Hi mateiz, I've created [SPARK-2277](https://issues.apache.org/jira/browse/SPARK-2277) to make TaskScheduler track hosts on each rack. Please help to review, thanks.

Author: Rui Li <rui.li@intel.com>

Closes #1212 from lirui-intel/trackHostOnRack and squashes the following commits:

2b4bd0f [Rui Li] SPARK-2277: refine UT
fbde838 [Rui Li] SPARK-2277: add UT
7bbe658 [Rui Li] SPARK-2277: rename the method
5e4ef62 [Rui Li] SPARK-2277: remove unnecessary import
79ac750 [Rui Li] SPARK-2277: make TaskScheduler track hosts on rack

JIRA issue: [SPARK-2119](https://issues.apache.org/jira/browse/SPARK-2119)

Essentially this PR fixed three issues to gain much better performance when reading large Parquet file off S3.

1. When reading the schema, fetching Parquet metadata from a part-file rather than the `_metadata` file

   The `_metadata` file contains metadata of all row groups, and can be very large if there are many row groups. Since schema information and row group metadata are coupled within a single Thrift object, we have to read the whole `_metadata` to fetch the schema. On the other hand, schema is replicated among footers of all part-files, which are fairly small.

1. Only add the root directory of the Parquet file rather than all the part-files to input paths

   HDFS API can automatically filter out all hidden files and underscore files (`_SUCCESS` & `_metadata`), there's no need to filter out all part-files and add them individually to input paths. What make it much worse is that, `FileInputFormat.listStatus()` calls `FileSystem.globStatus()` on each individual input path sequentially, each results a blocking remote S3 HTTP request.

1. Worked around [PARQUET-16](https://issues.apache.org/jira/browse/PARQUET-16)

   Essentially PARQUET-16 is similar to the above issue, and results lots of sequential `FileSystem.getFileStatus()` calls, which are further translated into a bunch of remote S3 HTTP requests.

   `FilteringParquetRowInputFormat` should be cleaned up once PARQUET-16 is fixed.

Below is the micro benchmark result. The dataset used is a S3 Parquet file consists of 3,793 partitions, about 110MB per partition in average. The benchmark is done with a 9-node AWS cluster.

- Creating a Parquet `SchemaRDD` (Parquet schema is fetched)

  ```scala
  val tweets = parquetFile(uri)
  ```

  - Before: 17.80s
  - After: 8.61s

- Fetching partition information

  ```scala
  tweets.getPartitions
  ```

  - Before: 700.87s
  - After: 21.47s

- Counting the whole file (both steps above are executed altogether)

  ```scala
  parquetFile(uri).count()
  ```

  - Before: ??? (haven't test yet)
  - After: 53.26s

Author: Cheng Lian <lian.cs.zju@gmail.com>

Closes #1370 from liancheng/faster-parquet and squashes the following commits:

94a2821 [Cheng Lian] Added comments about schema consistency
d2c4417 [Cheng Lian] Worked around PARQUET-16 to improve Parquet performance
1c0d1b9 [Cheng Lian] Accelerated Parquet schema retrieving
5bd3d29 [Cheng Lian] Fixed Parquet log level

This is a follow-up of #1359 with nullability narrowing.

Author: Takuya UESHIN <ueshin@happy-camper.st>

Closes #1426 from ueshin/issues/SPARK-2504 and squashes the following commits:

5157832 [Takuya UESHIN] Remove unnecessary white spaces.
80958ac [Takuya UESHIN] Fix nullability of Substring expression.

`Substring` including `null` literal cases could be added to `NullPropagation`.

Author: Takuya UESHIN <ueshin@happy-camper.st>

Closes #1428 from ueshin/issues/SPARK-2509 and squashes the following commits:

d9eb85f [Takuya UESHIN] Add Substring cases to NullPropagation.

[SPARK-2314][SQL] Override collect and take in JavaSchemaRDD, forwarding to SchemaRDD implementations.

Author: Aaron Staple <aaron.staple@gmail.com>

Closes #1421 from staple/SPARK-2314 and squashes the following commits:

73e04dc [Aaron Staple] [SPARK-2314] Override collect and take in JavaSchemaRDD, forwarding to SchemaRDD implementations.

http://legacy.python.org/dev/peps/pep-0008/
## Programming Recommendations
- Comparisons to singletons like None should always be done with is or is not, never the equality operators.

