`. A case match could thus short-circuit cleaning or serializing closures that had already been cleaned or serialized (both in `ClosureCleaner` and in the closure serializer). Cleaned-and-serialized closures would be represented by a boxed tuple of the original closure and a serialized copy (complete with an environment quiesced at transformation time). Additional implicit conversions could convert from `ClosureBox` instances to the underlying function type where appropriate. Tracking this sort of state in the type system seems like the right thing to do to me.
### Why we might not want to do that
_It's pretty invasive._ Every function type used by every `RDD` subclass would have to change to reflect that they expected a `ClosureBox[T=>U]` instead of a `T=>U`. This obscures what's going on and is not a little ugly. Although I really like the idea of using the type system to enforce the clean-or-serialize once discipline, it might not be worth adding another layer of types (even if we could hide some of the extra boilerplate with judicious application of implicit conversions).
_It statically guarantees a property whose absence is unlikely to cause any serious problems as it stands._ It appears that all closures are currently dynamically cleaned once and it's not obvious that repeated closure-cleaning is likely to be a problem in the future. Furthermore, serializing closures is relatively cheap, so doing it once to check for serialization and once again to actually ship them across the wire doesn't seem like a big deal.
Taken together, these seem like a high price to pay for statically guaranteeing that closures are operated upon only once.
## Other possibilities
I felt like the serialize-and-deserialize approach was best due to its obvious simplicity. But it would be possible to do a more sophisticated transformation within `ClosureCleaner.clean`. It might also be possible for `clean` to modify its argument in a way so that whether or not a given closure had been cleaned would be apparent upon inspection; this would buy us some of the operational benefits of the `ClosureBox` approach but not the static cleanliness.
I'm interested in any feedback or discussion on whether or not the problems with the type-based approach indeed outweigh the advantage, as well as of approaches to this issue and to closure handling in general.
Author: William Benton <willb@redhat.com>
Closes#189 from willb/spark-729 and squashes the following commits:
f4cafa0 [William Benton] Stylistic changes and cleanups
b3d9c86 [William Benton] Fixed style issues in tests
9b56ce0 [William Benton] Added array-element capture test
97e9d91 [William Benton] Split closure-serializability failure tests
12ef6e3 [William Benton] Skip proactive closure capture for runJob
8ee3ee7 [William Benton] Predictable closure environment capture
12c63a7 [William Benton] Added tests for variable capture in closures
d6e8dd6 [William Benton] Don't check serializability of DStream transforms.
4ecf841 [William Benton] Make proactive serializability checking optional.
d8df3db [William Benton] Adds proactive closure-serializablilty checking
21b4b06 [William Benton] Test cases for SPARK-897.
d5947b3 [William Benton] Ensure assertions in Graph.apply are asserted.
This patch marks some existing classes as private[spark] and adds two types of API annotations:
- `EXPERIMENTAL API` = experimental user-facing module
- `DEVELOPER API - UNSTABLE` = developer-facing API that might change
There is some discussion of the different mechanisms for doing this here:
https://issues.apache.org/jira/browse/SPARK-1081
I was pretty aggressive with marking things private. Keep in mind that if we want to open something up in the future we can, but we can never reduce visibility.
A few notes here:
- In the past we've been inconsistent with the visiblity of the X-RDD classes. This patch marks them private whenever there is an existing function in RDD that can directly creat them (e.g. CoalescedRDD and rdd.coalesce()). One trade-off here is users can't subclass them.
- Noted that compression and serialization formats don't have to be wire compatible across versions.
- Compression codecs and serialization formats are semi-private as users typically don't instantiate them directly.
