It's a follow-up of https://github.com/apache/spark/pull/5154, we can speed up scala udf evaluation by create type converter in advance.
Author: Wenchen Fan <cloud0fan@outlook.com>
Closes#6182 from cloud-fan/tmp and squashes the following commits:
241cfe9 [Wenchen Fan] use converter in ScalaUdf
(cherry picked from commit 2f22424e9f)
Signed-off-by: Yin Huai <yhuai@databricks.com>
JIRA: https://issues.apache.org/jira/browse/SPARK-7098
The WHERE clause with timstamp shows inconsistent results. This pr fixes it.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#5682 from viirya/consistent_timestamp and squashes the following commits:
171445a [Liang-Chi Hsieh] Merge remote-tracking branch 'upstream/master' into consistent_timestamp
4e98520 [Liang-Chi Hsieh] Make the WHERE clause with timestamp show consistent result.
(cherry picked from commit f9705d4613)
Signed-off-by: Reynold Xin <rxin@databricks.com>
Add an `explode` function for dataframes and modify the analyzer so that single table generating functions can be present in a select clause along with other expressions. There are currently the following restrictions:
- only top level TGFs are allowed (i.e. no `select(explode('list) + 1)`)
- only one may be present in a single select to avoid potentially confusing implicit Cartesian products.
TODO:
- [ ] Python
Author: Michael Armbrust <michael@databricks.com>
Closes#6107 from marmbrus/explodeFunction and squashes the following commits:
7ee2c87 [Michael Armbrust] whitespace
6f80ba3 [Michael Armbrust] Update dataframe.py
c176c89 [Michael Armbrust] Merge remote-tracking branch 'origin/master' into explodeFunction
81b5da3 [Michael Armbrust] style
d3faa05 [Michael Armbrust] fix self join case
f9e1e3e [Michael Armbrust] fix python, add since
4f0d0a9 [Michael Armbrust] Merge remote-tracking branch 'origin/master' into explodeFunction
e710fe4 [Michael Armbrust] add java and python
52ca0dc [Michael Armbrust] [SPARK-7548][SQL] Add explode function for dataframes.
(cherry picked from commit 6d0633e3ec)
Signed-off-by: Michael Armbrust <michael@databricks.com>
A follow-up of https://github.com/apache/spark/pull/5624
Author: Wenchen Fan <cloud0fan@outlook.com>
Closes#6142 from cloud-fan/tmp and squashes the following commits:
971a92b [Wenchen Fan] use plan instead of execute
24c5ffe [Wenchen Fan] rename apply
(cherry picked from commit f2cd00be35)
Signed-off-by: Reynold Xin <rxin@databricks.com>
for example:
table: src(key string, value string)
sql: with v1 as(select key, count(value) over (partition by key) cnt_val from src), v2 as(select v1.key, v1_lag.cnt_val from v1, v1 v1_lag where v1.key = v1_lag.key) select * from v2 limit 5;
then will analyze fail when resolving conflicting references in Join:
'Limit 5
'Project [*]
'Subquery v2
'Project ['v1.key,'v1_lag.cnt_val]
'Filter ('v1.key = 'v1_lag.key)
'Join Inner, None
Subquery v1
Project [key#95,cnt_val#94L]
Window [key#95,value#96], [HiveWindowFunction#org.apache.hadoop.hive.ql.udf.generic.GenericUDAFCount(value#96) WindowSpecDefinition [key#95], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING AS cnt_val#94L], WindowSpecDefinition [key#95], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
Project [key#95,value#96]
MetastoreRelation default, src, None
Subquery v1_lag
Subquery v1
Project [key#97,cnt_val#94L]
Window [key#97,value#98], [HiveWindowFunction#org.apache.hadoop.hive.ql.udf.generic.GenericUDAFCount(value#98) WindowSpecDefinition [key#97], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING AS cnt_val#94L], WindowSpecDefinition [key#97], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
Project [key#97,value#98]
MetastoreRelation default, src, None
Conflicting attributes: cnt_val#94L
Author: linweizhong <linweizhong@huawei.com>
Closes#6114 from Sephiroth-Lin/spark-7595 and squashes the following commits:
f8f2637 [linweizhong] Add unit test
dfe9169 [linweizhong] Handle windowExpression with self join
(cherry picked from commit 13e652b61a)
Signed-off-by: Michael Armbrust <michael@databricks.com>
JavaTypeInference into catalyst
types.DateUtils into catalyst
CacheManager into execution
DefaultParserDialect into catalyst
Author: Reynold Xin <rxin@databricks.com>
Closes#6108 from rxin/sql-rename and squashes the following commits:
3fc9613 [Reynold Xin] Fixed import ordering.
83d9ff4 [Reynold Xin] Fixed codegen tests.
e271e86 [Reynold Xin] mima
f4e24a6 [Reynold Xin] [SQL] Move some classes into packages that are more appropriate.
(cherry picked from commit e683182c3e)
Signed-off-by: Michael Armbrust <michael@databricks.com>
Optimize the case of `project(_, sort)` , a example is:
`select key from (select * from testData order by key) t`
before this PR:
```
== Parsed Logical Plan ==
'Project ['key]
'Subquery t
'Sort ['key ASC], true
'Project [*]
'UnresolvedRelation [testData], None
== Analyzed Logical Plan ==
Project [key#0]
Subquery t
Sort [key#0 ASC], true
Project [key#0,value#1]
Subquery testData
LogicalRDD [key#0,value#1], MapPartitionsRDD[1]
== Optimized Logical Plan ==
Project [key#0]
Sort [key#0 ASC], true
LogicalRDD [key#0,value#1], MapPartitionsRDD[1]
== Physical Plan ==
Project [key#0]
Sort [key#0 ASC], true
Exchange (RangePartitioning [key#0 ASC], 5), []
PhysicalRDD [key#0,value#1], MapPartitionsRDD[1]
```
after this PR
```
== Parsed Logical Plan ==
'Project ['key]
'Subquery t
'Sort ['key ASC], true
'Project [*]
'UnresolvedRelation [testData], None
== Analyzed Logical Plan ==
Project [key#0]
Subquery t
Sort [key#0 ASC], true
Project [key#0,value#1]
Subquery testData
LogicalRDD [key#0,value#1], MapPartitionsRDD[1]
== Optimized Logical Plan ==
Sort [key#0 ASC], true
Project [key#0]
LogicalRDD [key#0,value#1], MapPartitionsRDD[1]
== Physical Plan ==
Sort [key#0 ASC], true
Exchange (RangePartitioning [key#0 ASC], 5), []
Project [key#0]
PhysicalRDD [key#0,value#1], MapPartitionsRDD[1]
```
with this rule we will first do column pruning on the table and then do sorting.
