[SPARK-31291][SQL][TEST] SQLQueryTestSuite: Sharing test data and test tables among multiple test cases by beliefer · Pull Request #28060 · apache/spark

beliefer · 2020-03-28T04:39:59Z

What changes were proposed in this pull request?

SQLQueryTestSuite spend 35 minutes time to test.
I've listed the 10 test cases that took the longest time in the SQL module below.

Class	Spend time ↑	Skip	Pass	Total test case
SQLQueryTestSuite	35 minutes	1	230	231
TPCDSQuerySuite	3 minutes 8 seconds	0	156	156
SQLQuerySuite	2 minutes 52 seconds	0	185	185
DynamicPartitionPruningSuiteAEOff	1 minutes 52 seconds	0	22	22
DataFrameFunctionsSuite	1 minutes 37 seconds	0	102	102
DynamicPartitionPruningSuiteAEOn	1 minutes 24 seconds	0	22	22
DataFrameSuite	1 minutes 14 seconds	2	157	159
SubquerySuite	1 minutes 12 seconds	1	70	71
SingleLevelAggregateHashMapSuite	1 minutes 1 seconds	0	50	50
DataFrameAggregateSuite	59 seconds	0	50	50

I checked the code of SQLQueryTestSuite and found SQLQueryTestSuite load test data repeatedly.
This PR will improve the performance of SQLQueryTestSuite.

The total time run SQLQueryTestSuite before and after this PR show below.
Before

No	Time
1	20 minutes, 22 seconds
2	23 minutes, 21 seconds
3	21 minutes, 19 seconds
4	22 minutes, 26 seconds
5	20 minutes, 8 seconds

After

No	Time
1	20 minutes, 52 seconds
2	20 minutes, 47 seconds
3	20 minutes, 7 seconds
4	21 minutes, 10 seconds
5	20 minutes, 4 seconds

Why are the changes needed?

Improve the performance of SQLQueryTestSuite.

Does this PR introduce any user-facing change?

'No'.

How was this patch tested?

Jenkins test

SparkQA · 2020-03-28T04:52:04Z

Test build #120521 has finished for PR 28060 at commit 0511d99.

This patch fails to build.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2020-03-28T07:05:01Z

Test build #120522 has finished for PR 28060 at commit 8eac8d6.

This patch fails due to an unknown error code, -9.
This patch merges cleanly.
This patch adds no public classes.

wangyum · 2020-03-28T07:09:42Z

retest this please

maropu · 2020-03-28T07:50:03Z

Looks cool, thanks for the work, @beliefer ! btw, how long will SQLQueryTestSuite take with this fix? I just want to know the total running time.

SparkQA · 2020-03-28T11:16:24Z

Test build #120525 has finished for PR 28060 at commit 8eac8d6.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

beliefer · 2020-03-28T13:28:49Z

Looks cool, thanks for the work, @beliefer ! btw, how long will SQLQueryTestSuite take with this fix? I just want to know the total running time.

The total time after optimization is less than that before optimization by nearly one minute.

SparkQA · 2020-03-28T17:46:48Z

Test build #120530 has finished for PR 28060 at commit c392c03.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

maropu · 2020-03-30T02:36:06Z

+ if (testTables.contains("arraydata")) {
+ ((Seq(1, 2, 3), Seq(Seq(1, 2, 3))) :: (Seq(2, 3, 4), Seq(Seq(2, 3, 4))) :: Nil)
+ .toDF("arraycol", "nestedarraycol")
+ .createOrReplaceTempView("arraydata")


To avoid the overhead of per-session init, we cannot just move these local temp views into a session-independent place, e.g., global temp views?

The key point of conflict is not this. For example, test case A will create view testdata, and test case B will also create view testdata. However, the schema information of the two testdata is different. If the same session is shared globally, it will cause conflicts, especially in parallel execution.

If the same session is shared globally
I think that's not what @maropu means. We still create a fresh session for each testing file, but the testing views are created as global temp view, which are shared between all sessions.

@cloud-fan Thanks. I under @maropu 's suggestion now. I will try to use createGlobalTempView and shared these views between all sessions.

Sorry, but I missed your reply, @beliefer. Yea, that's what I wanted to say, thanks, @cloud-fan . I'll check later.

