### What changes were proposed in this pull request?
This:
1. switches Spark to use shaded Hadoop clients, namely hadoop-client-api and hadoop-client-runtime, for Hadoop 3.x.
2. upgrade built-in version for Hadoop 3.x to Hadoop 3.2.2
Note that for Hadoop 2.7, we'll still use the same modules such as hadoop-client.
In order to still keep default Hadoop profile to be hadoop-3.2, this defines the following Maven properties:
```
hadoop-client-api.artifact
hadoop-client-runtime.artifact
hadoop-client-minicluster.artifact
```
which default to:
```
hadoop-client-api
hadoop-client-runtime
hadoop-client-minicluster
```
but all switch to `hadoop-client` when the Hadoop profile is hadoop-2.7. A side affect from this is we'll import the same dependency multiple times. For this I have to disable Maven enforcer `banDuplicatePomDependencyVersions`.
Besides above, there are the following changes:
- explicitly add a few dependencies which are imported via transitive dependencies from Hadoop jars, but are removed from the shaded client jars.
- removed the use of `ProxyUriUtils.getPath` from `ApplicationMaster` which is a server-side/private API.
- modified `IsolatedClientLoader` to exclude `hadoop-auth` jars when Hadoop version is 3.x. This change should only matter when we're not sharing Hadoop classes with Spark (which is _mostly_ used in tests).
### Why are the changes needed?
Hadoop 3.2.2 is released with new features and bug fixes, so it's good for the Spark community to adopt it. However, latest Hadoop versions starting from Hadoop 3.2.1 have upgraded to use Guava 27+. In order to resolve Guava conflicts, this takes the approach by switching to shaded client jars provided by Hadoop. This also has the benefits of avoid pulling other 3rd party dependencies from Hadoop side so as to avoid more potential future conflicts.
### Does this PR introduce _any_ user-facing change?
When people use Spark with `hadoop-provided` option, they should make sure class path contains `hadoop-client-api` and `hadoop-client-runtime` jars. In addition, they may need to make sure these jars appear before other Hadoop jars in the order. Otherwise, classes may be loaded from the other non-shaded Hadoop jars and cause potential conflicts.
### How was this patch tested?
Relying on existing tests.
Closes#30701 from sunchao/test-hadoop-3.2.2.
Authored-by: Chao Sun <sunchao@apple.com>
Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
### What changes were proposed in this pull request?
The type-coercion for numeric types of average and sum is not necessary at all, as the resultType and sumType can prevent the overflow.
### Why are the changes needed?
rm unnecessary logic which may cause potential performance regressions
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
tpcds tests for plan
Closes#31079 from yaooqinn/SPARK-34037.
Authored-by: Kent Yao <yao@apache.org>
Signed-off-by: Liang-Chi Hsieh <viirya@gmail.com>
### What changes were proposed in this pull request?
Some local variables are declared as `var`, but they are never reassigned and should be declared as `val`, so this pr turn these from `var` to `val` except for `mockito` related cases.
### Why are the changes needed?
Use `val` instead of `var` when possible.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Pass the Jenkins or GitHub Action
Closes#31142 from LuciferYang/SPARK-33346.
Authored-by: yangjie01 <yangjie01@baidu.com>
Signed-off-by: Sean Owen <srowen@gmail.com>
### What changes were proposed in this pull request?
Add support for `orc.force.positional.evolution` config that forces ORC top level column matching by position rather than by name.
This does work in Hive:
```
> set orc.force.positional.evolution;
+--------------------------------------+
| set |
+--------------------------------------+
| orc.force.positional.evolution=true |
+--------------------------------------+
> create table t (c1 string, c2 string) stored as orc;
> insert into t values ('foo', 'bar');
> alter table t change c1 c3 string;
```
The orc file in this case contains the original `c1` and `c2` columns that doesn't match the metadata in HMS. But due to the positional evolution setting, Hive is capable to return all the data:
```
> select * from t;
+--------+--------+
| t.c3 | t.c2 |
+--------+--------+
| foo | bar |
+--------+--------+
```
Without this PR Spark returns `null`s for the renamed `c3` column.
After this PR Spark returns the data in `c3` column.
### Why are the changes needed?
Hive/ORC does support it.
### Does this PR introduce _any_ user-facing change?
Yes, we will support `orc.force.positional.evolution`.
### How was this patch tested?
New UT.
Closes#29737 from peter-toth/SPARK-32864-support-orc-forced-positional-evolution.
Lead-authored-by: Peter Toth <peter.toth@gmail.com>
Co-authored-by: Peter Toth <ptoth@cloudera.com>
Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
### What changes were proposed in this pull request?
On the read-side, we should respect the original data instead of trimming it first.
It brings extra overhead on the code-gen code side, trimming and padding for the same field, and it's also unnecessary and a bug
### Why are the changes needed?
bugfix and perf regression
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
new tests
Closes#31181 from yaooqinn/SPARK-34114.
Authored-by: Kent Yao <yao@apache.org>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This PR introduces a new property `spark.sql.cli.print.header` to let users change the behavior of printing header for spark-sql CLI by SET command.
### Why are the changes needed?
Like Hive CLI, spark-sql CLI accepts `hive.cli.print.header` property and we can change the behavior of printing header.
But spark-sql CLI doesn't allow users to change Hive specific configurations dynamically by SET command.
So, it's better to support the way to change the behavior by SET command.
### Does this PR introduce _any_ user-facing change?
Yes. Users can dynamically change the behavior by SET command.
### How was this patch tested?
I confirmed with the following commands/queries.