Author: scwf <wangfei1@huawei.com>
This patch had conflicts when merged, resolved by
Committer: Michael Armbrust <michael@databricks.com>
Closes#5838 from scwf/pruning and squashes the following commits:
b00d833 [scwf] address michael's comment
e230155 [scwf] fix tests failure
b09b895 [scwf] improve column pruning
(cherry picked from commit 59250fe514)
Signed-off-by: Michael Armbrust <michael@databricks.com>
Some third-party UDTF extensions generate additional rows in the "GenericUDTF.close()" method, which is supported / documented by Hive.
https://cwiki.apache.org/confluence/display/Hive/DeveloperGuide+UDTF
However, Spark SQL ignores the "GenericUDTF.close()", and it causes bug while porting job from Hive to Spark SQL.
Author: Cheng Hao <hao.cheng@intel.com>
Closes#5383 from chenghao-intel/udtf_close and squashes the following commits:
98b4e4b [Cheng Hao] Support UDTF.close
(cherry picked from commit 0da254fb29)
Signed-off-by: Cheng Lian <lian@databricks.com>
Author: Wenchen Fan <cloud0fan@outlook.com>
Closes#5831 from cloud-fan/7276 and squashes the following commits:
ee4a1e1 [Wenchen Fan] fix rebase mistake
a3b565d [Wenchen Fan] refactor
99deb5d [Wenchen Fan] add test
f1f67ad [Wenchen Fan] fix 7276
(cherry picked from commit 4e290522c2)
Signed-off-by: Michael Armbrust <michael@databricks.com>
Author: Wenchen Fan <cloud0fan@outlook.com>
Closes#6079 from cloud-fan/unapply and squashes the following commits:
40da442 [Wenchen Fan] one more
7d90a05 [Wenchen Fan] cleanup unapply in DataTypes
(cherry picked from commit 831504cf6b)
Signed-off-by: Reynold Xin <rxin@databricks.com>
As a follow-up to https://github.com/apache/spark/pull/5944
Author: Reynold Xin <rxin@databricks.com>
Closes#6064 from rxin/jointype-better-error and squashes the following commits:
7629bf7 [Reynold Xin] [SQL] Show better error messages for incorrect join types in DataFrames.
(cherry picked from commit 4f4dbb030c)
Signed-off-by: Reynold Xin <rxin@databricks.com>
It's the first step: generalize UnresolvedGetField to support all map, struct, and array
TODO: add `apply` in Scala and `__getitem__` in Python, and unify the `getItem` and `getField` methods to one single API(or should we keep them for compatibility?).
Author: Wenchen Fan <cloud0fan@outlook.com>
Closes#5744 from cloud-fan/generalize and squashes the following commits:
715c589 [Wenchen Fan] address comments
7ea5b31 [Wenchen Fan] fix python test
4f0833a [Wenchen Fan] add python test
f515d69 [Wenchen Fan] add apply method and test cases
8df6199 [Wenchen Fan] fix python test
239730c [Wenchen Fan] fix test compile
2a70526 [Wenchen Fan] use _bin_op in dataframe.py
6bf72bc [Wenchen Fan] address comments
3f880c3 [Wenchen Fan] add java doc
ab35ab5 [Wenchen Fan] fix python test
b5961a9 [Wenchen Fan] fix style
c9d85f5 [Wenchen Fan] generalize UnresolvedGetField to support all map, struct, and array
(cherry picked from commit 2d05f325dc)
Signed-off-by: Michael Armbrust <michael@databricks.com>
Added a new batch named `Substitution` before `Resolution` batch. The motivation for this is there are kind of cases we want to do some substitution on the parsed logical plan before resolve it.
Consider this two cases:
1 CTE, for cte we first build a row logical plan
```
'With Map(q1 -> 'Subquery q1
'Project ['key]
'UnresolvedRelation [src], None)
'Project [*]
'Filter ('key = 5)
'UnresolvedRelation [q1], None
```
In `With` logicalplan here is a map stored the (`q1-> subquery`), we want first take off the with command and substitute the `q1` of `UnresolvedRelation` by the `subquery`
2 Another example is Window function, in window function user may define some windows, we also need substitute the window name of child by the concrete window. this should also done in the Substitution batch.
Author: wangfei <wangfei1@huawei.com>
Closes#5776 from scwf/addbatch and squashes the following commits:
d4b962f [wangfei] added WindowsSubstitution
70f6932 [wangfei] Merge branch 'master' of https://github.com/apache/spark into addbatch
ecaeafb [wangfei] address yhuai's comments
553005a [wangfei] fix test case
0c54798 [wangfei] address comments
29aaaaf [wangfei] fix compile
1c9a092 [wangfei] added Substitution bastch
(cherry picked from commit f496bf3c53)
Signed-off-by: Yin Huai <yhuai@databricks.com>
This PR switches Spark SQL's Hive support to use the isolated hive client interface introduced by #5851, instead of directly interacting with the client. By using this isolated client we can now allow users to dynamically configure the version of Hive that they are connecting to by setting `spark.sql.hive.metastore.version` without the need recompile. This also greatly reduces the surface area for our interaction with the hive libraries, hopefully making it easier to support other versions in the future.