@cloud-fan
If use sparkSession.newSession(), test case failed. such as:
23:25:05.078 ERROR org.apache.spark.sql.SQLQueryTestSuite: Error using configs: [info] - operators.sql *** FAILED *** (5 seconds, 722 milliseconds) [info] operators.sql [info] Expected "struct<[(- key):int,(+ key):int]>", but got "struct<[]>" Schema did not match for query #4 [info] select -key, +key from testdata where key = 2: -- !query [info] select -key, +key from testdata where key = 2 [info] -- !query schema [info] struct<> [info] -- !query output [info] org.apache.spark.sql.AnalysisException [info] Table or view not found: testdata; line 1 pos 23 (SQLQueryTestSuite.scala:464) [info] org.scalatest.exceptions.TestFailedException: [info] at org.scalatest.Assertions.newAssertionFailedException(Assertions.scala:530)

like this? master...maropu:SPARK-31291

master...maropu:SPARK-31291
This is a method but will cause too many changes. After a discussion offline between @cloud-fan and me, I will try to use df.write.saveAsTable replace df.createTempView.

@cloud-fan @maropu I have used df.write.saveAsTable replace df.createTempView.
Because the origin temp view changed to tables, I have to regenerate some golden files.

dongjoon-hyun · 2020-03-30T03:01:02Z

Hi, All.
I converted this issue into a subtask of SPARK-25604.
FYI, technically, SPARK-25604 is already resolved by enhancing the test framework by parallelizing execution of the slow test suite and SQLQueryTestSuite and ThriftServerQueryTestSuite.scala are already parallelized in all SBT build and tests (including PRBuilder). So, this doesn't improve the total testing time in the SBT environment at all. The benefit of this PR is limited to only Maven environment.

cc @gatorsmile and @gengliangwang

gatorsmile · 2020-03-30T03:23:17Z

This was assigned to @beliefer after our offline talk. He is trying to find out the reasons why SQLQueryTestSuite took 35 minutes to finish. The time costs of each step/phase can help us locate the root cause. It would be interesting to know whether our compiler overhead are too big for these short queries.

beliefer · 2020-04-03T02:24:40Z

@dongjoon-hyun I think even in parallel execution, this PR will still help.

beliefer · 2020-04-07T01:57:31Z

cc @cloud-fan

SparkQA · 2020-04-08T18:25:18Z

Test build #120977 has finished for PR 28060 at commit 45bba4c.

This patch fails Spark unit tests.
This patch merges cleanly.
This patch adds no public classes.

cloud-fan · 2020-04-09T05:35:35Z

 -- !query output
-1	1
-2	2
+51	51


can we add a sort before limit?

I use repartition resolved the issue.

cloud-fan · 2020-04-09T05:36:47Z

 try {
 TimeZone.setDefault(originalTimeZone)
 Locale.setDefault(originalLocale)
+ unloadTestData(spark)


I'd prefer createTestTables and removeTestTables

cloud-fan · 2020-04-09T05:57:44Z

the change LGTM, can you regenerate the benchmark numbers?

beliefer · 2020-04-09T06:09:55Z

the change LGTM, can you regenerate the benchmark numbers?

OK.

SparkQA · 2020-04-09T07:05:01Z

Test build #121005 has finished for PR 28060 at commit af42f50.

This patch fails due to an unknown error code, -9.
This patch merges cleanly.
This patch adds no public classes.

beliefer · 2020-04-09T07:16:13Z

retest this please

SparkQA · 2020-04-09T12:02:23Z

Test build #121015 has finished for PR 28060 at commit af42f50.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

cloud-fan · 2020-04-09T12:16:34Z

thanks, merging to master/3.0!

