### What changes were proposed in this pull request?
Deprecate Python 3.6 in Spark documentation
### Why are the changes needed?
According to https://endoflife.date/python, Python 3.6 will be EOL on 23 Dec, 2021.
We should prepare for the deprecation of Python 3.6 support in Spark in advance.
### Does this PR introduce _any_ user-facing change?
N/A.
### How was this patch tested?
Manual tests.
Closes#33141 from xinrong-databricks/deprecate3.6_doc.
Authored-by: Xinrong Meng <xinrong.meng@databricks.com>
Signed-off-by: Hyukjin Kwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
There are two new Avro options `datetimeRebaseMode` and `positionalFieldMatching` after Spark 3.2.
We should document the since version so that users can know whether the option works in their Spark version.
### Why are the changes needed?
Better documentation.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Manual preview on local setup.
<img width="828" alt="Screen Shot 2021-06-30 at 5 05 54 PM" src="https://user-images.githubusercontent.com/1097932/123934000-ba833b00-d947-11eb-9ca5-ce8ff8add74b.png">
<img width="711" alt="Screen Shot 2021-06-30 at 5 06 34 PM" src="https://user-images.githubusercontent.com/1097932/123934126-d4bd1900-d947-11eb-8d80-69df8f3d9900.png">
Closes#33153 from gengliangwang/version.
Authored-by: Gengliang Wang <gengliang@apache.org>
Signed-off-by: Gengliang Wang <gengliang@apache.org>
### What changes were proposed in this pull request?
Provide the (configurable) ability to perform Avro-to-Catalyst schema field matching using the position of the fields instead of their names. A new `option` is added for the Avro datasource, `positionalFieldMatching`, which instructs `AvroSerializer`/`AvroDeserializer` to perform positional field matching instead of matching by name.
### Why are the changes needed?
This by-name matching is somewhat recent; prior to PR #24635, at least on the write path, schemas were matched by positionally ("structural" comparison). While by-name is better behavior as a default, it will be better to make this configurable by a user. Even at the time that PR #24635 was handled, there was [interest in making this behavior configurable](https://github.com/apache/spark/pull/24635#issuecomment-494205251), but it appears it went unaddressed.
There is precedence for configurability of this behavior as seen in PR #29737, which added this support for ORC. Besides this precedence, the behavior of Hive is to perform matching positionally ([ref](https://cwiki.apache.org/confluence/display/Hive/AvroSerDe#AvroSerDe-WritingtablestoAvrofiles)), so this is behavior that Hadoop/Hive ecosystem users are familiar with.
### Does this PR introduce _any_ user-facing change?
Yes, a new option is provided for the Avro datasource, `positionalFieldMatching`, which provides compatibility with Hive and pre-3.0.0 Spark behavior.
### How was this patch tested?
New unit tests are added within `AvroSuite`, `AvroSchemaHelperSuite`, and `AvroSerdeSuite`; and most of the existing tests within `AvroSerdeSuite` are adapted to perform the same test using by-name and positional matching to ensure feature parity.
Closes#31490 from xkrogen/xkrogen-SPARK-34365-avro-positional-field-matching.
Authored-by: Erik Krogen <xkrogen@apache.org>
Signed-off-by: Gengliang Wang <gengliang@apache.org>
### What changes were proposed in this pull request?
This adds two new additional metrics to `ExternalBlockHandler`:
- `blockTransferRate` -- for indicating the rate of transferring blocks, vs. the data within them
- `blockTransferAvgSize_1min` -- a 1-minute trailing average of block sizes transferred by the ESS
Additionally, this enhances `YarnShuffleServiceMetrics` to expose the histogram/`Snapshot` information from `Timer` metrics within `ExternalBlockHandler`.
### Why are the changes needed?
Currently `ExternalBlockHandler` exposes some useful metrics, but is lacking around metrics for the rate of block transfers. We have `blockTransferRateBytes` to tell us the rate of _bytes_, but no metric to tell us the rate of _blocks_, which is especially relevant when running the ESS on HDDs that are sensitive to random reads. Many small block transfers can have a negative impact on performance, but won't show up as a spike in `blockTransferRateBytes` since the sizes are small. Thus the new metrics to show information around average block size and block transfer rate are very useful to monitor the health/performance of the ESS, especially when running on HDDs.
For the `YarnShuffleServiceMetrics`, currently the three `Timer` metrics exposed by `ExternalBlockHandler` are being underutilized in a YARN-based environment -- they are basically treated as a `Meter`, only exposing rate-based information, when the metrics themselves are collected detailed histograms of timing information. We should expose this information for better observability.
### Does this PR introduce _any_ user-facing change?
Yes, there are two entirely new metrics for the ESS, as documented in `monitoring.md`. Additionally in a YARN environment, `Timer` metrics exposed by the ESS will include more rich timing information.
### How was this patch tested?
New unit tests are added to verify that new metrics are showing up as expected.
We have been running this patch internally for approx. 1 year and have found it to be useful for monitoring the health of ESS and diagnosing performance issues.
Closes#32388 from xkrogen/xkrogen-SPARK-35258-ess-new-metrics.
Authored-by: Erik Krogen <xkrogen@apache.org>
Signed-off-by: Mridul Muralidharan <mridul<at>gmail.com>
### What changes were proposed in this pull request?
The hyperlinks in Python code blocks in [Spark SQL Guide - Getting Started](https://spark.apache.org/docs/latest/sql-getting-started.html) currently point to invalid addresses and return 404. This pull request fixes that issue by pointing them to correct links in Python API docs.
### Why are the changes needed?
Error in documentation classifies as a bug and hence needs to be fixed.
### Does this PR introduce _any_ user-facing change?
Yes. This PR fixes documentation error in https://spark.apache.org/docs/latest/sql-getting-started.html
### How was this patch tested?
This patch was locally built after cloning the repo from scratch and then doing a clean build after fixing the required problems.