- Metrics sources are made private - user only interacts with them through Spark's reflection
Author: Patrick Wendell <pwendell@gmail.com>
Author: Andrew Or <andrewor14@gmail.com>
Closes#274 from pwendell/private-apis and squashes the following commits:
44179e4 [Patrick Wendell] Merge remote-tracking branch 'apache-github/master' into private-apis
042c803 [Patrick Wendell] spark.annotations -> spark.annotation
bfe7b52 [Patrick Wendell] Adding experimental for approximate counts
8d0c873 [Patrick Wendell] Warning in SparkEnv
99b223a [Patrick Wendell] Cleaning up annotations
e849f64 [Patrick Wendell] Merge pull request #2 from andrewor14/annotations
982a473 [Andrew Or] Generalize jQuery matching for non Spark-core API docs
a01c076 [Patrick Wendell] Merge pull request #1 from andrewor14/annotations
c1bcb41 [Andrew Or] DeveloperAPI -> DeveloperApi
0d48908 [Andrew Or] Comments and new lines (minor)
f3954e0 [Andrew Or] Add identifier tags in comments to work around scaladocs bug
99192ef [Andrew Or] Dynamically add badges based on annotations
824011b [Andrew Or] Add support for injecting arbitrary JavaScript to API docs
037755c [Patrick Wendell] Some changes after working with andrew or
f7d124f [Patrick Wendell] Small fixes
c318b24 [Patrick Wendell] Use CSS styles
e4c76b9 [Patrick Wendell] Logging
f390b13 [Patrick Wendell] Better visibility for workaround constructors
d6b0afd [Patrick Wendell] Small chang to existing constructor
403ba52 [Patrick Wendell] Style fix
870a7ba [Patrick Wendell] Work around for SI-8479
7fb13b2 [Patrick Wendell] Changes to UnionRDD and EmptyRDD
4a9e90c [Patrick Wendell] EXPERIMENTAL API --> EXPERIMENTAL
c581dce [Patrick Wendell] Changes after building against Shark.
8452309 [Patrick Wendell] Style fixes
1ed27d2 [Patrick Wendell] Formatting and coloring of badges
cd7a465 [Patrick Wendell] Code review feedback
2f706f1 [Patrick Wendell] Don't use floats
542a736 [Patrick Wendell] Small fixes
cf23ec6 [Patrick Wendell] Marking GraphX as alpha
d86818e [Patrick Wendell] Another naming change
5a76ed6 [Patrick Wendell] More visiblity clean-up
42c1f09 [Patrick Wendell] Using better labels
9d48cbf [Patrick Wendell] Initial pass
Another handful of small build changes to organize and standardize a bit, and avoid warnings:
- Update Maven plugin versions for good measure
- Since plugins need maven 3.0.4 already, require it explicitly (<3.0.4 had some bugs anyway)
- Use variables to define versions across dependencies where they should move in lock step
- ... and make this consistent between Maven/SBT
OK, I also updated the JIRA URL while I was at it here.
Author: Sean Owen <sowen@cloudera.com>
Closes#291 from srowen/SPARK-1387 and squashes the following commits:
461eca1 [Sean Owen] Couldn't resist also updating JIRA location to new one
c2d5cc5 [Sean Owen] Update plugins and Maven version; use variables consistently across Maven/SBT to define dependency versions that should stay in step.
This avoids a silent data corruption issue (https://spark-project.atlassian.net/browse/SPARK-1188) and has no performance impact by my measurements. It also simplifies the code. As far as I can tell the object re-use was nothing but premature optimization.
I did actual benchmarks for all the included changes, and there is no performance difference. I am not sure where to put the benchmarks. Does Spark not have a benchmark suite?
This is an example benchmark I did:
test("benchmark") {
val builder = new EdgePartitionBuilder[Int]
for (i <- (1 to 10000000)) {
builder.add(i.toLong, i.toLong, i)
}
val p = builder.toEdgePartition
p.map(_.attr + 1).iterator.toList
}
It ran for 10 seconds both before and after this change.
Author: Daniel Darabos <darabos.daniel@gmail.com>
Closes#276 from darabos/spark-1188 and squashes the following commits:
574302b [Daniel Darabos] Restore "manual" copying in EdgePartition.map(Iterator). Add comment to discourage novices like myself from trying to simplify the code.
4117a64 [Daniel Darabos] Revert EdgePartitionSuite.
4955697 [Daniel Darabos] Create a copy of the Edge objects in EdgeRDD.compute(). This avoids exposing the object re-use, while still enables the more efficient behavior for internal code.
4ec77f8 [Daniel Darabos] Add comments about object re-use to the affected functions.
2da5e87 [Daniel Darabos] Restore object re-use in EdgePartition.
0182f2b [Daniel Darabos] Do not re-use objects in the EdgePartition/EdgeTriplet iterators. This avoids a silent data corruption issue (SPARK-1188) and has no performance impact in my measurements. It also simplifies the code.
c55f52f [Daniel Darabos] Tests that reproduce the problems from SPARK-1188.