Author: Ken Takagiwa <ken@Kens-MacBook-Pro.local>

Closes #1422 from giwa/apache_master and squashes the following commits:

7b361f3 [Ken Takagiwa] follow pep8 None should be checked using is or is not

[SPARK-2500] Move the logInfo for registering BlockManager to BlockManagerMasterActor.register method

PR for SPARK-2500

Move the logInfo call for BlockManager to BlockManagerMasterActor.register instead of BlockManagerInfo constructor.

Previously the loginfo call for registering the registering a BlockManager is happening in the BlockManagerInfo constructor. This kind of confusing because the code could call "new BlockManagerInfo" without actually registering a BlockManager and could confuse when reading the log files.

Author: Henry Saputra <henry.saputra@gmail.com>

Closes #1424 from hsaputra/move_registerblockmanager_log_to_registration_method and squashes the following commits:

3370b4a [Henry Saputra] Move the loginfo for BlockManager to BlockManagerMasterActor.register instead of BlockManagerInfo constructor.

This reduces shuffle compression memory usage by 3x.

Author: Reynold Xin <rxin@apache.org>

Closes #1415 from rxin/snappy and squashes the following commits:

06c1a01 [Reynold Xin] SPARK-2469: Use Snappy (instead of LZF) for default shuffle compression codec.

JIRA ticket: https://issues.apache.org/jira/browse/SPARK-2498

Author: Zongheng Yang <zongheng.y@gmail.com>

Closes #1423 from concretevitamin/scala-ref-catalyst and squashes the following commits:

325a149 [Zongheng Yang] Synchronize on a lock when initializing data type objects in Catalyst.

Author: Michael Armbrust <michael@databricks.com>

Closes #1414 from marmbrus/exprIdResolution and squashes the following commits:

97b47bc [Michael Armbrust] Attribute equality comparisons should be done by exprId.

This replaces the Hive UDF for SUBSTR(ING) with an implementation in Catalyst
and adds tests to verify correct operation.

Author: William Benton <willb@redhat.com>

Closes #1359 from willb/internalSqlSubstring and squashes the following commits:

ccedc47 [William Benton] Fixed too-long line.
a30a037 [William Benton] replace view bounds with implicit parameters
ec35c80 [William Benton] Adds fixes from review:
4f3bfdb [William Benton] Added internal implementation of SQL SUBSTR()

[SPARK-2474][SQL] For a registered table in OverrideCatalog, the Analyzer failed to resolve references in the format of "tableName.fieldName"

Please refer to JIRA (https://issues.apache.org/jira/browse/SPARK-2474) for how to reproduce the problem and my understanding of the root cause.

Author: Yin Huai <huai@cse.ohio-state.edu>

Closes #1406 from yhuai/SPARK-2474 and squashes the following commits:

96b1627 [Yin Huai] Merge remote-tracking branch 'upstream/master' into SPARK-2474
af36d65 [Yin Huai] Fix comment.
be86ba9 [Yin Huai] Correct SQL console settings.
c43ad00 [Yin Huai] Wrap the relation in a Subquery named by the table name in OverrideCatalog.lookupRelation.
a5c2145 [Yin Huai] Support sql/console.

Author: Michael Armbrust <michael@databricks.com>

Closes #1396 from marmbrus/moreTests and squashes the following commits:

6660b60 [Michael Armbrust] Blacklist a test that requires DFS command.
8b6001c [Michael Armbrust] Add golden files.
ccd8f97 [Michael Armbrust] Whitelist more tests.

Author: Michael Armbrust <michael@databricks.com>

Closes #1411 from marmbrus/nestedRepeated and squashes the following commits:

044fa09 [Michael Armbrust] Fix parsing of repeated, nested data access.

The assembly jar (built by sbt) doesn't include jets3t if we set it to runtime only, but I don't know whether it was set this way for a particular reason.

CC: srowen ScrapCodes

Author: Xiangrui Meng <meng@databricks.com>

Closes #1402 from mengxr/jets3t and squashes the following commits:

bfa2d17 [Xiangrui Meng] remove runtime scope for jets3t

Author: Reynold Xin <rxin@apache.org>

Closes #1417 from rxin/lz4 and squashes the following commits:

472f6a1 [Reynold Xin] Set the proper default.
9cf0b2f [Reynold Xin] Added LZ4 to compression codec in configuration page.