Closes#33107 from dhruvildave/sql-doc.
Authored-by: Dhruvil Dave <dhruvil.dave@outlook.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
### What changes were proposed in this pull request?
Update "Caching Data in Memory" section, add suggestion to call DataFrame `unpersist` method to make it consistent with previous suggestion of using `persist` method.
### Why are the changes needed?
Keep documentation consistent.
### Does this PR introduce _any_ user-facing change?
Yes, fixes the user-facing docs.
### How was this patch tested?
Manually.
Closes#33069 from Silverlight42/caching-data-doc.
Authored-by: Carlos Peña <Cdpm42@gmail.com>
Signed-off-by: Hyukjin Kwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR changes the unionByName with null filling logic to append new nested struct fields from the right side of the union to the schema versus sorting fields alphabetically. It removes the need to use UpdateField expressions, and just directly projects new nested structs from each side of the union with the correct schema. This changes the union'd schema from being alphabetically sorted previously to now "left dominant", where the fields from the left side of the union are included and then the missing ones from the right are added in the same order found originally.
### Why are the changes needed?
Certain nested structs would cause unionByName with null filling to error out due to part of the logic for rewriting the expression tree to sort the structs.
### Does this PR introduce _any_ user-facing change?
Yes, nested struct fields will be in a different order after unionByName with null filling than before, though shouldn't cause much effective difference.
### How was this patch tested?
Updated existing tests based on the new StructField ordering and added a new test for the case that was broken originally.
Closes#33040 from Kimahriman/union-by-name-struct-order.
Authored-by: Adam Binford <adamq43@gmail.com>
Signed-off-by: Liang-Chi Hsieh <viirya@gmail.com>
### What changes were proposed in this pull request?
* Add a new repartition operator `RebalanceRepartition`.
* Support a new hint `REBALANCE`
After this patch, user can run this query:
```sql
SELECT /*+ REBALANCE(c) */ * FROM t
```
### Why are the changes needed?
Add a new hint to distingush if we can optimize it safely.
This new hint can let AQE optimize with `CustomShuffleReaderExec` safely. Currently, AQE can only coalesce shuffle partitions but can not expand shuffle partitions due to the semantics of output partitioning.
Let's say we have a query:
```sql
SELECT /*+ REPARTITION(col) */ * FROM t
```
AQE can not expand the shuffle partitions even if `col` is skewed because expanding shuffle partitions will break the hashed output paritioning of `RepartitionByExpression`. But if the query is use`REPARTITION_BY_AQE`, AQE can optimize it without considering the semantics of output partitioning.
### Does this PR introduce _any_ user-facing change?
Yes, a new hint.
### How was this patch tested?
Add test.
Closes#32932 from ulysses-you/SPARK-35786.
Authored-by: ulysses-you <ulyssesyou18@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This's a follow-up of https://github.com/apache/spark/pull/30710.
Rename the conf from `spark.speculation.min.threshold` to `spark.speculation.minTaskRuntime`.
### Why are the changes needed?
To follow the [config naming policy](https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/internal/config/ConfigEntry.scala#L21).
### Does this PR introduce _any_ user-facing change?
No (since Spark 3.2 hasn't been released).
### How was this patch tested?
Pass existing tests.
Closes#33037 from Ngone51/spark-33741-followup.
Authored-by: yi.wu <yi.wu@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This PR proposes to introduce the strategy on mismatched offset for start offset timestamp on Kafka data source.
Please read the section `Why are the changes needed?` to understand the rationalization of the functionality.
This would be pretty much helpful for the case where there's a skew between partitions and some partitions have older records.
* AS-IS: Spark simply fails the query and end users have to deal with workarounds requiring manual steps.
* TO-BE: Spark will assign the latest offset for these partitions, so that Spark can read newer records from these partitions in further micro-batches.
To retain the existing behavior and also give some help for the proposed "TO-BE" behavior, we'd like to introduce the strategy on mismatched offset for start offset timestamp to let end users choose from them.
The strategy will be added as source option, to ensure end users set the behavior explicitly (otherwise simply "known" default value).
* New source option to be added: startingOffsetsByTimestampStrategy
* Available values: `error` (fail the query as referred as AS-IS), `latest` (set the offset to the latest as referred as TO-BE)
Doc changes are following:
### Why are the changes needed?
We encountered a real-world case Spark fails the query if some of the partitions don't have matching offset by timestamp.
This is intended behavior to avoid bring unintended output for some cases like:
* timestamp 2 is presented as timestamp-offset, but the some of partitions don't have the record yet
* record with timestamp 1 comes "later" in the following micro-batch
which is possible since Kafka allows to specify the timestamp in record.
Here the unintended output we talked about was the risk of reading record with timestamp 1 in the next micro-batch despite the option specifying timestamp 2.
But for many cases end users just suppose timestamp is increasing monotonically with wall clocks are all in sync, and current behavior blocks these cases to make progress.
### Does this PR introduce _any_ user-facing change?
Yes, but not a breaking change. It's up to end users to choose the behavior which the default value is "error" (current behavior). And it's a source option (not config) so they need to explicitly set the behavior to let the functionality takes effect.
### How was this patch tested?
New UTs.
Closes#32747 from HeartSaVioR/SPARK-35611.
Authored-by: Jungtaek Lim <kabhwan.opensource@gmail.com>
Signed-off-by: Liang-Chi Hsieh <viirya@gmail.com>
### What changes were proposed in this pull request?
PySpark added pinned thread mode at https://github.com/apache/spark/pull/24898 to sync Python thread to JVM thread. Previously, one JVM thread could be reused which ends up with messed inheritance hierarchy such as thread local especially when multiple jobs run in parallel. To completely fix this, we should enable this mode by default.
### Why are the changes needed?
To correctly support parallel job submission and management.
### Does this PR introduce _any_ user-facing change?
Yes, now Python thread is mapped to JVM thread one to one.