Jars for the desired hive version can be configured using `spark.sql.hive.metastore.jars`, which accepts the following options:
- a colon-separated list of jar files or directories for hive and hadoop.
- `builtin` - attempt to discover the jars that were used to load Spark SQL and use those. This
option is only valid when using the execution version of Hive.
- `maven` - download the correct version of hive on demand from maven.
By default, `builtin` is used for Hive 13.
This PR also removes the test step for building against Hive 12, as this will no longer be required to talk to Hive 12 metastores. However, the full removal of the Shim is deferred until a later PR.
Remaining TODOs:
- Remove the Hive Shims and inline code for Hive 13.
- Several HiveCompatibility tests are not yet passing.
- `nullformatCTAS` - As detailed below, we now are handling CTAS parsing ourselves instead of hacking into the Hive semantic analyzer. However, we currently only handle the common cases and not things like CTAS where the null format is specified.
- `combine1` now leaks state about compression somehow, breaking all subsequent tests. As such we currently add it to the blacklist
- `part_inherit_tbl_props` and `part_inherit_tbl_props_with_star` do not work anymore. We are correctly propagating the information
- "load_dyn_part14.*" - These tests pass when run on their own, but fail when run with all other tests. It seems our `RESET` mechanism may not be as robust as it used to be?
Other required changes:
- `CreateTableAsSelect` no longer carries parts of the HiveQL AST with it through the query execution pipeline. Instead, we parse CTAS during the HiveQL conversion and construct a `HiveTable`. The full parsing here is not yet complete as detailed above in the remaining TODOs. Since the operator is Hive specific, it is moved to the hive package.
- `Command` is simplified to be a trait that simply acts as a marker for a LogicalPlan that should be eagerly evaluated.
Author: Michael Armbrust <michael@databricks.com>
Closes#5876 from marmbrus/useIsolatedClient and squashes the following commits:
258d000 [Michael Armbrust] really really correct path handling
e56fd4a [Michael Armbrust] getAbsolutePath
5a259f5 [Michael Armbrust] fix typos
81bb366 [Michael Armbrust] comments from vanzin
5f3945e [Michael Armbrust] Merge remote-tracking branch 'origin/master' into useIsolatedClient
4b5cd41 [Michael Armbrust] yin's comments
f5de7de [Michael Armbrust] cleanup
11e9c72 [Michael Armbrust] better coverage in versions suite
7e8f010 [Michael Armbrust] better error messages and jar handling
e7b3941 [Michael Armbrust] more permisive checking for function registration
da91ba7 [Michael Armbrust] Merge remote-tracking branch 'origin/master' into useIsolatedClient
5fe5894 [Michael Armbrust] fix serialization suite
81711c4 [Michael Armbrust] Initial support for running without maven
1d8ae44 [Michael Armbrust] fix final tests?
1c50813 [Michael Armbrust] more comments
a3bee70 [Michael Armbrust] Merge remote-tracking branch 'origin/master' into useIsolatedClient
a6f5df1 [Michael Armbrust] style
ab07f7e [Michael Armbrust] WIP
4d8bf02 [Michael Armbrust] Remove hive 12 compilation
8843a25 [Michael Armbrust] [SPARK-6908] [SQL] Use isolated Hive client
(cherry picked from commit cd1d4110cf)
Signed-off-by: Yin Huai <yhuai@databricks.com>
Avoid translating to CaseWhen and evaluate the key expression many times.
Author: Wenchen Fan <cloud0fan@outlook.com>
Closes#5979 from cloud-fan/condition and squashes the following commits:
3ce54e1 [Wenchen Fan] add CaseKeyWhen
(cherry picked from commit 35f0173b8f)
Signed-off-by: Michael Armbrust <michael@databricks.com>
Go through the context classloader when reflecting on user types in ScalaReflection.
Replaced calls to `typeOf` with `typeTag[T].in(mirror)`. The convenience method assumes
all types can be found in the classloader that loaded scala-reflect (the primordial
classloader). This assumption is not valid in all contexts (sbt console, Eclipse launchers).
Fixed SPARK-5281
Author: Iulian Dragos <jaguarul@gmail.com>
Closes#5981 from dragos/issue/mirrors-missing-requirement-error and squashes the following commits:
d103e70 [Iulian Dragos] Go through the context classloader when reflecting on user types in ScalaReflection
(cherry picked from commit 937ba798c5)
Signed-off-by: Michael Armbrust <michael@databricks.com>
`Star` and `MultiAlias` just used in `analyzer` and them will be substituted after analyze, So just like `Alias` they do not need extend `Attribute`
Author: scwf <wangfei1@huawei.com>
Closes#5928 from scwf/attribute and squashes the following commits:
73a0560 [scwf] star and multialias do not need extend attribute
(cherry picked from commit 97d1182af6)
Signed-off-by: Michael Armbrust <michael@databricks.com>
Address marmbrus and scwf's comments in #5604.
Author: Yin Huai <yhuai@databricks.com>
Closes#5945 from yhuai/windowFollowup and squashes the following commits:
0ef879d [Yin Huai] Add collectFirst to TreeNode.
2373968 [Yin Huai] wip
4a16df9 [Yin Huai] Address minor comments for [SPARK-1442].