```
spark-sql> select (1) as a, (2) as b, (3) as c, (4) as d;
1 2 3 4
Time taken: 3.218 seconds, Fetched 1 row(s)
spark-sql> set spark.sql.cli.print.header=true;
key value
spark.sql.cli.print.header true
Time taken: 1.506 seconds, Fetched 1 row(s)
spark-sql> select (1) as a, (2) as b, (3) as c, (4) as d;
a b c d
1 2 3 4
Time taken: 0.79 seconds, Fetched 1 row(s)
```
Closes#31173 from sarutak/spark-sql-print-header.
Authored-by: Kousuke Saruta <sarutak@oss.nttdata.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR changes cache refreshing of v2 tables in v2 commands. In particular, v2 table dependents are not removed from the cache after this PR. Comparing to current implementation, we just clear cached data of all dependents and keep them in the cache. So, the next actions will fill in the cached data of the original v2 table and its dependents. In more details:
1. Add new method `recacheTable()` to `DataSourceV2Strategy` and pass it the exec node where need to recache table. New method uses `recacheByPlan` to refresh data cache of v2 tables, and keeps table dependents still cached **while clearing their caches**.
2. Simplify `invalidateCache` (and rename it `invalidateTableCache`) by retargeting it for only table cache invalidation.
3. Modify a test for `REFRESH TABLE` and check that v2 table dependent is still cached after refreshing the base table.
### Why are the changes needed?
1. This should improve user experience with table/view caching. For example, let's imagine that an user has cached v2 table and cached view based on the table. And the user passed the table to external library which drops/renames/adds partitions in the v2 table. Unfortunately, the user gets the view uncached after that even he/she hasn't uncached the view explicitly.
2. Improve code maintenance.
3. Reduce the number of calls to the Cache Manager when need to recache a table. Before the changes, `invalidateCache()` invokes the Cache Manager 3 times: `lookupCachedData()`, `uncacheQuery()` and `cacheQuery()`.
4. Also this should speed up table recaching.
### Does this PR introduce _any_ user-facing change?
From the view of the correctness of query results, there are no behavior changes but the changes might influence on consuming memory and query execution time.
### How was this patch tested?
By running the existing test suites for v2 the add/drop/rename partition commands.
Closes#31172 from MaxGekk/dsv2-recache-table.
Authored-by: Max Gekk <max.gekk@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This pr use `exists` or `forall` to simplify `filter + emptiness check`, it's semantically consistent, but looks simpler. The rule as follow:
- `seq.filter(p).size == 0)` -> `!seq.exists(p)`
- `seq.filter(p).length > 0` -> `seq.exists(p)`
- `seq.filterNot(p).isEmpty` -> `seq.forall(p)`
- `seq.filterNot(p).nonEmpty` -> `!seq.forall(p)`
### Why are the changes needed?
Code Simpilefications.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Pass the Jenkins or GitHub Action
Closes#31184 from LuciferYang/SPARK-34118.
Authored-by: yangjie01 <yangjie01@baidu.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This patch proposes to pull the test of `numKeys` metric into a separate test in `StateStoreSuite`.
### Why are the changes needed?
Right now in `StateStoreSuite`, the tests of get/put/remove/commit are mixed with `numKeys` metric test. I found it is flaky when I was testing with other `StateStore` implementation.
Current test logic is tightly bound to the in-memory map behavior of `HDFSBackedStateStore`. For example, put can immediately show up in the `numKeys` metric.
But for a `StateStore` implementation relying on external storage, e.g. RocksDB, the metric might be updated once the data is actually committed. And `StateStoreSuite` should be a common test suite for all kinds of StateStore implementations.
Specifically, we also are able to check these metrics after state store is updated (committed). So I think we can refactor the test a little bit to make it easier to incorporate other `StateStore` externally.
### Does this PR introduce _any_ user-facing change?
No, dev only.
### How was this patch tested?
Unit test.
Closes#31183 from viirya/SPARK-34116.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
### What changes were proposed in this pull request?
This PR aims to strip auto-generated cast. The main logic is:
1. Add tag if Cast is specified by user.
2. Wrap `PrettyAttribute` in usePrettyExpression.
### Why are the changes needed?
Make sql consistent with dsl. Here is an inconsistent example before this PR:
```
-- output field name: FLOOR(1)
spark.emptyDataFrame.select(floor(lit(1)))
-- output field name: FLOOR(CAST(1 AS DOUBLE))
spark.sql("select floor(1)")
```
Note that, we don't remove the `Cast` so the auto-generated `Cast` can still work. The only changed place is `usePrettyExpression`, we use `PrettyAttribute` replace `Cast` to give a better sql string.
### Does this PR introduce _any_ user-facing change?
Yes, the default field name may change.
### How was this patch tested?
Add test and pass exists test.
Closes#31034 from ulysses-you/SPARK-33989.
Authored-by: ulysses-you <ulyssesyou18@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
PartitionPruning push down pruningHasBenefit function into insertPredicate function to decrease calculate time
### Why are the changes needed?
to accelerate PartitionPruning prune calculate
### Does this PR introduce _any_ user-facing change?
NO
### How was this patch tested?
existed unit test
Closes#31122 from monkeyboy123/optimize-dynamic-pruning.
Authored-by: Dereck Li <monkeyboy.ljh@gmail.com>
Signed-off-by: Yuming Wang <yumwang@ebay.com>
### What changes were proposed in this pull request?
Fix a regression from https://github.com/apache/spark/pull/29959.
In Spark, the following file paths are considered as hidden paths and they are ignored on file reads:
1. starts with "_" and doesn't contain "="
2. starts with "."