Author: witgo <witgo@qq.com>

Closes #1112 from witgo/SPARK-1291 and squashes the following commits:

6022bcd [witgo] review commit
1fbb925 [witgo] add addAmIpFilter to yarn alpha
210299c [witgo] review commit
1b92a07 [witgo] review commit
6896586 [witgo] Add comments to addWebUIFilter
3e9630b [witgo] review commit
142ee29 [witgo] review commit
1fe7710 [witgo] Link the spark UI to RM ui in yarn-client mode

Author: witgo <witgo@qq.com>

Closes #1404 from witgo/run-tests and squashes the following commits:

f703aee [witgo] fix Note: implicit method fromPairDStream is not applicable here because it comes after the application point and it lacks an explicit result type
2944f51 [witgo] Remove "NOTE: SPARK_YARN is deprecated, please use -Pyarn flag"
ef59c70 [witgo] fix Note: implicit method fromPairDStream is not applicable here because it comes after the application point and it lacks an explicit result type
6cefee5 [witgo] Remove "NOTE: SPARK_YARN is deprecated, please use -Pyarn flag"

Author: William Benton <willb@redhat.com>

Closes #1419 from willb/reformat-2486 and squashes the following commits:

2676231 [William Benton] Reformat multi-line closure argument.

Adding two classes:
1) MulticlassMetrics implements various multiclass evaluation metrics
2) MulticlassMetricsSuite implements unit tests for MulticlassMetrics

Author: Alexander Ulanov <nashb@yandex.ru>
Author: unknown <ulanov@ULANOV1.emea.hpqcorp.net>
Author: Xiangrui Meng <meng@databricks.com>

Closes #1155 from avulanov/master and squashes the following commits:

2eae80f [Alexander Ulanov] Merge pull request #1 from mengxr/avulanov-master
5ebeb08 [Xiangrui Meng] minor updates
79c3555 [Alexander Ulanov] Addressing reviewers comments mengxr
0fa9511 [Alexander Ulanov] Addressing reviewers comments mengxr
f0dadc9 [Alexander Ulanov] Addressing reviewers comments mengxr
4811378 [Alexander Ulanov] Removing println
87fb11f [Alexander Ulanov] Addressing reviewers comments mengxr. Added confusion matrix
e3db569 [Alexander Ulanov] Addressing reviewers comments mengxr. Added true positive rate and false positive rate. Test suite code style.
a7e8bf0 [Alexander Ulanov] Addressing reviewers comments mengxr
c3a77ad [Alexander Ulanov] Addressing reviewers comments mengxr
e2c91c3 [Alexander Ulanov] Fixes to mutliclass metics
d5ce981 [unknown] Comments about Double
a5c8ba4 [unknown] Unit tests. Class rename
fcee82d [unknown] Unit tests. Class rename
d535d62 [unknown] Multiclass evaluation

(cherry picked from commit 401083be9f010f95110a819a49837ecae7d9c4ec)
Signed-off-by: Reynold Xin <rxin@apache.org>

Instead of using prependOne currently in GeneralizedLinearAlgorithm, we would like to use appendBias for 1) keeping the indices of original training set unchanged by adding the intercept into the last element of vector and 2) using the same public API for consistently adding intercept.

Author: DB Tsai <dbtsai@alpinenow.com>

Closes #1410 from dbtsai/SPARK-2477_intercept_with_appendBias and squashes the following commits:

011432c [DB Tsai] From Alpine Data Labs

Based on Greg Bowyer's patch from JIRA https://issues.apache.org/jira/browse/SPARK-2399

Author: Reynold Xin <rxin@apache.org>

Closes #1416 from rxin/lz4 and squashes the following commits:

6c8fefe [Reynold Xin] Fixed typo.
8a14d38 [Reynold Xin] [SPARK-2399] Add support for LZ4 compression.

When completedDrivers number exceeds the threshold, the first Max(spark.deploy.retainedDrivers, 1) will be discarded.

Author: lianhuiwang <lianhuiwang09@gmail.com>

Closes #1114 from lianhuiwang/retained-drivers and squashes the following commits:

8789418 [lianhuiwang] discarded exceeded completedDrivers

Author: Michael Armbrust <michael@databricks.com>

Closes #1412 from marmbrus/lockHiveClient and squashes the following commits:

4bc9d5a [Michael Armbrust] protected[hive]
22e9177 [Michael Armbrust] Add comments.
7aa8554 [Michael Armbrust] Don't lock on hive's object.
a6edc5f [Michael Armbrust] Lock usage of hive client.