(cherry picked from commit 5784c8d955)
Signed-off-by: Michael Armbrust <michael@databricks.com>
This patch comprises of a few related pieces of work:
* Schema inference is performed directly on the JSON token stream
* `String => Row` conversion populate Spark SQL structures without intermediate types
* Projection pushdown is implemented via CatalystScan for DataFrame queries
* Support for the legacy parser by setting `spark.sql.json.useJacksonStreamingAPI` to `false`
Performance improvements depend on the schema and queries being executed, but it should be faster across the board. Below are benchmarks using the last.fm Million Song dataset:
```
Command | Baseline | Patched
---------------------------------------------------|----------|--------
import sqlContext.implicits._ | |
val df = sqlContext.jsonFile("/tmp/lastfm.json") | 70.0s | 14.6s
df.count() | 28.8s | 6.2s
df.rdd.count() | 35.3s | 21.5s
df.where($"artist" === "Robert Hood").collect() | 28.3s | 16.9s
```
To prepare this dataset for benchmarking, follow these steps:
```
# Fetch the datasets from http://labrosa.ee.columbia.edu/millionsong/lastfm
wget http://labrosa.ee.columbia.edu/millionsong/sites/default/files/lastfm/lastfm_test.zip \
http://labrosa.ee.columbia.edu/millionsong/sites/default/files/lastfm/lastfm_train.zip
# Decompress and combine, pipe through `jq -c` to ensure there is one record per line
unzip -p lastfm_test.zip lastfm_train.zip | jq -c . > lastfm.json
```
Author: Nathan Howell <nhowell@godaddy.com>
Closes#5801 from NathanHowell/json-performance and squashes the following commits:
26fea31 [Nathan Howell] Recreate the baseRDD each for each scan operation
a7ebeb2 [Nathan Howell] Increase coverage of inserts into a JSONRelation
e06a1dd [Nathan Howell] Add comments to the `useJacksonStreamingAPI` config flag
6822712 [Nathan Howell] Split up JsonRDD2 into multiple objects
fa8234f [Nathan Howell] Wrap long lines
b31917b [Nathan Howell] Rename `useJsonRDD2` to `useJacksonStreamingAPI`
15c5d1b [Nathan Howell] JSONRelation's baseRDD need not be lazy
f8add6e [Nathan Howell] Add comments on lack of support for precision and scale DecimalTypes
fa0be47 [Nathan Howell] Remove unused default case in the field parser
80dba17 [Nathan Howell] Add comments regarding null handling and empty strings
842846d [Nathan Howell] Point the empty schema inference test at JsonRDD2
ab6ee87 [Nathan Howell] Add projection pushdown support to JsonRDD/JsonRDD2
f636c14 [Nathan Howell] Enable JsonRDD2 by default, add a flag to switch back to JsonRDD
0bbc445 [Nathan Howell] Improve JSON parsing and type inference performance
7ca70c1 [Nathan Howell] Eliminate arrow pattern, replace with pattern matches
(cherry picked from commit 2d6612cc8b)
Signed-off-by: Yin Huai <yhuai@databricks.com>
Adding more information about the implementation...
This PR is adding the support of window functions to Spark SQL (specifically OVER and WINDOW clause). For every expression having a OVER clause, we use a WindowExpression as the container of a WindowFunction and the corresponding WindowSpecDefinition (the definition of a window frame, i.e. partition specification, order specification, and frame specification appearing in a OVER clause).
# Implementation #
The high level work flow of the implementation is described as follows.
* Query parsing: In the query parse process, all WindowExpressions are originally placed in the projectList of a Project operator or the aggregateExpressions of an Aggregate operator. It makes our changes to simple and keep all of parsing rules for window functions at a single place (nodesToWindowSpecification). For the WINDOWclause in a query, we use a WithWindowDefinition as the container as the mapping from the name of a window specification to a WindowSpecDefinition. This changes is similar with our common table expression support.
* Analysis: The query analysis process has three steps for window functions.
* Resolve all WindowSpecReferences by replacing them with WindowSpecReferences according to the mapping table stored in the node of WithWindowDefinition.
* Resolve WindowFunctions in the projectList of a Project operator or the aggregateExpressions of an Aggregate operator. For this PR, we use Hive's functions for window functions because we will have a major refactoring of our internal UDAFs and it is better to switch our UDAFs after that refactoring work.
* Once we have resolved all WindowFunctions, we will use ResolveWindowFunction to extract WindowExpressions from projectList and aggregateExpressions and then create a Window operator for every distinct WindowSpecDefinition. With this choice, at the execution time, we can rely on the Exchange operator to do all of work on reorganizing the table and we do not need to worry about it in the physical Window operator. An example analyzed plan is shown as follows
```
sql("""
SELECT
year, country, product, sales,
avg(sales) over(partition by product) avg_product,
sum(sales) over(partition by country) sum_country
FROM sales
ORDER BY year, country, product
""").explain(true)
== Analyzed Logical Plan ==
Sort [year#34 ASC,country#35 ASC,product#36 ASC], true
Project [year#34,country#35,product#36,sales#37,avg_product#27,sum_country#28]
Window [year#34,country#35,product#36,sales#37,avg_product#27], [HiveWindowFunction#org.apache.hadoop.hive.ql.udf.generic.GenericUDAFSum(sales#37) WindowSpecDefinition [country#35], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING AS sum_country#28], WindowSpecDefinition [country#35], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
Window [year#34,country#35,product#36,sales#37], [HiveWindowFunction#org.apache.hadoop.hive.ql.udf.generic.GenericUDAFAverage(sales#37) WindowSpecDefinition [product#36], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING AS avg_product#27], WindowSpecDefinition [product#36], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
Project [year#34,country#35,product#36,sales#37]
MetastoreRelation default, sales, None
```
* Query planning: In the process of query planning, we simple generate the physical Window operator based on the logical Window operator. Then, to prepare the executedPlan, the EnsureRequirements rule will add Exchange and Sort operators if necessary. The EnsureRequirements rule will analyze the data properties and try to not add unnecessary shuffle and sort. The physical plan for the above example query is shown below.