However, after the refactoring PR https://github.com/apache/spark/pull/29959, the hidden paths are not filtered out on partition inference: https://github.com/apache/spark/pull/29959/files#r556432426
This PR is to fix the bug. To archive the goal, the method `InMemoryFileIndex.shouldFilterOut` is refactored as `HadoopFSUtils.shouldFilterOutPathName`
### Why are the changes needed?
Bugfix
### Does this PR introduce _any_ user-facing change?
Yes, it fixes a bug for reading file paths with partitions.
### How was this patch tested?
Unit test
Closes#31169 from gengliangwang/fileListingBug.
Authored-by: Gengliang Wang <gengliang.wang@databricks.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR aims to fix `MiMaExcludes` rule by moving SPARK-23429 from 2.4 to 3.0.
### Why are the changes needed?
SPARK-23429 was added at Apache Spark 3.0.0.
This should land on `master` and `branch-3.1` and `branch-3.0`.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Pass the MiMa rule.
Closes#31174 from dongjoon-hyun/SPARK-34103.
Authored-by: Dongjoon Hyun <dhyun@apple.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This is a followup PR for SPARK-33690 (#30647) .
In addition to the original PR, this PR intends to escape the following meta-characters in `Dataset#showString`.
* `\r` (carrige ret)
* `\f` (form feed)
* `\b` (backspace)
* `\u000B` (vertical tab)
* `\u0007` (bell)
### Why are the changes needed?
To avoid breaking the layout of `Dataset#showString`.
`\u0007` does not break the layout of `Dataset#showString` but it's noisy (beeps for each row) so it should be also escaped.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Modified the existing tests.
I also build the documents and check the generated html for `sql-migration-guide.md`.
Closes#31144 from sarutak/escape-metacharacters-in-getRows.
Authored-by: Kousuke Saruta <sarutak@oss.nttdata.com>
Signed-off-by: Sean Owen <srowen@gmail.com>
### What changes were proposed in this pull request?
<!--
Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue.
If possible, please consider writing useful notes for better and faster reviews in your PR. See the examples below.
1. If you refactor some codes with changing classes, showing the class hierarchy will help reviewers.
2. If you fix some SQL features, you can provide some references of other DBMSes.
3. If there is design documentation, please add the link.
4. If there is a discussion in the mailing list, please add the link.
-->
This PR adds a feature which supports 32-bit unicode escape in string literals like PostgreSQL or some modern programming languages do (e.g, Python3, C++11 and Rust).
In addition to the feature which supports 16-bit unicode escape like `"\u0041"`, users can express unicode characters like `"\U00020BB7"` with this change.
### Why are the changes needed?
<!--
Please clarify why the changes are needed. For instance,
1. If you propose a new API, clarify the use case for a new API.
2. If you fix a bug, you can clarify why it is a bug.
-->
Users can express unicode characters straightly without surrogate pair.
### Does this PR introduce _any_ user-facing change?
<!--
Note that it means *any* user-facing change including all aspects such as the documentation fix.
If yes, please clarify the previous behavior and the change this PR proposes - provide the console output, description and/or an example to show the behavior difference if possible.
If possible, please also clarify if this is a user-facing change compared to the released Spark versions or within the unreleased branches such as master.
If no, write 'No'.
-->
Yes. Users an express all the unicode characters straightly.
### How was this patch tested?
<!--
If tests were added, say they were added here. Please make sure to add some test cases that check the changes thoroughly including negative and positive cases if possible.
If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future.
If tests were not added, please describe why they were not added and/or why it was difficult to add.
-->
Added new assertions to the existing test case.
Closes#31096 from sarutak/32-bit-unicode-escape.
Authored-by: Kousuke Saruta <sarutak@oss.nttdata.com>
Signed-off-by: Sean Owen <srowen@gmail.com>
### What changes were proposed in this pull request?
There are some redundant collection conversion can be removed, for version compatibility, clean up these with Scala-2.13 profile.
### Why are the changes needed?
Remove redundant collection conversion
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- Pass the Jenkins or GitHub Action
- Manual test `core`, `graphx`, `mllib`, `mllib-local`, `sql`, `yarn`,`kafka-0-10` in Scala 2.13 passed
Closes#31125 from LuciferYang/SPARK-34068.
Authored-by: yangjie01 <yangjie01@baidu.com>
Signed-off-by: Sean Owen <srowen@gmail.com>
### What changes were proposed in this pull request?
This pr use `exists` to simplify `find + emptiness check`, it's semantically consistent, but looks simpler.
**Before**
```
seq.find(p).isDefined
or
seq.find(p).isEmpty
```
**After**
```
seq.exists(p)
or
!seq.exists(p)
```
### Why are the changes needed?
Code Simpilefications.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Pass the Jenkins or GitHub Action
Closes#31130 from LuciferYang/SPARK-34070.
Authored-by: yangjie01 <yangjie01@baidu.com>
Signed-off-by: Sean Owen <srowen@gmail.com>
### What changes were proposed in this pull request?
Add min threshold time speculation config
### Why are the changes needed?
When we turn on speculation with default configs we have the last 10% of the tasks subject to speculation. There are a lot of stages where the stage runs for few seconds to minutes. Also in general we don't want to speculate tasks that run within a minimum threshold. By setting a minimum threshold for speculation config gives us better control for speculative tasks
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Unit test
Closes#30710 from redsanket/SPARK-33741.