When running jobs with YARN Cluster mode and using HistoryServer, the files in the Staging Directory (~/.sparkStaging on HDFS) cannot be deleted.
HistoryServer uses directory where event log is written, and the directory is represented as a instance of o.a.h.f.FileSystem created by using FileSystem.get.

On the other hand, ApplicationMaster has a instance named fs, which also created by using FileSystem.get.

FileSystem.get returns cached same instance when URI passed to the method represents same file system and the method is called by same user.
Because of the behavior, when the directory for event log is on HDFS, fs of ApplicationMaster and fileSystem of FileLogger is same instance.
When shutting down ApplicationMaster, fileSystem.close is called in FileLogger#stop, which is invoked by SparkContext#stop indirectly.

And ApplicationMaster#cleanupStagingDir also called by JVM shutdown hook. In this method, fs.delete(stagingDirPath) is invoked.
Because fs.delete in ApplicationMaster is called after fileSystem.close in FileLogger, fs.delete fails and results not deleting files in the staging directory.

I think, calling fileSystem.delete is not needed.

Author: Kousuke Saruta <sarutak@oss.nttdata.co.jp>

Closes #1326 from sarutak/SPARK-2390 and squashes the following commits:

10e1a88 [Kousuke Saruta] Removed fileSystem.close from FileLogger.scala not to prevent any other FileSystem operation

groupBy()/groupByKey() is notorious for being a very convenient API that can lead to poor performance when used incorrectly.

This PR just makes it clear that users should be cautious not to rely on this API when they really want a different (more performant) one, such as reduceByKey().

(Note that one source of confusion is the name; this groupBy() is not the same as a SQL GROUP-BY, which is used for aggregation and is more similar in nature to Spark's reduceByKey().)

Author: Aaron Davidson <aaron@databricks.com>

Closes #1380 from aarondav/warning and squashes the following commits:

f60da39 [Aaron Davidson] Give better advice
d0afb68 [Aaron Davidson] Add/increase severity of warning in documentation of groupBy()

When running Spark under certain instrumenting profilers,
Utils.getCallSite could crash with an NPE.  This commit
makes it more resilient to failures occurring while inspecting
stack frames.

Author: William Benton <willb@redhat.com>

Closes #1413 from willb/spark-2486 and squashes the following commits:

b7c0274 [William Benton] Use explicit null checks instead of Try()
0f0c1ae [William Benton] Utils.getCallSite is now resilient to bogus frames

Author: Takuya UESHIN <ueshin@happy-camper.st>

Closes #1398 from ueshin/issues/SPARK-2467 and squashes the following commits:

7f01d58 [Takuya UESHIN] Revert SparkBuild to publish-local to both .m2 and .ivy2.

Note that this commit changes the semantics when loading in data that was created with prior versions of Spark SQL.  Before, we were writing out strings as Binary data without adding any other annotations. Thus, when data is read in from prior versions, data that was StringType will now become BinaryType.  Users that need strings can CAST that column to a String.  It was decided that while this breaks compatibility, it does make us compatible with other systems (Hive, Thrift, etc) and adds support for Binary data, so this is the right decision long term.

To support `BinaryType`, the following changes are needed:
- Make `StringType` use `OriginalType.UTF8`
- Add `BinaryType` using `PrimitiveTypeName.BINARY` without `OriginalType`

Author: Takuya UESHIN <ueshin@happy-camper.st>

Closes #1373 from ueshin/issues/SPARK-2446 and squashes the following commits:

ecacb92 [Takuya UESHIN] Add BinaryType support to Parquet I/O.
616e04a [Takuya UESHIN] Make StringType use OriginalType.UTF8.

Because submitting tasks and registering executors are asynchronous, in most situation, early stages' tasks run without preferred locality.

A simple solution is sleeping few seconds in application, so that executors have enough time to register.

The PR add 2 configuration properties to make TaskScheduler submit tasks after a few of executors have been registered.