```
== Physical Plan ==
Sort [year#34 ASC,country#35 ASC,product#36 ASC], true
Exchange (RangePartitioning [year#34 ASC,country#35 ASC,product#36 ASC], 200), []
Window [year#34,country#35,product#36,sales#37,avg_product#27], [HiveWindowFunction#org.apache.hadoop.hive.ql.udf.generic.GenericUDAFSum(sales#37) WindowSpecDefinition [country#35], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING AS sum_country#28], WindowSpecDefinition [country#35], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
Exchange (HashPartitioning [country#35], 200), [country#35 ASC]
Window [year#34,country#35,product#36,sales#37], [HiveWindowFunction#org.apache.hadoop.hive.ql.udf.generic.GenericUDAFAverage(sales#37) WindowSpecDefinition [product#36], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING AS avg_product#27], WindowSpecDefinition [product#36], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
Exchange (HashPartitioning [product#36], 200), [product#36 ASC]
HiveTableScan [year#34,country#35,product#36,sales#37], (MetastoreRelation default, sales, None), None
```
* Execution time: At execution time, a physical Window operator buffers all rows in a partition specified in the partition spec of a OVER clause. If necessary, it also maintains a sliding window frame. The current implementation tries to buffer the input parameters of a window function according to the window frame to avoid evaluating a row multiple times.
# Future work #
Here are three improvements that are not hard to add:
* Taking advantage of the window frame specification to reduce the number of rows buffered in the physical Window operator. For some cases, we only need to buffer the rows appearing in the sliding window. But for other cases, we will not be able to reduce the number of rows buffered (e.g. ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING).
* When aRAGEN frame is used, for <value> PRECEDING and <value> FOLLOWING, it will be great if the <value> part is an expression (we can start with Literal). So, when the data type of ORDER BY expression is a FractionalType, we can support FractionalType as the type <value> (<value> still needs to be evaluated as a positive value).
* When aRAGEN frame is used, we need to support DateType and TimestampType as the data type of the expression appearing in the order specification. Then, the <value> part of <value> PRECEDING and <value> FOLLOWING can support interval types (once we support them).
This is a joint work with guowei2 and yhuai
Thanks hbutani hvanhovell for his comments
Thanks scwf for his comments and unit tests
Author: Yin Huai <yhuai@databricks.com>
Closes#5604 from guowei2/windowImplement and squashes the following commits:
76fe1c8 [Yin Huai] Implementation.
aa2b0ae [Yin Huai] Tests.
(cherry picked from commit f2c47082c3)
Signed-off-by: Michael Armbrust <michael@databricks.com>
huangjs
Acutally spark sql will first go through analysis period, in which we do widen types and promote strings, and then optimization, where constant IN will be converted into INSET.
So it turn out that we only need to fix this for IN.
Author: Daoyuan Wang <daoyuan.wang@intel.com>
Closes#4945 from adrian-wang/inset and squashes the following commits:
71e05cc [Daoyuan Wang] minor fix
581fa1c [Daoyuan Wang] mysql way
f3f7baf [Daoyuan Wang] address comments
5eed4bc [Daoyuan Wang] promote string and do widen types for IN
(cherry picked from commit c3eb441f54)
Signed-off-by: Yin Huai <yhuai@databricks.com>
After a discussion on the user mailing list, it was decided to put all UDF's under `o.a.s.sql.functions`
cc rxin
Author: Burak Yavuz <brkyvz@gmail.com>
Closes#5923 from brkyvz/move-math-funcs and squashes the following commits:
a8dc3f7 [Burak Yavuz] address comments
cf7a7bb [Burak Yavuz] [SPARK-7358] Move DataFrame mathfunctions into functions
(cherry picked from commit ba2b56614d)
Signed-off-by: Reynold Xin <rxin@databricks.com>
See the comment in join function for more information.
Author: Reynold Xin <rxin@databricks.com>
Closes#5919 from rxin/self-join-resolve and squashes the following commits:
e2fb0da [Reynold Xin] Updated SQLConf comment.
7233a86 [Reynold Xin] Updated comment.
6be2b4d [Reynold Xin] Removed println
9f6b72f [Reynold Xin] [SPARK-6231][SQL/DF] Automatically resolve ambiguity in join condition for self-joins.
make StringComparison extends ExpectsInputTypes and added expectedChildTypes, so do not need override expectedChildTypes in each subclass
Author: wangfei <wangfei1@huawei.com>
Closes#5905 from scwf/ExpectsInputTypes and squashes the following commits:
b374ddf [wangfei] make stringcomparison extends ExpectsInputTypes
(cherry picked from commit 3059291e20)
Signed-off-by: Reynold Xin <rxin@databricks.com>
Two minor doc errors in `BytesToBytesMap` and `UnsafeRow`.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#5906 from viirya/minor_doc and squashes the following commits:
27f9089 [Liang-Chi Hsieh] Minor update for doc.
(cherry picked from commit b83091ae45)
Signed-off-by: Sean Owen <sowen@cloudera.com>
This should gives us better analysis time error messages (rather than runtime) and automatic type casting.
Author: Reynold Xin <rxin@databricks.com>
Closes#5796 from rxin/expected-input-types and squashes the following commits:
c900760 [Reynold Xin] [SPARK-7266] Add ExpectsInputTypes to expressions when possible.
(cherry picked from commit 678c4da0fa)
Signed-off-by: Reynold Xin <rxin@databricks.com>
This PR adds initial support for loading multiple versions of Hive in a single JVM and provides a common interface for extracting metadata from the `HiveMetastoreClient` for a given version. This is accomplished by creating an isolated `ClassLoader` that operates according to the following rules:
- __Shared Classes__: Java, Scala, logging, and Spark classes are delegated to `baseClassLoader`
allowing the results of calls to the `ClientInterface` to be visible externally.
- __Hive Classes__: new instances are loaded from `execJars`. These classes are not
accessible externally due to their custom loading.
- __Barrier Classes__: Classes such as `ClientWrapper` are defined in Spark but must link to a specific version of Hive. As a result, the bytecode is acquired from the Spark `ClassLoader` but a new copy is created for each instance of `IsolatedClientLoader`.