Lead-authored-by: schintap <schintap@verizonmedia.com>
Co-authored-by: Sanket Chintapalli <chintapalli.sanketreddy@gmail.com>
Signed-off-by: Thomas Graves <tgraves@apache.org>
### What changes were proposed in this pull request?
This changes `CatalogImpl.dropTempView` and `CatalogImpl.dropGlobalTempView` use analyzed logical plan instead of `viewDef` which is unresolved.
### Why are the changes needed?
Currently, `CatalogImpl.dropTempView` is implemented as following:
```scala
override def dropTempView(viewName: String): Boolean = {
sparkSession.sessionState.catalog.getTempView(viewName).exists { viewDef =>
sparkSession.sharedState.cacheManager.uncacheQuery(
sparkSession, viewDef, cascade = false)
sessionCatalog.dropTempView(viewName)
}
}
```
Here, the logical plan `viewDef` is not resolved, and when passing to `uncacheQuery`, it could fail at `sameResult` call, where canonicalized plan is compared. The error message looks like:
```
Invalid call to qualifier on unresolved object, tree: 'key
```
This can be reproduced via:
```scala
sql(s"CREATE TEMPORARY VIEW $v AS SELECT key FROM src LIMIT 10")
sql(s"CREATE TABLE $t AS SELECT * FROM src")
sql(s"CACHE TABLE $t")
dropTempTable(v)
```
### Does this PR introduce _any_ user-facing change?
The only user-facing change is that, previously `SQLContext.dropTempTable` may fail in the above scenario but will work with this fix.
### How was this patch tested?
Added new unit tests.
Closes#31136 from sunchao/SPARK-34076.
Authored-by: Chao Sun <sunchao@apple.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
#24595 introduced `private val runId: UUID = UUID.randomUUID` in `BroadcastExchangeExec` to cancel the broadcast execution in the Future when timeout happens. Since the runId is a random UUID instead of inheriting the job group id, when a SQL statement is cancelled, these broadcast sub-jobs are still executing. This PR uses the job group id of the outside thread as its `runId` to abort these broadcast sub-jobs when the SQL statement is cancelled.
### Why are the changes needed?
When broadcasting a table takes too long and the SQL statement is cancelled. However, the background Spark job is still running and it wastes resources.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Manually test.
Since broadcasting a table is too fast to cancel in UT, but it is very easy to verify manually:
1. Start a Spark thrift-server with less resource in YARN.
2. When the driver is running but no executors are launched, submit a SQL which will broadcast tables from beeline.
3. Cancel the SQL in beeline
Without the patch, broadcast sub-jobs won't be cancelled.
With this patch, broadcast sub-jobs will be cancelled.
Closes#31119 from LantaoJin/SPARK-34064.
Authored-by: LantaoJin <jinlantao@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
Increase test timeout.
### Why are the changes needed?
It's more reasonable to use 60s instead of 6s since many code place use the 60s in `TestUtils.waitUntilExecutorsUp`.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Pass exists test.
Closes#31166 from ulysses-you/SPARK-32850-FOLLOWUP.
Authored-by: ulysses-you <ulyssesyou18@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/133928/testReport/org.apache.spark.sql.execution/LogicalPlanTagInSparkPlanSuite/q41/
We can reduce more than 8000 bytes by removing the unnecessary CONCAT expression.
W/ this fix, for q41 in TPCDS with [Using TPCDS original definitions for char/varchar columns](https://github.com/apache/spark/pull/31012) applied, we can reduce the stage code-gen size from 22523 to 14369
```
14369 - 22523 = - 8154
```
### Why are the changes needed?
fix the perf regression(we need other improvements for q41 works), there will be a huge performance regression if codegen fails
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
modified uts
Closes#31150 from yaooqinn/SPARK-34086.
Authored-by: Kent Yao <yao@apache.org>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
Port the test added by https://github.com/apache/spark/pull/31112 to:
1. v1 In-Memory catalog for `ALTER TABLE .. DROP PARTITION`
2. v1 In-Memory and Hive external catalogs for `ALTER TABLE .. ADD PARTITION`
3. v1 In-Memory and Hive external catalogs for `ALTER TABLE .. RENAME PARTITION`
### Why are the changes needed?
To improve test coverage.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
By running the modified test suites:
```
$ build/sbt -Phive-2.3 -Phive-thriftserver "test:testOnly *.AlterTableAddPartitionSuite"
$ build/sbt -Phive-2.3 -Phive-thriftserver "test:testOnly *.AlterTableDropPartitionSuite"
$ build/sbt -Phive-2.3 -Phive-thriftserver "test:testOnly *.AlterTableRenamePartitionSuite"
```
Closes#31131 from MaxGekk/cache-stats-tests.
Authored-by: Max Gekk <max.gekk@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
`HadoopDelegationTokenManager.isServiceEnabled` is quite a time consuming operation which is called in `KafkaTokenUtil.needTokenUpdate` often which slowed down Kafka processing heavily. SPARK-33635 changed the if condition in order to overcome this issue when no delegation token is used but in case of delegation token usage the problem still exists. In this PR I'm caching the `HadoopDelegationTokenManager.isServiceEnabled` result in the `KafkaDataConsumer` instances which solves the issue. There would be another solution, namely caching the result inside `HadoopDelegationTokenManager` but since it's an object function and several application is running inside a JVM, different `SparkConf` instances will arrive. Caching the result per `SparkConf` instance would be an overkill.
### Why are the changes needed?
Kafka stream processing is slow with delegation token.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
* Existing unit tests
* In Kafka to Kafka live query I've double checked that `HadoopDelegationTokenManager.isServiceEnabled` call executed only when new `KafkaDataConsumer` created (new delegation token arrives or task failure).
Closes#31154 from gaborgsomogyi/SPARK-34090.