### How was this patch tested?
Existing tests should cover it.
Closes#32429 from HyukjinKwon/SPARK-35303.
Authored-by: HyukjinKwon <gurwls223@apache.org>
Signed-off-by: Hyukjin Kwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This is a followup of https://github.com/apache/spark/pull/32513
It's hard to keep the command execution name for `DataFrameWriter`, as the command logical plan is a bit messy (DS v1, file source and hive and different command logical plans) and sometimes it's hard to distinguish "insert" and "save".
However, `DataFrameWriterV2` only produce v2 commands which are pretty clean. It's easy to keep the command execution name for them.
### Why are the changes needed?
less breaking changes.
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
N/A
Closes#32919 from cloud-fan/follow.
Authored-by: Wenchen Fan <wenchen@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This PR updates the document about building Spark with Hadoop for Hadoop 3.x and Hadoop 3.2.
### Why are the changes needed?
The document says about how to build like as follows:
```
./build/mvn -Pyarn -Dhadoop.version=2.8.5 -DskipTests clean package
```
But this command fails because the default build settings are for Hadoop 3.x.
So, we need to modify the command example.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
I confirmed both of these commands successfully finished.
```
./build/mvn -Pyarn -Dhadoop.version=3.3.0 -DskipTests package
./build/mvn -Phadoop-2.7 -Pyarn -Dhadoop.version=2.8.5 -DskipTests package
```
I also built the document and confirmed the result.
This is before:
And this is after:
Closes#32917 from sarutak/fix-build-doc-with-hadoop.
Authored-by: Kousuke Saruta <sarutak@oss.nttdata.com>
Signed-off-by: Hyukjin Kwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR adda a feature which allow the parser parse any day-time interval types in SQL.
### Why are the changes needed?
To comply with ANSI standard, we additionally need to support the following types.
* INTERVAL DAY
* INTERVAL DAY TO HOUR
* INTERVAL DAY TO MINUTE
* INTERVAL HOUR
* INTERVAL HOUR TO MINUTE
* INTERVAL HOUR TO SECOND
* INTERVAL MINUTE
* INTERVAL MINUTE TO SECOND
* INTERVAL SECOND
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
New tests.
Closes#32893 from sarutak/parse-any-day-time.
Authored-by: Kousuke Saruta <sarutak@oss.nttdata.com>
Signed-off-by: Max Gekk <max.gekk@gmail.com>
### What changes were proposed in this pull request?
The STRUCT type syntax is defined like this:
STRUCT(fieldNmae: fileType [NOT NULL][COMMENT stringLiteral][,.....])
So the field list is nearly the same as a column list
if we could make ':' optional it would be so much cleaner an less proprietary
### Why are the changes needed?
ease of use
### Does this PR introduce _any_ user-facing change?
Yes, you can use Struct type list is nearly the same as a column list
### How was this patch tested?
unit tests
Closes#32858 from jerqi/master.
Authored-by: RoryQi <1242949407@qq.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This pr upgrades built-in Hive to 2.3.9. Hive 2.3.9 changes:
- [HIVE-17155] - findConfFile() in HiveConf.java has some issues with the conf path
- [HIVE-24797] - Disable validate default values when parsing Avro schemas
- [HIVE-24608] - Switch back to get_table in HMS client for Hive 2.3.x
- [HIVE-21200] - Vectorization: date column throwing java.lang.UnsupportedOperationException for parquet
- [HIVE-21563] - Improve Table#getEmptyTable performance by disabling registerAllFunctionsOnce
- [HIVE-19228] - Remove commons-httpclient 3.x usage
### Why are the changes needed?
Fix regression caused by AVRO-2035.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Unit test.
Closes#32750 from wangyum/SPARK-34512.
Authored-by: Yuming Wang <yumwang@ebay.com>
Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
### What changes were proposed in this pull request?
This PR updates documentation by removing reference to [hashicorp/nomad-spark](https://github.com/hashicorp/nomad-spark) which has been deprecated in April 2020, and will not be developed any longer.
### Why are the changes needed?
To keep the documentation updated and remove confusion for potential users being interested in running with Nomad.
### Does this PR introduce _any_ user-facing change?
Yes. A change to the documentation.
### How was this patch tested?
Generated to documentation, and checked everything is alright in the output.
Closes#32860 from pptaszynski/doc/remove-spark-nomad-project-reference.
Authored-by: Pawel Ptaszynski <pawel.ptaszynski@bolt.eu>
Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
### What changes were proposed in this pull request?
This PR adds support for lateral subqueries. A lateral subquery is a subquery preceded by the `LATERAL` keyword in the FROM clause of a query that can reference columns in the preceding FROM items. For example:
```sql
SELECT * FROM t1, LATERAL (SELECT * FROM t2 WHERE t1.a = t2.c)
```
A new subquery expression`LateralSubquery` is used to represent a lateral subquery. It is similar to `ScalarSubquery` but can return multiple rows and columns. A new logical unary node `LateralJoin` is used to represent a lateral join.
Here is the analyzed plan for the above query:
```scala
Project [a, b, c, d]
+- LateralJoin lateral-subquery [a], Inner
: +- Project [c, d]
: +- Filter (outer(a) = c)
: +- Relation [c, d]
+- Relation [a, b]
```
Similar to a correlated subquery, a lateral subquery can be viewed as a dependent (nested loop) join where the evaluation of the right subtree depends on the current value of the left subtree. The same technique to decorrelate a subquery is used to decorrelate a lateral join:
```scala
Project [a, b, c, d]
+- LateralJoin lateral-subquery [a && a = c], Inner // pull up correlated predicates as join conditions
: +- Project [c, d]
: +- Relation [c, d]
+- Relation [a, b]
```
Then the lateral join can be rewritten into a normal join:
```scala
Join Inner (a = c)
:- Relation [a, b]
+- Relation [c, d]
```
#### Follow-ups:
1. Similar to rewriting correlated scalar subqueries, rewriting lateral joins is also subject to the COUNT bug (See SPARK-15370 for more details). This is **not** handled in the current PR as it requires a sizeable amount of refactoring. It will be addressed in a subsequent PR (SPARK-35551).