This new instance is able to see a specific version of hive without using reflection where ever hive is consistent across versions. Since
this is a unique instance, it is not visible externally other than as a generic
`ClientInterface`, unless `isolationOn` is set to `false`.
In addition to the unit tests, I have also tested this locally against mysql instances of the Hive Metastore. I've also successfully ported Spark SQL to run with this client, but due to the size of the changes, that will come in a follow-up PR.
By default, Hive jars are currently downloaded from Maven automatically for a given version to ease packaging and testing. However, there is also support for specifying their location manually for deployments without internet.
Author: Michael Armbrust <michael@databricks.com>
Closes#5851 from marmbrus/isolatedClient and squashes the following commits:
c72f6ac [Michael Armbrust] rxins comments
1e271fa [Michael Armbrust] [SPARK-6907][SQL] Isolated client for HiveMetastore
based on #4015, we should not delete `sqlParser` from sqlcontext, that leads to mima failed. Users implement dialect to give a fallback for `sqlParser` and we should construct `sqlParser` in sqlcontext according to the dialect
`protected[sql] val sqlParser = new SparkSQLParser(getSQLDialect().parse(_))`
Author: Cheng Hao <hao.cheng@intel.com>
Author: scwf <wangfei1@huawei.com>
Closes#5827 from scwf/sqlparser1 and squashes the following commits:
81b9737 [scwf] comment fix
0878bd1 [scwf] remove comments
c19780b [scwf] fix mima tests
c2895cf [scwf] Merge branch 'master' of https://github.com/apache/spark into sqlparser1
493775c [Cheng Hao] update the code as feedback
81a731f [Cheng Hao] remove the unecessary comment
aab0b0b [Cheng Hao] polish the code a little bit
49b9d81 [Cheng Hao] shrink the comment for rebasing
Adds the functions `rand` (Uniform Dist) and `randn` (Normal Dist.) as expressions to DataFrames.
cc mengxr rxin
Author: Burak Yavuz <brkyvz@gmail.com>
Closes#5819 from brkyvz/df-rng and squashes the following commits:
50d69d4 [Burak Yavuz] add seed for test that failed
4234c3a [Burak Yavuz] fix Rand expression
13cad5c [Burak Yavuz] couple fixes
7d53953 [Burak Yavuz] waiting for hive tests
b453716 [Burak Yavuz] move radn with seed down
03637f0 [Burak Yavuz] fix broken hive func
c5909eb [Burak Yavuz] deleted old implementation of Rand
6d43895 [Burak Yavuz] implemented random generators
Run following sql get error
`SELECT r.*
FROM testData l join testData2 r on (l.key = r.a)`
Author: scwf <wangfei1@huawei.com>
Closes#5690 from scwf/tablestar and squashes the following commits:
3b2e2b6 [scwf] support table.star
This PR aims to make the SQL Parser Pluggable, and user can register it's own parser via Spark SQL CLI.
```
# add the jar into the classpath
$hchengmydesktop:spark>bin/spark-sql --jars sql99.jar
-- switch to "hiveql" dialect
spark-sql>SET spark.sql.dialect=hiveql;
spark-sql>SELECT * FROM src LIMIT 1;
-- switch to "sql" dialect
spark-sql>SET spark.sql.dialect=sql;
spark-sql>SELECT * FROM src LIMIT 1;
-- switch to a custom dialect
spark-sql>SET spark.sql.dialect=com.xxx.xxx.SQL99Dialect;
spark-sql>SELECT * FROM src LIMIT 1;
-- register the non-exist SQL dialect
spark-sql> SET spark.sql.dialect=NotExistedClass;
spark-sql> SELECT * FROM src LIMIT 1;
-- Exception will be thrown and switch to default sql dialect ("sql" for SQLContext and "hiveql" for HiveContext)
```
Author: Cheng Hao <hao.cheng@intel.com>
Closes#4015 from chenghao-intel/sqlparser and squashes the following commits:
493775c [Cheng Hao] update the code as feedback
81a731f [Cheng Hao] remove the unecessary comment
aab0b0b [Cheng Hao] polish the code a little bit
49b9d81 [Cheng Hao] shrink the comment for rebasing
Now in spark sql optimizer we only push down right side filter for left semi join, actually we can push down left side filter because left semi join is doing filter on left table essentially.
Author: wangfei <wangfei1@huawei.com>
Author: scwf <wangfei1@huawei.com>
Closes#5677 from scwf/leftsemi and squashes the following commits:
483d205 [wangfei] update with master to fix compile issue
82df0e1 [wangfei] Merge branch 'master' of https://github.com/apache/spark into leftsemi
d68a053 [wangfei] added apply
8f48a3d [scwf] added test
ebadaa9 [wangfei] left filter push down for left semi join
Author: 云峤 <chensong.cs@alibaba-inc.com>
Closes#5778 from kaka1992/fix_codegenon_datetype_mismatch and squashes the following commits:
1ad4cff [云峤] SPARK-7234 fix dateType mismatch
Author: Cheng Hao <hao.cheng@intel.com>
Closes#5772 from chenghao-intel/specific_row and squashes the following commits:
2cd064d [Cheng Hao] scala style issue
60347a2 [Cheng Hao] SpecificMutableRow should take integer type as internal representation for DateType
This is built on top of kaka1992 's PR #5711 using Logical plans.