Authored-by: Gabor Somogyi <gabor.g.somogyi@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR is a follow-up of #29138 and #29195 to add more tests for `slice` function.
### Why are the changes needed?
The original PRs are missing tests with column-based arguments instead of literals.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Added tests and existing tests.
Closes#31159 from ueshin/issues/SPARK-32338/slice_tests.
Authored-by: Takuya UESHIN <ueshin@databricks.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Add shouldInterruptTaskThread check when kill barrier task.
### Why are the changes needed?
We should interrupt task thread if user set local property `SPARK_JOB_INTERRUPT_ON_CANCEL` to true.
### Does this PR introduce _any_ user-facing change?
Yes, task will be interrupted if user set `SPARK_JOB_INTERRUPT_ON_CANCEL` to true.
### How was this patch tested?
Add test.
Closes#31127 from ulysses-you/SPARK-34069.
Authored-by: ulysses-you <ulyssesyou18@gmail.com>
Signed-off-by: Mridul Muralidharan <mridul<at>gmail.com>
### What changes were proposed in this pull request?
Fix the setting issue of shuffle batch fetch in `ShuffledRowRDD`.
### Why are the changes needed?
Currently, we can not disable the shuffle batch fetch mode once the batch fetch mode has been enabled. This PR fixes the issue to make `ShuffledRowRDD` respects the `spark.sql.adaptive.fetchShuffleBlocksInBatch` at runtime.
### Does this PR introduce _any_ user-facing change?
Yes. Before this PR, users can not disable batch fetch if they enabled first. After this PR, they can.
### How was this patch tested?
Added unit test.
Closes#31155 from Ngone51/fix-batchfetch-set-issue.
Authored-by: yi.wu <yi.wu@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
Fix an issue in `ALTER TABLE .. ADD PARTITION` which happens when:
- A table doesn't have stats
- `spark.sql.statistics.size.autoUpdate.enabled` is `true`
In that case, `ALTER TABLE .. ADD PARTITION` does not update table stats automatically.
### Why are the changes needed?
The changes fix the issue demonstrated by the example:
```sql
spark-sql> create table tbl (col0 int, part int) partitioned by (part);
spark-sql> insert into tbl partition (part = 0) select 0;
spark-sql> set spark.sql.statistics.size.autoUpdate.enabled=true;
spark-sql> alter table tbl add partition (part = 1);
```
the `add partition` command should update table stats but it does not. There is no stats in the output of:
```
spark-sql> describe table extended tbl;
```
### Does this PR introduce _any_ user-facing change?
Yes. After the changes, `ALTER TABLE .. ADD PARTITION` updates stats even when a table does have them before the command:
```sql
spark-sql> alter table tbl add partition (part = 1);
spark-sql> describe table extended tbl;
col0 int NULL
part int NULL
# Partition Information
# col_name data_type comment
part int NULL
# Detailed Table Information
...
Statistics 2 bytes
```
### How was this patch tested?
By running new UT and existing test suites:
```
$ build/sbt -Phive-2.3 -Phive-thriftserver "test:testOnly *.AlterTableAddPartitionSuite"
```
Closes#31149 from MaxGekk/fix-stats-in-add-partition.
Authored-by: Max Gekk <max.gekk@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
Changes the configuration of the cancel workflow action to skip schedule/push events from canceling. This has the effect that duplicates of all direct pushes (master merges or direct pushes to the spark repository are not cancelled)
### What changes were proposed in this pull request?
Update to CI cancel policy to skip direct pushes. Duplicates will only be cancelled for Pull Requests.
### Why are the changes needed?
[Apparenlty](https://github.com/apache/spark/pull/31104#issuecomment-758318463) the aggressive behavior of the cancel action which also cancels duplicate master builds is too agressive for spark community. This change spares merges to master and scheduled builds from duplicate checking (as a result all merges to master will be always build to completion).
The previous behavior of the action was that in case of subsequent merges to master, only the latest one was guaranteed to complete. Other changes could be cancelled before they complete to save job queue.
### Does this PR introduce _any_ user-facing change?
No, except if the master builds were somehow facing the users (but it's unlikely taking into account the ASF release policy).
There was a potential that some changes that could be detected by specific master merge failing could be detected later (in one of the subsequent builds) which could make investigation of the root cause of failure a bit more difficult, because it could have been introduced in one of the commits between two completed builds merges. But this is at most impacting the timeline of release close to release itself, not the release itself.
### How was this patch tested?
This configuration parameter has been tested earlier in Airflow. We used it initially, but since our master builds are heavy and we have extensive tests in the PRs and investigation for failed builds is not as difficult we found that limiting the strain on Github Action and cancelling master builds was more important for the health of the whole ASF community and we removed that configuration.
Tested in https://github.com/potiuk/spark/runs/1688506527?check_suite_focus=true#step:2:46 where the action found other master builds in progress but did not add them as candidates to cancel.
Closes#31153 from potiuk/skip-schedule-push-branches.