…t tables among multiple test cases ### What changes were proposed in this pull request? `SQLQueryTestSuite` spend 35 minutes time to test. I've listed the 10 test cases that took the longest time in the `SQL` module below. Class | Spend time ↑ | Failure | Skip | Pass | Total test case -- | -- | -- | -- | -- | -- SQLQueryTestSuite | 35 minutes | 0 | 1 | 230 | 231 TPCDSQuerySuite | 3 minutes 8 seconds | 0 | 0 | 156 | 156 SQLQuerySuite | 2 minutes 52 seconds | 0 | 0 | 185 | 185 DynamicPartitionPruningSuiteAEOff | 1 minutes 52 seconds | 0 | 0 | 22 | 22 DataFrameFunctionsSuite | 1 minutes 37 seconds | 0 | 0 | 102 | 102 DynamicPartitionPruningSuiteAEOn | 1 minutes 24 seconds | 0 | 0 | 22 | 22 DataFrameSuite | 1 minutes 14 seconds | 0 | 2 | 157 | 159 SubquerySuite | 1 minutes 12 seconds | 0 | 1 | 70 | 71 SingleLevelAggregateHashMapSuite | 1 minutes 1 seconds | 0 | 0 | 50 | 50 DataFrameAggregateSuite | 59 seconds | 0 | 0 | 50 | 50 I checked the code of `SQLQueryTestSuite` and found `SQLQueryTestSuite` load test data repeatedly. This PR will improve the performance of `SQLQueryTestSuite`. The total time run `SQLQueryTestSuite` before and after this PR show below. Before No | Time -- | -- 1 | 20 minutes, 22 seconds 2 | 23 minutes, 21 seconds 3 | 21 minutes, 19 seconds 4 | 22 minutes, 26 seconds 5 | 20 minutes, 8 seconds After No | Time -- | -- 1 | 20 minutes, 52 seconds 2 | 20 minutes, 47 seconds 3 | 20 minutes, 7 seconds 4 | 21 minutes, 10 seconds 5 | 20 minutes, 4 seconds ### Why are the changes needed? Improve the performance of `SQLQueryTestSuite`. ### Does this PR introduce any user-facing change? 'No'. ### How was this patch tested? Jenkins test Closes#28060 from beliefer/avoid-load-test-data-repeatedly. Lead-authored-by: gengjiaan <gengjiaan@360.cn> Co-authored-by: beliefer <beliefer@163.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com> (cherry picked from commit 014d335) Signed-off-by: Wenchen Fan <wenchen@databricks.com>

beliefer · 2020-04-09T14:07:14Z

@cloud-fan @maropu Thanks for review this PR.
@dongjoon-hyun @gatorsmile @wangyum Thanks for your help!

…a and test tables among multiple test cases ### What changes were proposed in this pull request? This PR is related to #28060. `ThriftServerQueryTestSuite` spend 17 minutes time to test. I checked the code and found `ThriftServerQueryTestSuite` load test data repeatedly. I've listed all the test cases order by time with desc in the `hive-thriftserver` module below. Class | Spend time ↑ | Failure | Skip | Pass | Total test case -- | -- | -- | -- | -- | -- ThriftServerQueryTestSuite | 17 minutes | 0 | 15 | 140 | 155 CliSuite | 8 minutes 24 seconds | 0 | 0 | 24 | 24 SparkThriftServerProtocolVersionsSuite | 59 seconds | 0 | 0 | 210 | 210 HiveThriftBinaryServerSuite | 36 seconds | 0 | 1 | 21 | 22 SparkMetadataOperationSuite | 19 seconds | 0 | 0 | 7 | 7 HiveCliSessionStateSuite | 16 seconds | 0 | 0 | 2 | 2 SparkSQLEnvSuite | 16 seconds | 0 | 0 | 1 | 1 HiveThriftHttpServerSuite | 15 seconds | 0 | 0 | 3 | 3 SingleSessionSuite | 14 seconds | 0 | 0 | 3 | 3 JdbcConnectionUriSuite | 2.1 seconds | 0 | 0 | 1 | 1 ThriftServerWithSparkContextSuite | 1.4 seconds | 0 | 0 | 1 | 1 SparkExecuteStatementOperationSuite | 63 millseconds | 0 | 0 | 2 | 2 UISeleniumSuite | -1 millseconds | 0 | 1 | 0 | 1 I checked the code of `ThriftServerQueryTestSuite` and found `ThriftServerQueryTestSuite` load test data repeatedly. This PR will improve the performance of `ThriftServerQueryTestSuite`. Because #28060 provides `createTestTables`(https://github.com/apache/spark/blob/e42a3945acd614a26c7941a9eed161b500fb4520/sql/core/src/test/scala/org/apache/spark/sql/SQLQueryTestSuite.scala#L574) and `removeTestTables`(https://github.com/apache/spark/blob/e42a3945acd614a26c7941a9eed161b500fb4520/sql/core/src/test/scala/org/apache/spark/sql/SQLQueryTestSuite.scala#L666), this PR will still uses them. The total time run `ThriftServerQueryTestSuite` before and after this PR show below. Before No | Time -- | -- 1 | 18 minutes, 8 seconds 2 | 22 minutes, 44 seconds 3 | 17 minutes, 48 seconds 4 | 18 minutes, 30 seconds After No | Time -- | -- 1 | 16 minutes, 11 seconds 2 | 17 minutes, 19 seconds 3 | 18 minutes, 15 seconds 4 | 17 minutes, 27 seconds ### Why are the changes needed? Improve the performance of `ThriftServerQueryTestSuite`. ### Does this PR introduce any user-facing change? 'No'. ### How was this patch tested? Jenkins test Closes#28180 from beliefer/avoid-load-thrift-test-data-repeatedly. Authored-by: beliefer <beliefer@163.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>