Author: Burak Yavuz <brkyvz@gmail.com>
Closes#5761 from brkyvz/random-sample and squashes the following commits:
a1fb0aa [Burak Yavuz] remove unrelated file
69669c3 [Burak Yavuz] fix broken test
1ddb3da [Burak Yavuz] copy base
6000328 [Burak Yavuz] added python api and fixed test
3c11d1b [Burak Yavuz] fixed broken test
f400ade [Burak Yavuz] fix build errors
2384266 [Burak Yavuz] addressed comments v0.1
e98ebac [Burak Yavuz] [SPARK-7156][SQL] support RandomSplit in DataFrames
This patch adds managed-memory-based aggregation to Spark SQL / DataFrames. Instead of working with Java objects, this new aggregation path uses `sun.misc.Unsafe` to manipulate raw memory. This reduces the memory footprint for aggregations, resulting in fewer spills, OutOfMemoryErrors, and garbage collection pauses. As a result, this allows for higher memory utilization. It can also result in better cache locality since objects will be stored closer together in memory.
This feature can be eanbled by setting `spark.sql.unsafe.enabled=true`. For now, this feature is only supported when codegen is enabled and only supports aggregations for which the grouping columns are primitive numeric types or strings and aggregated values are numeric.
### Managing memory with sun.misc.Unsafe
This patch supports both on- and off-heap managed memory.
- In on-heap mode, memory addresses are identified by the combination of a base Object and an offset within that object.
- In off-heap mode, memory is addressed directly with 64-bit long addresses.
To support both modes, functions that manipulate memory accept both `baseObject` and `baseOffset` fields. In off-heap mode, we simply pass `null` as `baseObject`.
We allocate memory in large chunks, so memory fragmentation and allocation speed are not significant bottlenecks.
By default, we use on-heap mode. To enable off-heap mode, set `spark.unsafe.offHeap=true`.
To track allocated memory, this patch extends `SparkEnv` with an `ExecutorMemoryManager` and supplies each `TaskContext` with a `TaskMemoryManager`. These classes work together to track allocations and detect memory leaks.
### Compact tuple format
This patch introduces `UnsafeRow`, a compact row layout. In this format, each tuple has three parts: a null bit set, fixed length values, and variable-length values:
- Rows are always 8-byte word aligned (so their sizes will always be a multiple of 8 bytes)
- The bit set is used for null tracking:
- Position _i_ is set if and only if field _i_ is null
- The bit set is aligned to an 8-byte word boundary.
- Every field appears as an 8-byte word in the fixed-length values part:
- If a field is null, we zero out the values.
- If a field is variable-length, the word stores a relative offset (w.r.t. the base of the tuple) that points to the beginning of the field's data in the variable-length part.
- Each variable-length data type can have its own encoding:
- For strings, the first word stores the length of the string and is followed by UTF-8 encoded bytes. If necessary, the end of the string is padded with empty bytes in order to ensure word-alignment.
For example, a tuple that consists 3 fields of type (int, string, string), with value (null, “data”, “bricks”) would look like this:
This format allows us to compare tuples for equality by directly comparing their raw bytes. This also enables fast hashing of tuples.
### Hash map for performing aggregations
This patch introduces `UnsafeFixedWidthAggregationMap`, a hash map for performing aggregations where the aggregation result columns are fixed-with. This map's keys and values are `Row` objects. `UnsafeFixedWidthAggregationMap` is implemented on top of `BytesToBytesMap`, an append-only map which supports byte-array keys and values.
`BytesToBytesMap` stores pointers to key and value tuples. For each record with a new key, we copy the key and create the aggregation value buffer for that key and put them in a buffer. The hash table then simply stores pointers to the key and value. For each record with an existing key, we simply run the aggregation function to update the values in place.
This map is implemented using open hashing with triangular sequence probing. Each entry stores two words in a long array: the first word stores the address of the key and the second word stores the relative offset from the key tuple to the value tuple, as well as the key's 32-bit hashcode. By storing the full hashcode, we reduce the number of equality checks that need to be performed to handle position collisions ()since the chance of hashcode collision is much lower than position collision).
`UnsafeFixedWidthAggregationMap` allows regular Spark SQL `Row` objects to be used when probing the map. Internally, it encodes these rows into `UnsafeRow` format using `UnsafeRowConverter`. This conversion has a small overhead that can be eliminated in the future once we use UnsafeRows in other operators.
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/apache/spark/5725)
<!-- Reviewable:end -->
Author: Josh Rosen <joshrosen@databricks.com>
Closes#5725 from JoshRosen/unsafe and squashes the following commits:
eeee512 [Josh Rosen] Add converters for Null, Boolean, Byte, and Short columns.
81f34f8 [Josh Rosen] Follow 'place children last' convention for GeneratedAggregate
1bc36cc [Josh Rosen] Refactor UnsafeRowConverter to avoid unnecessary boxing.
017b2dc [Josh Rosen] Remove BytesToBytesMap.finalize()
50e9671 [Josh Rosen] Throw memory leak warning even in case of error; add warning about code duplication
70a39e4 [Josh Rosen] Split MemoryManager into ExecutorMemoryManager and TaskMemoryManager:
6e4b192 [Josh Rosen] Remove an unused method from ByteArrayMethods.
de5e001 [Josh Rosen] Fix debug vs. trace in logging message.
a19e066 [Josh Rosen] Rename unsafe Java test suites to match Scala test naming convention.
78a5b84 [Josh Rosen] Add logging to MemoryManager
ce3c565 [Josh Rosen] More comments, formatting, and code cleanup.
529e571 [Josh Rosen] Measure timeSpentResizing in nanoseconds instead of milliseconds.
3ca84b2 [Josh Rosen] Only zero the used portion of groupingKeyConversionScratchSpace
162caf7 [Josh Rosen] Fix test compilation
b45f070 [Josh Rosen] Don't redundantly store the offset from key to value, since we can compute this from the key size.
a8e4a3f [Josh Rosen] Introduce MemoryManager interface; add to SparkEnv.
0925847 [Josh Rosen] Disable MiMa checks for new unsafe module
cde4132 [Josh Rosen] Add missing pom.xml
9c19fc0 [Josh Rosen] Add configuration options for heap vs. offheap
6ffdaa1 [Josh Rosen] Null handling improvements in UnsafeRow.