\# Submit tasks only after (registered executors / total executors) arrived the ratio, default value is 0
spark.scheduler.minRegisteredExecutorsRatio = 0.8

\# Whatever minRegisteredExecutorsRatio is arrived, submit tasks after the maxRegisteredWaitingTime(millisecond), default value is 30000
spark.scheduler.maxRegisteredExecutorsWaitingTime = 5000

Author: li-zhihui <zhihui.li@intel.com>

Closes #900 from li-zhihui/master and squashes the following commits:

b9f8326 [li-zhihui] Add logs & edit docs
1ac08b1 [li-zhihui] Add new configs to user docs
22ead12 [li-zhihui] Move waitBackendReady to postStartHook
c6f0522 [li-zhihui] Bug fix: numExecutors wasn't set & use constant DEFAULT_NUMBER_EXECUTORS
4d6d847 [li-zhihui] Move waitBackendReady to TaskSchedulerImpl.start & some code refactor
0ecee9a [li-zhihui] Move waitBackendReady from DAGScheduler.submitStage to TaskSchedulerImpl.submitTasks
4261454 [li-zhihui] Add docs for new configs & code style
ce0868a [li-zhihui] Code style, rename configuration property name of minRegisteredRatio & maxRegisteredWaitingTime
6cfb9ec [li-zhihui] Code style, revert default minRegisteredRatio of yarn to 0, driver get --num-executors in yarn/alpha
812c33c [li-zhihui] Fix driver lost --num-executors option in yarn-cluster mode
e7b6272 [li-zhihui] support yarn-cluster
37f7dc2 [li-zhihui] support yarn mode(percentage style)
3f8c941 [li-zhihui] submit stage after (configured ratio of) executors have been registered

This fix obtains a comparable performance boost as [PR #1390](https://github.com/apache/spark/pull/1390) by moving an array update and deserializer initialization out of a potentially very long loop. Suggested by yhuai. The below results are updated for this fix.

## Benchmarks
Generated a local text file with 10M rows of simple key-value pairs. The data is loaded as a table through Hive. Results are obtained on my local machine using hive/console.

Without the fix:

Type | Non-partitioned | Partitioned (1 part)
------------ | ------------ | -------------
First run | 9.52s end-to-end (1.64s Spark job) | 36.6s (28.3s)
Stablized runs | 1.21s (1.18s) | 27.6s (27.5s)

With this fix:

Type | Non-partitioned | Partitioned (1 part)
------------ | ------------ | -------------
First run | 9.57s (1.46s) | 11.0s (1.69s)
Stablized runs | 1.13s (1.10s) | 1.23s (1.19s)

Author: Zongheng Yang <zongheng.y@gmail.com>

Closes #1408 from concretevitamin/slow-read-2 and squashes the following commits:

d86e437 [Zongheng Yang] Move update & initialization out of potentially long loop.

Just move some test suite to corresponding package

Author: Daoyuan <daoyuan.wang@intel.com>

Closes #1401 from adrian-wang/movetestfiles and squashes the following commits:

d1a6803 [Daoyuan] move some test file to match src code

Author: Prashant Sharma <prashant.s@imaginea.com>

Closes #1354 from ScrapCodes/pep8-comp-1 and squashes the following commits:

9858ea8 [Prashant Sharma] Code Review
d8851b7 [Prashant Sharma] Found # noqa works even inside comment blocks. Not sure if it works with all versions of python.
10c0cef [Prashant Sharma] Made rdd.py pep8 complaint by using Autopep8 and a little manual tweaking.

(Just made a PR for this, mengxr was the reporter of:)

MLlib has sample data under serveral folders:
1) data/mllib
2) data/
3) mllib/data/*
Per previous discussion with Matei Zaharia, we want to put them under `data/mllib` and clean outdated files.

Author: Sean Owen <sowen@cloudera.com>

Closes #1394 from srowen/SPARK-2363 and squashes the following commits:

54313dd [Sean Owen] Move ML example data from /mllib/data/ and /data/ into /data/mllib/

Apologies if there's an already-discussed reason I missed for why this doesn't make sense.

Author: Sandy Ryza <sandy@cloudera.com>

Closes #1389 from sryza/sandy-spark-2462 and squashes the following commits:

2e5e201 [Sandy Ryza] SPARK-2462.  Make Vector.apply public.

Reuse byte buffers when creating unique attributes for multiple instances of an InMemoryRelation in a single query plan.

Author: Michael Armbrust <michael@databricks.com>

Closes #1332 from marmbrus/doubleCache and squashes the following commits:

4a19609 [Michael Armbrust] Clean up concurrency story by calculating buffersn the constructor.
b39c931 [Michael Armbrust] Allocations are kind of a side effect.
f67eff7 [Michael Armbrust] Reusue same byte buffers when creating new instance of InMemoryRelation

Author: Michael Armbrust <michael@databricks.com>

Closes #1366 from marmbrus/partialDistinct and squashes the following commits:

12a31ab [Michael Armbrust] Add more efficient distinct operator.

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