…a and test tables among multiple test cases ### What changes were proposed in this pull request? This PR is related to #28060. `ThriftServerQueryTestSuite` spend 17 minutes time to test. I checked the code and found `ThriftServerQueryTestSuite` load test data repeatedly. I've listed all the test cases order by time with desc in the `hive-thriftserver` module below. Class | Spend time ↑ | Failure | Skip | Pass | Total test case -- | -- | -- | -- | -- | -- ThriftServerQueryTestSuite | 17 minutes | 0 | 15 | 140 | 155 CliSuite | 8 minutes 24 seconds | 0 | 0 | 24 | 24 SparkThriftServerProtocolVersionsSuite | 59 seconds | 0 | 0 | 210 | 210 HiveThriftBinaryServerSuite | 36 seconds | 0 | 1 | 21 | 22 SparkMetadataOperationSuite | 19 seconds | 0 | 0 | 7 | 7 HiveCliSessionStateSuite | 16 seconds | 0 | 0 | 2 | 2 SparkSQLEnvSuite | 16 seconds | 0 | 0 | 1 | 1 HiveThriftHttpServerSuite | 15 seconds | 0 | 0 | 3 | 3 SingleSessionSuite | 14 seconds | 0 | 0 | 3 | 3 JdbcConnectionUriSuite | 2.1 seconds | 0 | 0 | 1 | 1 ThriftServerWithSparkContextSuite | 1.4 seconds | 0 | 0 | 1 | 1 SparkExecuteStatementOperationSuite | 63 millseconds | 0 | 0 | 2 | 2 UISeleniumSuite | -1 millseconds | 0 | 1 | 0 | 1 I checked the code of `ThriftServerQueryTestSuite` and found `ThriftServerQueryTestSuite` load test data repeatedly. This PR will improve the performance of `ThriftServerQueryTestSuite`. Because #28060 provides `createTestTables`(https://github.com/apache/spark/blob/e42a3945acd614a26c7941a9eed161b500fb4520/sql/core/src/test/scala/org/apache/spark/sql/SQLQueryTestSuite.scala#L574) and `removeTestTables`(https://github.com/apache/spark/blob/e42a3945acd614a26c7941a9eed161b500fb4520/sql/core/src/test/scala/org/apache/spark/sql/SQLQueryTestSuite.scala#L666), this PR will still uses them. The total time run `ThriftServerQueryTestSuite` before and after this PR show below. Before No | Time -- | -- 1 | 18 minutes, 8 seconds 2 | 22 minutes, 44 seconds 3 | 17 minutes, 48 seconds 4 | 18 minutes, 30 seconds After No | Time -- | -- 1 | 16 minutes, 11 seconds 2 | 17 minutes, 19 seconds 3 | 18 minutes, 15 seconds 4 | 17 minutes, 27 seconds ### Why are the changes needed? Improve the performance of `ThriftServerQueryTestSuite`. ### Does this PR introduce any user-facing change? 'No'. ### How was this patch tested? Jenkins test Closes#28180 from beliefer/avoid-load-thrift-test-data-repeatedly. Authored-by: beliefer <beliefer@163.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com> (cherry picked from commit 2d3692e) Signed-off-by: Wenchen Fan <wenchen@databricks.com>

dongjoon-hyun · 2020-04-11T03:30:39Z

Hi, all.
This seems to break all Maven Jenkins jobs in both master and branch-3.0. The following is the example.

https://amplab.cs.berkeley.edu/jenkins/view/Spark%20QA%20Test%20(Dashboard)/job/spark-branch-3.0-test-maven-hadoop-2.7-hive-2.3/211/

org.apache.spark.sql.hive.thriftserver.ThriftServerQueryTestSuite *** ABORTED ***

Could you take a look?