2. Currently Spark does use outer query references to resolve star expressions in subqueries. This is not lateral subquery specific and can be handled in a separate PR (SPARK-35618)
### Why are the changes needed?
To support an ANSI SQL feature.
### Does this PR introduce _any_ user-facing change?
Yes. It allows users to use lateral subqueries in the FROM clause of a query.
### How was this patch tested?
- Parser test: `PlanParserSuite.scala`
- Analyzer test: `ResolveSubquerySuite.scala`
- Optimizer test: `PullupCorrelatedPredicatesSuite.scala`
- SQL test: `join-lateral.sql`, `postgreSQL/join.sql`
Closes#32303 from allisonwang-db/spark-34382-lateral.
Lead-authored-by: allisonwang-db <66282705+allisonwang-db@users.noreply.github.com>
Co-authored-by: Wenchen Fan <cloud0fan@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This PR enhances `RepartitionByExpression` to make it coalesce partitions efficiently by AQE. Usually used to merge small files.
The basic logic is: Spark first tries to coalesce partitions, if it cannot be coalesced, then use the local shuffle reader to read data to avoid exchange the data over the network.
Usage:
```sql
SELECT /*+ REPARTITION */ * FROM t
```
```scala
df.repartition()
```
For example:
coalesce small output files | local shuffle reader
--- | ---
| 
### Why are the changes needed?
Coalesce partitions efficiently.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Unit test.
Closes#32781 from wangyum/SPARK-35650.
Authored-by: Yuming Wang <yumwang@ebay.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
Currently, Spark eagerly executes commands on the caller side of `QueryExecution`, which is a bit hacky as `QueryExecution` is not aware of it and leads to confusion.
For example, if you run `sql("show tables").collect()`, you will see two queries with identical query plans in the web UI.
The first query is triggered at `Dataset.logicalPlan`, which eagerly executes the command.
The second query is triggered at `Dataset.collect`, which is the normal query execution.
From the web UI, it's hard to tell that these two queries are caused by eager command execution.
This PR proposes to move the eager command execution to `QueryExecution`, and turn the command plan to `CommandResult` to indicate that command has been executed already. Now `sql("show tables").collect()` still triggers two queries, but the quey plans are not identical. The second query becomes:
In addition to the UI improvements, this PR also has other benefits:
1. Simplifies code as caller side no need to worry about eager command execution. `QueryExecution` takes care of it.
2. It helps https://github.com/apache/spark/pull/32442 , where there can be more plan nodes above commands, and we need to replace commands with something like local relation that produces unsafe rows.
### Why are the changes needed?
Explained above.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
existing tests
Closes#32513 from beliefer/SPARK-35378.
Lead-authored-by: gengjiaan <gengjiaan@360.cn>
Co-authored-by: beliefer <beliefer@163.com>
Co-authored-by: Jiaan Geng <beliefer@163.com>
Co-authored-by: Wenchen Fan <cloud0fan@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This patch introduces a new option to specify the minimum number of offsets to read per trigger i.e. minOffsetsPerTrigger and maxTriggerDelay to avoid the infinite wait for the trigger.
This new option will allow skipping trigger/batch when the number of records available in Kafka is low. This is a very useful feature in cases where we have a sudden burst of data at certain intervals in a day and data volume is low for the rest of the day.
'maxTriggerDelay' option will help to avoid cases of infinite delay in scheduling trigger and the trigger will happen irrespective of records available if the maxTriggerDelay time exceeds the last trigger. It would be an optional parameter with a default value of 15 mins. This option will be only applicable if minOffsetsPerTrigger is set.
minOffsetsPerTrigger option would be optional of course, but once specified it would take precedence over maxOffestsPerTrigger which will be honored only after minOffsetsPerTrigger is satisfied.
### Why are the changes needed?
There are many scenarios where there is a sudden burst of data at certain intervals in a day and data volume is low for the rest of the day. Tunning such jobs is difficult as decreasing trigger processing time increasing the number of batches and hence cluster resource usage and adds to small file issues. Increasing trigger processing time adds consumer lag. This patch tries to address this issue.
### How was this patch tested?
This patch was tested by adding test cases as well as manually on a cluster where the job was running for a full one day with a data burst happening once a day.
Here is the picture of databurst and hence consumer lag:
<img width="1198" alt="Screenshot 2021-04-29 at 11 39 35 PM" src="https://user-images.githubusercontent.com/1044003/116997587-9b2ab180-acfa-11eb-91fd-524802ce3316.png">
This is how the job behaved at burst time running every 4.5 mins (which is the specified trigger time):
<img width="1154" alt="Burst Time" src="https://user-images.githubusercontent.com/1044003/116997919-12f8dc00-acfb-11eb-9b0a-98387fc67560.png">
This is job behavior during the non-burst time where it is skipping 2 to 3 triggers and running once every 9 to 13.5 mins
<img width="1154" alt="Non Burst Time" src="https://user-images.githubusercontent.com/1044003/116998244-8b5f9d00-acfb-11eb-8340-33d47149ef81.png">
Here are some more stats from the two-run i.e. one normal run and the other with minOffsetsperTrigger set:
| Run | Data Size | Number of Batch Runs | Number of Files |
| ------------- | ------------- |------------- |------------- |
| Normal Run | 54.2 GB | 320 | 21968 |
| Run with minOffsetsperTrigger | 54.2 GB | 120 | 12104 |
Closes#32653 from satishgopalani/SPARK-35312.