Authored-by: Jarek Potiuk <potiuk@apache.org>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
In 0f8e5dd445, we partially fix the rule conflicts between `PaddingAndLengthCheckForCharVarchar` and `ResolveAggregateFunctions`, as error still exists in
sql like ```SELECT substr(v, 1, 2), sum(i) FROM t GROUP BY v ORDER BY substr(v, 1, 2)```
```sql
[info] Failed to analyze query: org.apache.spark.sql.AnalysisException: expression 'spark_catalog.default.t.`v`' is neither present in the group by, nor is it an aggregate function. Add to group by or wrap in first() (or first_value) if you don't care which value you get.;
[info] Project [substr(v, 1, 2)#100, sum(i)#101L]
[info] +- Sort [aggOrder#102 ASC NULLS FIRST], true
[info] +- !Aggregate [v#106], [substr(v#106, 1, 2) AS substr(v, 1, 2)#100, sum(cast(i#98 as bigint)) AS sum(i)#101L, substr(v#103, 1, 2) AS aggOrder#102
[info] +- SubqueryAlias spark_catalog.default.t
[info] +- Project [if ((length(v#97) <= 3)) v#97 else if ((length(rtrim(v#97, None)) > 3)) cast(raise_error(concat(input string of length , cast(length(v#97) as string), exceeds varchar type length limitation: 3)) as string) else rpad(rtrim(v#97, None), 3, ) AS v#106, i#98]
[info] +- Relation[v#97,i#98] parquet
[info]
[info] Project [substr(v, 1, 2)#100, sum(i)#101L]
[info] +- Sort [aggOrder#102 ASC NULLS FIRST], true
[info] +- !Aggregate [v#106], [substr(v#106, 1, 2) AS substr(v, 1, 2)#100, sum(cast(i#98 as bigint)) AS sum(i)#101L, substr(v#103, 1, 2) AS aggOrder#102
[info] +- SubqueryAlias spark_catalog.default.t
[info] +- Project [if ((length(v#97) <= 3)) v#97 else if ((length(rtrim(v#97, None)) > 3)) cast(raise_error(concat(input string of length , cast(length(v#97) as string), exceeds varchar type length limitation: 3)) as string) else rpad(rtrim(v#97, None), 3, ) AS v#106, i#98]
[info] +- Relation[v#97,i#98] parquet
```
We need to look recursively into children to find char/varchars.
In this PR, we try to resolve the full attributes including the original `Aggregate` expressions and the candidates in `SortOrder` together, then use the new re-resolved `Aggregate` expressions to determine which candidate in the `SortOrder` shall be pushed. This can avoid mismatch for the same attributes w/o this change, as the expressions returned by `executeSameContext` will change when `PaddingAndLengthCheckForCharVarchar` takes effects. W/ this change, the expressions can be matched correctly.
For those unmatched, w need to look recursively into children to find char/varchars instead of the expression itself only.
### Why are the changes needed?
bugfix
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
add new tests
Closes#31129 from yaooqinn/SPARK-34003-F.
Authored-by: Kent Yao <yao@apache.org>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
Add a legacy configuration `spark.sql.legacy.allowParameterlessCount` in case users need the parameterless count.
This is a follow-up for https://github.com/apache/spark/pull/30541.
### Why are the changes needed?
There can be some users depends on the legacy behavior. We need a legacy flag for it.
### Does this PR introduce _any_ user-facing change?
Yes, adding a legacy flag `spark.sql.legacy.allowParameterlessCount`.
### How was this patch tested?
Unit tests
Closes#31143 from gengliangwang/countLegacy.
Authored-by: Gengliang Wang <gengliang.wang@databricks.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Do not alter table stats if they are the same as in the catalog (at least since the recent retrieve).
### Why are the changes needed?
The changes reduce the number of calls to Hive external catalog.
### Does this PR introduce _any_ user-facing change?
Should not.
### How was this patch tested?
By running the modified test suites:
```
$ build/sbt -Phive-2.3 -Phive-thriftserver "test:testOnly *AlterTableAddPartitionSuite"
$ build/sbt -Phive-2.3 -Phive-thriftserver "test:testOnly *AlterTableDropPartitionSuite"
```
Closes#31135 from MaxGekk/optimize-updateTableStats.
Authored-by: Max Gekk <max.gekk@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
use a tmp model in OneVsRestModel.transform, to avoid calling directly setter of model
### Why are the changes needed?
params of model (submodels) should not be changed in transform
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
added testsuite
Closes#31086 from zhengruifeng/ovr_transform_tmp_model.
Authored-by: Ruifeng Zheng <ruifengz@foxmail.com>
Signed-off-by: Ruifeng Zheng <ruifengz@foxmail.com>
### What changes were proposed in this pull request?
Missing POD detection is extended by timestamp (and time limit) based check to avoid wrongfully detection of missing POD detection.
The two new timestamps:
- `fullSnapshotTs` is introduced for the `ExecutorPodsSnapshot` which only updated by the pod polling snapshot source
- `registrationTs` is introduced for the `ExecutorData` and it is initialized at the executor registration at the scheduler backend
Moreover a new config `spark.kubernetes.executor.missingPodDetectDelta` is used to specify the accepted delta between the two.
### Why are the changes needed?
Watching a POD (`ExecutorPodsWatchSnapshotSource`) only inform about single POD changes. This could wrongfully lead to detecting of missing PODs (PODs known by scheduler backend but missing from POD snapshots) by the executor POD lifecycle manager.
A key indicator of this error is seeing this log message:
> "The executor with ID [some_id] was not found in the cluster but we didn't get a reason why. Marking the executor as failed. The executor may have been deleted but the driver missed the deletion event."
So one of the problem is running the missing POD detection check even when a single POD is changed without having a full consistent snapshot about all the PODs (see `ExecutorPodsPollingSnapshotSource`).
The other problem could be the race between the executor POD lifecycle manager and the scheduler backend: so even in case of a having a full snapshot the registration at the scheduler backend could precede the snapshot polling (and processing of those polled snapshots).
### Does this PR introduce _any_ user-facing change?
Yes. When the POD is missing then the reason message explaining the executor's exit is extended with both timestamps (the polling time and the executor registration time) and even the new config is mentioned.