dongjoon-hyun · 2020-04-11T03:38:12Z

 .format("csv")
 .options(Map("delimiter" -> "\t", "header" -> "false"))
 .schema("a int, b float")
 .load(testFile("test-data/postgresql/agg.data"))


This seems to be the root cause of failure.
java.lang.IllegalArgumentException: java.net.URISyntaxException: Relative path in absolute URI: jar:file:/home/jenkins/workspace/spark-branch-3.0-test-maven-hadoop-2.7-hive-2.3/sql/core/target/spark-sql_2.12-3.0.1-SNAPSHOT-tests.jar!/test-data/postgresql/agg.data

dongjoon-hyun · 2020-04-11T03:41:59Z

Since I found the root cause, I'll make a follow-up PR soon.

dongjoon-hyun · 2020-04-11T03:53:40Z

The one quick fix is copying the test file from jar:file:/home/jenkins/workspace/spark-branch-3.0-test-maven-hadoop-2.7-hive-2.3/sql/core/target/spark-sql_2.12-3.0.1-SNAPSHOT-tests.jar!/test-data/postgresql/agg.data to the local file.

Since this PR is about the performance, the fix will increase the test time a little for copying.

dongjoon-hyun · 2020-04-11T04:23:58Z

I made a follow-up PR to recover master and branch-3.0.

[SPARK-31291][SQL][TEST][FOLLOWUP] Fix resource loading error in ThriftServerQueryTestSuite #28186

…ftServerQueryTestSuite ### What changes were proposed in this pull request? [SPARK-31291](#28060) broke `ThriftServerQueryTestSuite` in Maven environment. This PR fixes it by copying the resource file from jars to local temp file. ### Why are the changes needed? To recover the Jenkins jobs in `master` and `branch-3.0`. - https://amplab.cs.berkeley.edu/jenkins/view/Spark%20QA%20Test%20(Dashboard)/job/spark-branch-3.0-test-maven-hadoop-2.7-hive-2.3/211/ ``` org.apache.spark.sql.hive.thriftserver.ThriftServerQueryTestSuite *** ABORTED *** ... java.lang.IllegalArgumentException: java.net.URISyntaxException: Relative path in absolute URI: jar:file:/home/jenkins/workspace/spark-branch-3.0-test-maven-hadoop-2.7-hive-2.3/sql/core/target/ spark-sql_2.12-3.0.1-SNAPSHOT-tests.jar!/test-data/postgresql/agg.data ``` ![Screen Shot 2020-04-10 at 9 54 28 PM](https://user-images.githubusercontent.com/9700541/79035702-f03ad900-7b75-11ea-9eee-0c1581a28838.png) ### Does this PR introduce any user-facing change? No. ### How was this patch tested? Pass the Jenkins with SBT and Maven. - [x] Sbt (`Test build #121117` #28186 (comment)) - [x] Maven (`Test build #121118` #28186 (comment)) Closes#28186 from dongjoon-hyun/SPARK-31291. Authored-by: Dongjoon Hyun <dongjoon@apache.org> Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>

…ftServerQueryTestSuite ### What changes were proposed in this pull request? [SPARK-31291](#28060) broke `ThriftServerQueryTestSuite` in Maven environment. This PR fixes it by copying the resource file from jars to local temp file. ### Why are the changes needed? To recover the Jenkins jobs in `master` and `branch-3.0`. - https://amplab.cs.berkeley.edu/jenkins/view/Spark%20QA%20Test%20(Dashboard)/job/spark-branch-3.0-test-maven-hadoop-2.7-hive-2.3/211/ ``` org.apache.spark.sql.hive.thriftserver.ThriftServerQueryTestSuite *** ABORTED *** ... java.lang.IllegalArgumentException: java.net.URISyntaxException: Relative path in absolute URI: jar:file:/home/jenkins/workspace/spark-branch-3.0-test-maven-hadoop-2.7-hive-2.3/sql/core/target/ spark-sql_2.12-3.0.1-SNAPSHOT-tests.jar!/test-data/postgresql/agg.data ``` ![Screen Shot 2020-04-10 at 9 54 28 PM](https://user-images.githubusercontent.com/9700541/79035702-f03ad900-7b75-11ea-9eee-0c1581a28838.png) ### Does this PR introduce any user-facing change? No. ### How was this patch tested? Pass the Jenkins with SBT and Maven. - [x] Sbt (`Test build #121117` #28186 (comment)) - [x] Maven (`Test build #121118` #28186 (comment)) Closes#28186 from dongjoon-hyun/SPARK-31291. Authored-by: Dongjoon Hyun <dongjoon@apache.org> Signed-off-by: Dongjoon Hyun <dongjoon@apache.org> (cherry picked from commit b4c438a) Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>