Authored-by: Satish Gopalani <satish.gopalani@pubmatic.com>
Signed-off-by: Jungtaek Lim <kabhwan.opensource@gmail.com>
### What changes were proposed in this pull request?
Override `getJDBCType` method in `MySQLDialect` so that `FloatType` is mapped to `FLOAT` instead of `REAL`
### Why are the changes needed?
MySQL treats `REAL` as a synonym to `DOUBLE` by default (see https://dev.mysql.com/doc/refman/8.0/en/numeric-types.html). Therefore, when creating a table with a column of `REAL` type, it will be created as `DOUBLE`. However, currently, `MySQLDialect` does not provide an implementation for `getJDBCType`, and will thus ultimately fall back to `JdbcUtils.getCommonJDBCType`, which maps `FloatType` to `REAL`. This change is needed so that we can properly map the `FloatType` to `FLOAT` for MySQL.
### Does this PR introduce _any_ user-facing change?
Prior to this PR, when writing a dataframe with a `FloatType` column to a MySQL table, it will create a `DOUBLE` column. After the PR, it will create a `FLOAT` column.
### How was this patch tested?
Added a test case in `JDBCSuite` that verifies the mapping.
Closes#32605 from mariosmeim-db/SPARK-35446.
Authored-by: Marios Meimaris <marios.meimaris@databricks.com>
Signed-off-by: Hyukjin Kwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR proposes fix the default value in Data Source Option page based on the Scala documentation.
### Why are the changes needed?
Some of the existing default value in Data Source Option page follow the Python documentation, which has `None` as the default value for all options.
### Does this PR introduce _any_ user-facing change?
Yes, the default value in the Data Source Option page is fixed (from `None` to proper default value)
- Before
<img width="361" alt="Screen Shot 2021-06-02 at 6 31 12 PM" src="https://user-images.githubusercontent.com/44108233/120456594-b8719f00-c3d0-11eb-9778-071ab2ba9f45.png">
- After
<img width="562" alt="Screen Shot 2021-06-02 at 6 32 47 PM" src="https://user-images.githubusercontent.com/44108233/120456844-f1117880-c3d0-11eb-9c7c-9dcd66776444.png">
### How was this patch tested?
Manually built the docs and checked one by one.
Closes#32745 from itholic/SPARK-35523.
Lead-authored-by: itholic <haejoon.lee@databricks.com>
Co-authored-by: Haejoon Lee <44108233+itholic@users.noreply.github.com>
Signed-off-by: Hyukjin Kwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR proposes to port Koalas documentation to PySpark documentation as its initial step.
It ports almost as is except these differences:
- Renamed import from `databricks.koalas` to `pyspark.pandas`.
- Renamed `to_koalas` -> `to_pandas_on_spark`
- Renamed `(Series|DataFrame).koalas` -> `(Series|DataFrame).pandas_on_spark`
- Added a `ps_` prefix in the RST file names of Koalas documentation
Other then that,
- Excluded `python/docs/build/html` in linter
- Fixed GA dependency installataion
### Why are the changes needed?
To document pandas APIs on Spark.
### Does this PR introduce _any_ user-facing change?
Yes, it adds new documentations.
### How was this patch tested?
Manually built the docs and checked the output.
Closes#32726 from HyukjinKwon/SPARK-35587.
Authored-by: Hyukjin Kwon <gurwls223@apache.org>
Signed-off-by: Hyukjin Kwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR proposes adding more methods to set data source option to `Data Source Option` page for each data source.
For example, Data Source Option page for JSON as below:
- Before
<img width="322" alt="Screen Shot 2021-06-03 at 10 51 54 AM" src="https://user-images.githubusercontent.com/44108233/120574245-eb13aa00-c459-11eb-9f81-0b356023bcb5.png">
- After
<img width="470" alt="Screen Shot 2021-06-03 at 10 52 21 AM" src="https://user-images.githubusercontent.com/44108233/120574253-ed760400-c459-11eb-9008-1f075e0b9267.png">
### Why are the changes needed?
To provide users various options when they set options for data source.
### Does this PR introduce _any_ user-facing change?
Yes, now the document provides more methods for setting options than before, as in above screen capture.
### How was this patch tested?
Manually built the docs and check one by one.
Closes#32757 from itholic/SPARK-35528.
Authored-by: itholic <haejoon.lee@databricks.com>
Signed-off-by: Hyukjin Kwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
In the PR, I propose to support special datetime values introduced by #25708 and by #25716 only in typed literals, and don't recognize them in parsing strings to dates/timestamps. The following string values are supported only in typed timestamp literals:
- `epoch [zoneId]` - `1970-01-01 00:00:00+00 (Unix system time zero)`
- `today [zoneId]` - midnight today.
- `yesterday [zoneId]` - midnight yesterday
- `tomorrow [zoneId]` - midnight tomorrow
- `now` - current query start time.
For example:
```sql
spark-sql> SELECT timestamp 'tomorrow';
2019-09-07 00:00:00
```
Similarly, the following special date values are supported only in typed date literals:
- `epoch [zoneId]` - `1970-01-01`
- `today [zoneId]` - the current date in the time zone specified by `spark.sql.session.timeZone`.
- `yesterday [zoneId]` - the current date -1
- `tomorrow [zoneId]` - the current date + 1
- `now` - the date of running the current query. It has the same notion as `today`.
For example:
```sql
spark-sql> SELECT date 'tomorrow' - date 'yesterday';
2
```
### Why are the changes needed?
In the current implementation, Spark supports the special date/timestamp value in any input strings casted to dates/timestamps that leads to the following problems:
- If executors have different system time, the result is inconsistent, and random. Column values depend on where the conversions were performed.
- The special values play the role of distributed non-deterministic functions though users might think of the values as constants.
### Does this PR introduce _any_ user-facing change?
Yes but the probability should be small.