### How was this patch tested?
The existing unit tests are extended.
Closes#30675 from attilapiros/SPARK-33711.
Authored-by: “attilapiros” <piros.attila.zsolt@gmail.com>
Signed-off-by: Holden Karau <hkarau@apple.com>
### What changes were proposed in this pull request?
This is a follow-up to replace `version.toDouble > 2` with `version >= "2.0"`
### Why are the changes needed?
`toDouble` has some assumption and can cause `java.lang.NumberFormatException`.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Pass the CIs.
Closes#31134 from dongjoon-hyun/SPARK-33970-FOLLOWUP.
Authored-by: Dongjoon Hyun <dhyun@apple.com>
Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
### What changes were proposed in this pull request?
This patch adds a few description to SS doc about offset reset and data loss.
### Why are the changes needed?
During recent SS test, the behavior of gradual reducing input rows are confusing me. Comparing with Flink, I do not see a similar behavior. After looking into the code and doing some tests, I feel it is better to add some more description there in SS doc.
### Does this PR introduce _any_ user-facing change?
No, doc only.
### How was this patch tested?
Doc only.
Closes#31089 from viirya/ss-minor-5.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
### What changes were proposed in this pull request?
This patch fixes few issues when using encoders to serialize input/output in `ScalaUDF`.
### Why are the changes needed?
This fixes a bug when using encoders in Scala UDF. First, the output data type should be corrected to the corresponding data type of the object serializer. Second, `catalystConverter` should not serialize `Option[_]` as the ordinary row because in `ScalaUDF` case it is serialized to a column, not the top-level row. Otherwise, there will be a redundant `value` struct wrapping the serialized `Option[_]` object.
### Does this PR introduce _any_ user-facing change?
Yes, fixing a bug of `ScalaUDF`.
### How was this patch tested?
Unit test.
Closes#31103 from viirya/SPARK-34002.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
### What changes were proposed in this pull request?
For `AllJobsPage `class, `AllJobsPage` gets the schedulingMode of enumerated type by loading the `spark.scheduler.mode `configuration from Sparkconf, but an enumeration conversion error occurs when I set the value of this configuration to lowercase.
The reason for this problem is that the value of the `SchedulingMode `enumeration class is uppercase, which occurs when I configure `spark.scheduler.mode` to be lowercase.
I saw that the `#org.apache.spark.scheduler.TaskSchedulerImpl` class convert the `spark. scheduler.mode` value to uppercase, so I think it should be converted in `AllJobsPage `as well.
### Why are the changes needed?
An enumerated conversion error occurred with Spark when I set the value of this configuration to lowercase.
### How was this patch tested?
Existing tests.
Closes#31015 from yikf/master.
Authored-by: yikf <13468507104@163.com>
Signed-off-by: Sean Owen <srowen@gmail.com>
### What changes were proposed in this pull request?
According to https://github.com/apache/spark/pull/29966#discussion_r554514344
Use wrong name about suite file, this pr to fix this problem.
And change to use some fake ivy link for this test
### Why are the changes needed?
Follow file name rule
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
No
Closes#31118 from AngersZhuuuu/SPARK-33084-FOLLOW-UP.
Authored-by: angerszhu <angers.zhu@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This is kind of a followup of https://github.com/apache/spark/pull/31104 but I decided to track it separately with a separate JIRA.
Currently the jobs are being canceled in main repo branches. If a commit is merged, for example, to master branch before the test finishes, it cancels the previous builds. This is a problem because we cannot, for example, detect logical conflict properly. We should only cancel the jobs in PRs:
This PR proposes to don't do this in the main repo branch commits but only do it in PRs.
### Why are the changes needed?
- To keep the test coverage
- To run the test in the synced master branch instead of relying on the builds made in each PR with an outdated master branch
- To detect test failures from logical conflicts from merging two conflicting PRs at the same time.
### Does this PR introduce _any_ user-facing change?
No, dev-only.
### How was this patch tested?
I manually tested in
- https://github.com/HyukjinKwon/spark/pull/27
- https://github.com/HyukjinKwon/spark/pull/28
I added Yi Wu as a co-author since he helped verifying the current fix in the PR above.
I checked that it does not cancel in the main repo branch:
I checked it cancels in PRs:
Closes#31121 from HyukjinKwon/SPARK-34065.
Lead-authored-by: hyukjinkwon <gurwls223@apache.org>
Co-authored-by: yi.wu <yi.wu@databricks.com>
Co-authored-by: HyukjinKwon <gurwls223@apache.org>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR takes over https://github.com/apache/spark/pull/28780.
1. Counted the spilled memory size when creating the `UnsafeExternalSorter` with the existing `InMemorySorter`
2. Accumulate the `totalSpillBytes` when merging two `UnsafeExternalSorter`
### Why are the changes needed?
As mentioned in https://github.com/apache/spark/pull/28780:
> It happends when hash aggregate downgrades to sort based aggregate.
`UnsafeExternalSorter.createWithExistingInMemorySorter` calls spill on an `InMemorySorter` immediately, but the memory pointed by `InMemorySorter` is acquired by outside `BytesToBytesMap`, instead the allocatedPages in `UnsafeExternalSorter`. So the memory spill bytes metric is always 0, but disk bytes spill metric is right.
Besides, this PR also fixes the `UnsafeExternalSorter.merge` by accumulating the `totalSpillBytes` of two sorters. Thus, we can report the correct spilled size in `HashAggregateExec.finishAggregate`.