31eaabc [Josh Rosen] Lots of TODO and doc cleanup.
a95291e [Josh Rosen] Cleanups to string handling code
afe8dca [Josh Rosen] Some Javadoc cleanup
f3dcbfe [Josh Rosen] More mod replacement
854201a [Josh Rosen] Import and comment cleanup
06e929d [Josh Rosen] More warning cleanup
ef6b3d3 [Josh Rosen] Fix a bunch of FindBugs and IntelliJ inspections
29a7575 [Josh Rosen] Remove debug logging
49aed30 [Josh Rosen] More long -> int conversion.
b26f1d3 [Josh Rosen] Fix bug in murmur hash implementation.
765243d [Josh Rosen] Enable optional performance metrics for hash map.
23a440a [Josh Rosen] Bump up default hash map size
628f936 [Josh Rosen] Use ints intead of longs for indexing.
92d5a06 [Josh Rosen] Address a number of minor code review comments.
1f4b716 [Josh Rosen] Merge Unsafe code into the regular GeneratedAggregate, guarded by a configuration flag; integrate planner support and re-enable all tests.
d85eeff [Josh Rosen] Add basic sanity test for UnsafeFixedWidthAggregationMap
bade966 [Josh Rosen] Comment update (bumping to refresh GitHub cache...)
b3eaccd [Josh Rosen] Extract aggregation map into its own class.
d2bb986 [Josh Rosen] Update to implement new Row methods added upstream
58ac393 [Josh Rosen] Use UNSAFE allocator in GeneratedAggregate (TODO: make this configurable)
7df6008 [Josh Rosen] Optimizations related to zeroing out memory:
c1b3813 [Josh Rosen] Fix bug in UnsafeMemoryAllocator.free():
738fa33 [Josh Rosen] Add feature flag to guard UnsafeGeneratedAggregate
c55bf66 [Josh Rosen] Free buffer once iterator has been fully consumed.
62ab054 [Josh Rosen] Optimize for fact that get() is only called on String columns.
c7f0b56 [Josh Rosen] Reuse UnsafeRow pointer in UnsafeRowConverter
ae39694 [Josh Rosen] Add finalizer as "cleanup method of last resort"
c754ae1 [Josh Rosen] Now that the store*() contract has been stregthened, we can remove an extra lookup
f764d13 [Josh Rosen] Simplify address + length calculation in Location.
079f1bf [Josh Rosen] Some clarification of the BytesToBytesMap.lookup() / set() contract.
1a483c5 [Josh Rosen] First version that passes some aggregation tests:
fc4c3a8 [Josh Rosen] Sketch how the converters will be used in UnsafeGeneratedAggregate
53ba9b7 [Josh Rosen] Start prototyping Java Row -> UnsafeRow converters
1ff814d [Josh Rosen] Add reminder to free memory on iterator completion
8a8f9df [Josh Rosen] Add skeleton for GeneratedAggregate integration.
5d55cef [Josh Rosen] Add skeleton for Row implementation.
f03e9c1 [Josh Rosen] Play around with Unsafe implementations of more string methods.
ab68e08 [Josh Rosen] Begin merging the UTF8String implementations.
480a74a [Josh Rosen] Initial import of code from Databricks unsafe utils repo.
Adds support for the math functions for DataFrames in PySpark.
rxin I love Davies.
Author: Burak Yavuz <brkyvz@gmail.com>
Closes#5750 from brkyvz/python-math-udfs and squashes the following commits:
7c4f563 [Burak Yavuz] removed is_math
3c4adde [Burak Yavuz] cleanup imports
d5dca3f [Burak Yavuz] moved math functions to mathfunctions
25e6534 [Burak Yavuz] addressed comments v2.0
d3f7e0f [Burak Yavuz] addressed comments and added tests
7b7d7c4 [Burak Yavuz] remove tests for removed methods
33c2c15 [Burak Yavuz] fixed python style
3ee0c05 [Burak Yavuz] added python functions
Coalesce and repartition now show up as part of the query plan, rather than resulting in a new `DataFrame`.
cc rxin
Author: Burak Yavuz <brkyvz@gmail.com>
Closes#5762 from brkyvz/df-repartition and squashes the following commits:
b1e76dd [Burak Yavuz] added documentation on repartitions
5807e35 [Burak Yavuz] renamed coalescepartitions
fa4509f [Burak Yavuz] rename coalesce
2c349b5 [Burak Yavuz] address comments
f2e6af1 [Burak Yavuz] add ticks
686c90b [Burak Yavuz] made coalesce and repartition a part of the query plan
Implemented almost all math functions found in scala.math (max, min and abs were already present).
cc mengxr marmbrus
Author: Burak Yavuz <brkyvz@gmail.com>
Closes#5616 from brkyvz/math-udfs and squashes the following commits:
fb27153 [Burak Yavuz] reverted exception message
836a098 [Burak Yavuz] fixed test and addressed small comment
e5f0d13 [Burak Yavuz] addressed code review v2.2
b26c5fb [Burak Yavuz] addressed review v2.1
2761f08 [Burak Yavuz] addressed review v2
6588a5b [Burak Yavuz] fixed merge conflicts
b084e10 [Burak Yavuz] Addressed code review
029e739 [Burak Yavuz] fixed atan2 test
534cc11 [Burak Yavuz] added more tests, addressed comments
fa68dbe [Burak Yavuz] added double specific test data
937d5a5 [Burak Yavuz] use doubles instead of ints
8e28fff [Burak Yavuz] Added apache header
7ec8f7f [Burak Yavuz] Added math functions for DataFrames
rename DataTypeParser.apply to DataTypeParser.parse to make it more clear and readable.
/cc rxin
Author: wangfei <wangfei1@huawei.com>
Closes#5710 from scwf/apply and squashes the following commits:
c319977 [wangfei] rename apply to parse