### How was this patch tested?
By running existing test suites:
```
$ build/sbt "sql/testOnly org.apache.spark.sql.SQLQueryTestSuite -- -z interval.sql"
$ build/sbt "sql/testOnly org.apache.spark.sql.SQLQueryTestSuite -- -z date.sql"
$ build/sbt "sql/testOnly org.apache.spark.sql.SQLQueryTestSuite -- -z timestamp.sql"
$ build/sbt "test:testOnly *DateTimeUtilsSuite"
```
Closes#32714 from MaxGekk/remove-datetime-special-values.
Lead-authored-by: Max Gekk <max.gekk@gmail.com>
Co-authored-by: Maxim Gekk <max.gekk@gmail.com>
Signed-off-by: Max Gekk <max.gekk@gmail.com>
Fixed some places in cluster-overview that are obsolete (i.e. not mentioning Kubernetes), and also fixed the Yarn spark-submit sample command in submitting-applications.
### What changes were proposed in this pull request?
This is to fix the docs in "Cluster Overview" and "Submitting Applications" for places where Kubernetes is missed (mostly due to obsolete docs that haven't got updated) and where Yarn sample spark-submit command is incorrectly written.
### Why are the changes needed?
To help the Spark users who uses Kubernetes as cluster manager to have a correct idea when reading the "Cluster Overview" doc page. Also to make the sample spark-submit command for Yarn actually runnable in the "Submitting Applications" doc page, by removing the invalid comment after line continuation char `\`.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
No test, as this is doc fix.
Closes#32701 from huskysun/doc-fix.
Authored-by: Shiqi Sun <s.sun@salesforce.com>
Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
### What changes were proposed in this pull request?
This patch is a follow-up of SPARK-26848 (#23747). In SPARK-26848, we decided to open possibility to let end users set individual timestamp per partition. But in many cases, specifying timestamp represents the intention that we would want to go back to specific timestamp and reprocess records, which should be applied to all topics and partitions.
This patch proposes to provide a way to set a global timestamp across topic-partitions which the source is subscribing to, so that end users can set all offsets by specific timestamp easily. To provide the way to config the timestamp easier, the new options only receive "a" timestamp for start/end timestamp.
New options introduced in this PR:
* startingTimestamp
* endingTimestamp
All two options receive timestamp as string.
There're priorities for options regarding starting/ending offset as we will have three options for start offsets and another three options for end offsets. Priorities are following:
* starting offsets: startingTimestamp -> startingOffsetsByTimestamp -> startingOffsets
* ending offsets: startingTimestamp -> startingOffsetsByTimestamp -> startingOffsets
### Why are the changes needed?
Existing option to specify timestamp as offset is quite verbose if there're a lot of partitions across topics. Suppose there're 100s of partitions in a topic, the json should contain 100s of times of the same timestamp.
Also, the number of partitions can also change, which requires either:
* fixing the code if the json is statically created
* introducing the dependencies on Kafka client and deal with Kafka API on crafting json programmatically
Both approaches are even not "acceptable" if we're dealing with ad-hoc query; anyone doesn't want to write the code more complicated than the query itself. Flink [provides the option](https://ci.apache.org/projects/flink/flink-docs-release-1.13/docs/connectors/datastream/kafka/#kafka-consumers-start-position-configuration) to specify a timestamp for all topic-partitions like this PR, and even doesn't provide the option to specify the timestamp per topic-partition.
With this PR, end users are only required to provide a single timestamp value. No more complicated JSON format end users need to know about the structure.
### Does this PR introduce _any_ user-facing change?
Yes, this PR introduces two new options, described in above section.
Doc changes are following:
### How was this patch tested?
New UTs covering new functionalities. Also manually tested via simple batch & streaming queries.
Closes#32609 from HeartSaVioR/SPARK-29223-v2.
Authored-by: Jungtaek Lim <kabhwan.opensource@gmail.com>
Signed-off-by: Jungtaek Lim <kabhwan.opensource@gmail.com>
### What changes were proposed in this pull request?
This PR upgrades Dropwizard metrics to 4.2.0.
I also modified the corresponding links in `docs/monitoring.md`.
### Why are the changes needed?
The latest version was released last week and it contains some improvements.
https://github.com/dropwizard/metrics/releases/tag/v4.2.0
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Build succeeds and all the modified links are reachable.
Closes#32628 from sarutak/upgrade-dropwizard-4.2.0.
Authored-by: Kousuke Saruta <sarutak@oss.nttdata.com>
Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
### What changes were proposed in this pull request?
CTAS with location clause acts as an insert overwrite. This can cause problems when there are subdirectories within a location directory.
This causes some users to accidentally wipe out directories with very important data. We should not allow CTAS with location to a non-empty directory.
### Why are the changes needed?
Hive already handled this scenario: HIVE-11319
Steps to reproduce:
```scala
sql("""create external table `demo_CTAS`( `comment` string) PARTITIONED BY (`col1` string, `col2` string) STORED AS parquet location '/tmp/u1/demo_CTAS'""")
sql("""INSERT OVERWRITE TABLE demo_CTAS partition (col1='1',col2='1') VALUES ('abc')""")
sql("select* from demo_CTAS").show
sql("""create table ctas1 location '/tmp/u2/ctas1' as select * from demo_CTAS""")
sql("select* from ctas1").show
sql("""create table ctas2 location '/tmp/u2' as select * from demo_CTAS""")
```
Before the fix: Both create table operations will succeed. But values in table ctas1 will be replaced by ctas2 accidentally.
After the fix: `create table ctas2...` will throw `AnalysisException`:
```
org.apache.spark.sql.AnalysisException: CREATE-TABLE-AS-SELECT cannot create table with location to a non-empty directory /tmp/u2 . To allow overwriting the existing non-empty directory, set 'spark.sql.legacy.allowNonEmptyLocationInCTAS' to true.