Issues can be reproduced by the following step by checking the SQL metrics in UI:
```
bin/spark-shell --driver-memory 512m --executor-memory 512m --executor-cores 1 --conf "spark.default.parallelism=1"
scala> sql("select id, count(1) from range(10000000) group by id").write.csv("/tmp/result.json")
```
Before:
<img width="200" alt="WeChatfe5146180d91015e03b9a27852e9a443" src="https://user-images.githubusercontent.com/16397174/103625414-e6fc6280-4f75-11eb-8b93-c55095bdb5b8.png">
After:
<img width="200" alt="WeChat42ab0e73c5fbc3b14c12ab85d232071d" src="https://user-images.githubusercontent.com/16397174/103625420-e8c62600-4f75-11eb-8e1f-6f5e8ab561b9.png">
### Does this PR introduce _any_ user-facing change?
Yes, users can see the correct spill metrics after this PR.
### How was this patch tested?
Tested manually and added UTs.
Closes#31035 from Ngone51/SPARK-31952.
Lead-authored-by: yi.wu <yi.wu@databricks.com>
Co-authored-by: wangguangxin.cn <wangguangxin.cn@bytedance.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
Fix canonicalisation of `HiveTableRelation` by normalisation of `CatalogTable`, and exclude table stats and temporary fields from the canonicalized plan.
### Why are the changes needed?
This fixes the issue demonstrated by the example below:
```scala
scala> spark.conf.set("spark.sql.statistics.size.autoUpdate.enabled", true)
scala> sql(s"CREATE TABLE tbl (id int, part int) USING hive PARTITIONED BY (part)")
scala> sql("INSERT INTO tbl PARTITION (part=0) SELECT 0")
scala> sql("INSERT INTO tbl PARTITION (part=1) SELECT 1")
scala> sql("CACHE TABLE tbl")
scala> sql("SELECT * FROM tbl").show(false)
+---+----+
|id |part|
+---+----+
|0 |0 |
|1 |1 |
+---+----+
scala> spark.catalog.isCached("tbl")
scala> sql("ALTER TABLE tbl DROP PARTITION (part=0)")
scala> spark.catalog.isCached("tbl")
res19: Boolean = false
```
`ALTER TABLE .. DROP PARTITION` must keep the table in the cache.
### Does this PR introduce _any_ user-facing change?
Yes. After the changes, the drop partition command keeps the table in the cache while updating table stats:
```scala
scala> sql("ALTER TABLE tbl DROP PARTITION (part=0)")
scala> spark.catalog.isCached("tbl")
res19: Boolean = true
```
### How was this patch tested?
By running new UT in `AlterTableDropPartitionSuite`.
Closes#31112 from MaxGekk/fix-caching-hive-table-2.
Authored-by: Max Gekk <max.gekk@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
Replace `USING parquet` by `$defaultUsing` which is `USING parquet` for v1 In-Memory catalog and `USING hive` for v1 Hive external catalog.
### Why are the changes needed?
The PR https://github.com/apache/spark/pull/31101 added UT test but it checks only v1 In-Memory catalog. This PR runs this test for Hive external catalog as well to improve test coverage.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
By running the affected test suites:
```
$ build/sbt -Phive-2.3 -Phive-thriftserver "test:testOnly *.AlterTableAddPartitionSuite"
```
Closes#31117 from MaxGekk/add-partition-refresh-cache-2-followup-2.
Authored-by: Max Gekk <max.gekk@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This pr add test default partition in metastoredirectsql.
### Why are the changes needed?
Improve test.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
N/A
Closes#31109 from wangyum/SPARK-33970.
Authored-by: Yuming Wang <yumwang@ebay.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Currently, there are many DDL commands where the position of the unresolved identifiers are incorrect:
```
scala> sql("DROP TABLE unknown")
org.apache.spark.sql.AnalysisException: Table or view not found: unknown; line 1 pos 0;
```
, whereas the `pos` should be `11`.
This PR proposes to fix this issue for commands using `UnresolvedTableOrView`:
```
DROP TABLE unknown
DESCRIBE TABLE unknown
ANALYZE TABLE unknown COMPUTE STATISTICS
ANALYZE TABLE unknown COMPUTE STATISTICS FOR COLUMNS col
ANALYZE TABLE unknown COMPUTE STATISTICS FOR ALL COLUMNS
SHOW CREATE TABLE unknown
REFRESH TABLE unknown
SHOW COLUMNS FROM unknown
SHOW COLUMNS FROM unknown IN db
ALTER TABLE unknown RENAME TO t
ALTER VIEW unknown RENAME TO v
```
### Why are the changes needed?
To fix a bug.
### Does this PR introduce _any_ user-facing change?
Yes, now the above example will print the following:
```
org.apache.spark.sql.AnalysisException: Table or view not found: unknown; line 1 pos 11;
```
### How was this patch tested?
Add a new test.
Closes#31106 from imback82/unresolved_table_or_view_message.
Authored-by: Terry Kim <yuminkim@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
Improve flaky NaiveBayes test
Current test may sometimes fail under different BLAS library. Due to some absTol check. Error like
```
Expected 0.7 and 0.6485507246376814 to be within 0.05 using absolute tolerance...
```
* Change absTol to relTol: The `absTol 0.05` in some cases (such as compare 0.1 and 0.05) is a big difference
* Remove the `exp` when comparing params. The `exp` will amplify the relative error.
### Why are the changes needed?
Flaky test
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
N/A
Closes#31004 from WeichenXu123/improve_bayes_tests.
Authored-by: Weichen Xu <weichen.xu@databricks.com>
Signed-off-by: Ruifeng Zheng <ruifengz@foxmail.com>