```
### Does this PR introduce _any_ user-facing change?
Yes, if the location directory is not empty, CTAS with location will throw AnalysisException
```
sql("""create table ctas2 location '/tmp/u2' as select * from demo_CTAS""")
```
```
org.apache.spark.sql.AnalysisException: CREATE-TABLE-AS-SELECT cannot create table with location to a non-empty directory /tmp/u2 . To allow overwriting the existing non-empty directory, set 'spark.sql.legacy.allowNonEmptyLocationInCTAS' to true.
```
`CREATE TABLE AS SELECT` with non-empty `LOCATION` will throw `AnalysisException`. To restore the behavior before Spark 3.2, need to set `spark.sql.legacy.allowNonEmptyLocationInCTAS` to `true`. , default value is `false`.
Updated SQL migration guide.
### How was this patch tested?
Test case added in SQLQuerySuite.scala
Closes#32411 from vinodkc/br_fixCTAS_nonempty_dir.
Authored-by: Vinod KC <vinod.kc.in@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
The following two things are done in this PR.
* Add note about Jinja2 as a required dependency for document build.
* Add Jinja2 dependency for the document build to `spark-rm/Dockerfile`
### Why are the changes needed?
SPARK-35375(#32509) confined the version of Jinja to <3.0.0.
So it's good to note about it in `docs/README.md` and add the dependency to `spark-rm/Dockerfile`.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
I confimed that `make html` succeed under `python/docs` with the following command.
```
sudo pip install 'sphinx<3.1.0' mkdocs numpy pydata_sphinx_theme ipython nbsphinx numpydoc 'jinja2<3.0.0'
```
Closes#32573 from sarutak/required-module-for-python-doc.
Authored-by: Kousuke Saruta <sarutak@oss.nttdata.com>
Signed-off-by: Kousuke Saruta <sarutak@oss.nttdata.com>
### What changes were proposed in this pull request?
Fix incorrect statement that state is no longer needed in the event of executor failure and document that it is needed in the case of a flaky app causing occasional executor failure.
SO [discussion](https://stackoverflow.com/questions/67466878/can-spark-with-external-shuffle-service-use-saved-shuffle-files-in-the-event-of/67507439#67507439).
### Why are the changes needed?
To fix the documentation and guide users as to additional use case for the Shuffle Service.
### Does this PR introduce _any_ user-facing change?
Documentation only.
### How was this patch tested?
N/A.
Closes#32538 from chrisheaththomas/shuffle-service-and-executor-failure.
Authored-by: Chris Thomas <chrisheaththomas@hotmail.com>
Signed-off-by: Sean Owen <srowen@gmail.com>
### What changes were proposed in this pull request?
Add toc tag on monitoring.md
### Why are the changes needed?
fix doc
### Does this PR introduce _any_ user-facing change?
yes, the table of content of the monitoring page will be shown on the official doc site.
### How was this patch tested?
pass GA doc build
Closes#32545 from yaooqinn/minor.
Authored-by: Kent Yao <yao@apache.org>
Signed-off-by: Kent Yao <yao@apache.org>
### What changes were proposed in this pull request?
Generally, we would expect that x = y => hash( x ) = hash( y ). However +-0 hash to different values for floating point types.
```
scala> spark.sql("select hash(cast('0.0' as double)), hash(cast('-0.0' as double))").show
+-------------------------+--------------------------+
|hash(CAST(0.0 AS DOUBLE))|hash(CAST(-0.0 AS DOUBLE))|
+-------------------------+--------------------------+
| -1670924195| -853646085|
+-------------------------+--------------------------+
scala> spark.sql("select cast('0.0' as double) == cast('-0.0' as double)").show
+--------------------------------------------+
|(CAST(0.0 AS DOUBLE) = CAST(-0.0 AS DOUBLE))|
+--------------------------------------------+
| true|
+--------------------------------------------+
```
Here is an extract from IEEE 754:
> The two zeros are distinguishable arithmetically only by either division-byzero ( producing appropriately signed infinities ) or else by the CopySign function recommended by IEEE 754 /854. Infinities, SNaNs, NaNs and Subnormal numbers necessitate four more special cases
From this, I deduce that the hash function must produce the same result for 0 and -0.
### Why are the changes needed?
It is a correctness issue
### Does this PR introduce _any_ user-facing change?
This changes only affect to the hash function applied to -0 value in float and double types
### How was this patch tested?
Unit testing and manual testing
Closes#32496 from planga82/feature/spark35207_hashnegativezero.
Authored-by: Pablo Langa <soypab@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
In this PR I'm adding Structured Streaming Web UI state information documentation.
### Why are the changes needed?
Missing documentation.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
```
cd docs/
SKIP_API=1 bundle exec jekyll build
```
Manual webpage check.
Closes#32433 from gaborgsomogyi/SPARK-35311.
Authored-by: Gabor Somogyi <gabor.g.somogyi@gmail.com>
Signed-off-by: Jungtaek Lim <kabhwan.opensource@gmail.com>
### What changes were proposed in this pull request?
This proposes to document the available metrics for ExecutorAllocationManager in the Spark monitoring documentation.
### Why are the changes needed?
The ExecutorAllocationManager is instrumented with metrics using the Spark metrics system.
The relevant work is in SPARK-7007 and SPARK-33763
ExecutorAllocationManager metrics are currently undocumented.
### Does this PR introduce _any_ user-facing change?
This PR adds documentation only.
### How was this patch tested?
na
Closes#32500 from LucaCanali/followupMetricsDocSPARK33763.
Authored-by: Luca Canali <luca.canali@cern.ch>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
### What changes were proposed in this pull request?
This PR proposes to introduces three new configurations to limit the maximum number of jobs/stages/executors on the timeline view.
### Why are the changes needed?
If the number of items on the timeline view grows +1000, rendering can be significantly slow.
https://issues.apache.org/jira/browse/SPARK-35229
The maximum number of tasks on the timeline is already limited by `spark.ui.timeline.tasks.maximum` so l proposed to mitigate this issue with the same manner.
### Does this PR introduce _any_ user-facing change?
Yes. the maximum number of items shown on the timeline view is limited.
I proposed the default value 500 for jobs and stages, and 250 for executors.
A executor has at most 2 items (added and removed) 250 is chosen.
### How was this patch tested?
I manually confirm this change works with the following procedures.
```
# launch a cluster
$ bin/spark-shell --conf spark.ui.retainedDeadExecutors=300 --master "local-cluster[4, 1, 1024]"
// Confirm the maximum number of jobs
(1 to 1000).foreach { _ => sc.parallelize(List(1)).collect }
// Confirm the maximum number of stages
var df = sc.parallelize(1 to 2)
(1 to 1000).foreach { i => df = df.repartition(i % 5 + 1) }
df.collect
// Confirm the maximum number of executors
(1 to 300).foreach { _ => try sc.parallelize(List(1)).foreach { _ => System.exit(0) } catch { case e => }}
```
Screenshots here.
Closes#32381 from sarutak/mitigate-timeline-issue.
Authored-by: Kousuke Saruta <sarutak@oss.nttdata.com>
Signed-off-by: Gengliang Wang <ltnwgl@gmail.com>
### What changes were proposed in this pull request?
1. Extend Spark SQL parser to support parsing of:
- `INTERVAL YEAR TO MONTH` to `YearMonthIntervalType`
- `INTERVAL DAY TO SECOND` to `DayTimeIntervalType`
2. Assign new names to the ANSI interval types according to the SQL standard to be able to parse the names back by Spark SQL parser. Override the `typeName()` name of `YearMonthIntervalType`/`DayTimeIntervalType`.
### Why are the changes needed?
To be able to use new ANSI interval types in SQL. The SQL standard requires the types to be defined according to the rules:
```
<interval type> ::= INTERVAL <interval qualifier>
<interval qualifier> ::= <start field> TO <end field> | <single datetime field>
<start field> ::= <non-second primary datetime field> [ <left paren> <interval leading field precision> <right paren> ]
<end field> ::= <non-second primary datetime field> | SECOND [ <left paren> <interval fractional seconds precision> <right paren> ]
<primary datetime field> ::= <non-second primary datetime field | SECOND
<non-second primary datetime field> ::= YEAR | MONTH | DAY | HOUR | MINUTE
<interval fractional seconds precision> ::= <unsigned integer>
<interval leading field precision> ::= <unsigned integer>
```
Currently, Spark SQL supports only `YEAR TO MONTH` and `DAY TO SECOND` as `<interval qualifier>`.
### Does this PR introduce _any_ user-facing change?
Should not since the types has not been released yet.
### How was this patch tested?
By running the affected tests such as:
```
$ build/sbt "sql/testOnly *SQLQueryTestSuite -- -z interval.sql"
$ build/sbt "sql/testOnly *SQLQueryTestSuite -- -z datetime.sql"
$ build/sbt "test:testOnly *ExpressionTypeCheckingSuite"
$ build/sbt "sql/testOnly *SQLQueryTestSuite -- -z windowFrameCoercion.sql"
$ build/sbt "sql/testOnly *SQLQueryTestSuite -- -z literals.sql"
```
Closes#32409 from MaxGekk/parse-ansi-interval-types.
Authored-by: Max Gekk <max.gekk@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR extends `ADD FILE/JAR/ARCHIVE` commands to be able to take multiple path arguments like Hive.
### Why are the changes needed?
To make those commands more useful.
### Does this PR introduce _any_ user-facing change?
Yes. In the current implementation, those commands can take a path which contains whitespaces without enclose it by neither `'` nor `"` but after this change, users need to enclose such paths.
I've note this incompatibility in the migration guide.
### How was this patch tested?
New tests.
Closes#32205 from sarutak/add-multiple-files.
Authored-by: Kousuke Saruta <sarutak@oss.nttdata.com>
Signed-off-by: Kousuke Saruta <sarutak@oss.nttdata.com>
### What changes were proposed in this pull request?
This PR proposes to introduce a new JDBC option `refreshKrb5Config` which allows to reflect the change of `krb5.conf`.
### Why are the changes needed?
In the current master, JDBC datasources can't accept `refreshKrb5Config` which is defined in `Krb5LoginModule`.
So even if we change the `krb5.conf` after establishing a connection, the change will not be reflected.
The similar issue happens when we run multiple `*KrbIntegrationSuites` at the same time.
`MiniKDC` starts and stops every KerberosIntegrationSuite and different port number is recorded to `krb5.conf`.
Due to `SecureConnectionProvider.JDBCConfiguration` doesn't take `refreshKrb5Config`, KerberosIntegrationSuites except the first running one see the wrong port so those suites fail.
You can easily confirm with the following command.
```
build/sbt -Phive Phive-thriftserver -Pdocker-integration-tests "testOnly org.apache.spark.sql.jdbc.*KrbIntegrationSuite"
```
### Does this PR introduce _any_ user-facing change?
Yes. Users can set `refreshKrb5Config` to refresh krb5 relevant configuration.
### How was this patch tested?
New test.
Closes#32344 from sarutak/kerberos-refresh-issue.
Authored-by: Kousuke Saruta <sarutak@oss.nttdata.com>
Signed-off-by: Kousuke Saruta <sarutak@oss.nttdata.com>
Xinrong Meng
Gengliang Wang
Erik Krogen
Dhruvil Dave
Carlos Peña
Adam Binford
ulysses-you
yi.wu
Jungtaek Lim
HyukjinKwon
Wenchen Fan
Kousuke Saruta
RoryQi
Yuming Wang
Pawel Ptaszynski
allisonwang-db
gengjiaan
Satish Gopalani
Marios Meimaris
itholic
fornaix
Max Gekk
Shiqi Sun
Dongjoon Hyun
Vinod KC
Chris Thomas
Oleksandr Shevchenko
Kent Yao
Pablo Langa
Gabor Somogyi
Luca Canali
Ludovic Henry