…t tables among multiple test cases ### What changes were proposed in this pull request? `SQLQueryTestSuite` spend 35 minutes time to test. I've listed the 10 test cases that took the longest time in the `SQL` module below. Class | Spend time ↑ | Failure | Skip | Pass | Total test case -- | -- | -- | -- | -- | -- SQLQueryTestSuite | 35 minutes | 0 | 1 | 230 | 231 TPCDSQuerySuite | 3 minutes 8 seconds | 0 | 0 | 156 | 156 SQLQuerySuite | 2 minutes 52 seconds | 0 | 0 | 185 | 185 DynamicPartitionPruningSuiteAEOff | 1 minutes 52 seconds | 0 | 0 | 22 | 22 DataFrameFunctionsSuite | 1 minutes 37 seconds | 0 | 0 | 102 | 102 DynamicPartitionPruningSuiteAEOn | 1 minutes 24 seconds | 0 | 0 | 22 | 22 DataFrameSuite | 1 minutes 14 seconds | 0 | 2 | 157 | 159 SubquerySuite | 1 minutes 12 seconds | 0 | 1 | 70 | 71 SingleLevelAggregateHashMapSuite | 1 minutes 1 seconds | 0 | 0 | 50 | 50 DataFrameAggregateSuite | 59 seconds | 0 | 0 | 50 | 50 I checked the code of `SQLQueryTestSuite` and found `SQLQueryTestSuite` load test data repeatedly. This PR will improve the performance of `SQLQueryTestSuite`. The total time run `SQLQueryTestSuite` before and after this PR show below. Before No | Time -- | -- 1 | 20 minutes, 22 seconds 2 | 23 minutes, 21 seconds 3 | 21 minutes, 19 seconds 4 | 22 minutes, 26 seconds 5 | 20 minutes, 8 seconds After No | Time -- | -- 1 | 20 minutes, 52 seconds 2 | 20 minutes, 47 seconds 3 | 20 minutes, 7 seconds 4 | 21 minutes, 10 seconds 5 | 20 minutes, 4 seconds ### Why are the changes needed? Improve the performance of `SQLQueryTestSuite`. ### Does this PR introduce any user-facing change? 'No'. ### How was this patch tested? Jenkins test Closesapache#28060 from beliefer/avoid-load-test-data-repeatedly. Lead-authored-by: gengjiaan <gengjiaan@360.cn> Co-authored-by: beliefer <beliefer@163.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>

…a and test tables among multiple test cases ### What changes were proposed in this pull request? This PR is related to apache#28060. `ThriftServerQueryTestSuite` spend 17 minutes time to test. I checked the code and found `ThriftServerQueryTestSuite` load test data repeatedly. I've listed all the test cases order by time with desc in the `hive-thriftserver` module below. Class | Spend time ↑ | Failure | Skip | Pass | Total test case -- | -- | -- | -- | -- | -- ThriftServerQueryTestSuite | 17 minutes | 0 | 15 | 140 | 155 CliSuite | 8 minutes 24 seconds | 0 | 0 | 24 | 24 SparkThriftServerProtocolVersionsSuite | 59 seconds | 0 | 0 | 210 | 210 HiveThriftBinaryServerSuite | 36 seconds | 0 | 1 | 21 | 22 SparkMetadataOperationSuite | 19 seconds | 0 | 0 | 7 | 7 HiveCliSessionStateSuite | 16 seconds | 0 | 0 | 2 | 2 SparkSQLEnvSuite | 16 seconds | 0 | 0 | 1 | 1 HiveThriftHttpServerSuite | 15 seconds | 0 | 0 | 3 | 3 SingleSessionSuite | 14 seconds | 0 | 0 | 3 | 3 JdbcConnectionUriSuite | 2.1 seconds | 0 | 0 | 1 | 1 ThriftServerWithSparkContextSuite | 1.4 seconds | 0 | 0 | 1 | 1 SparkExecuteStatementOperationSuite | 63 millseconds | 0 | 0 | 2 | 2 UISeleniumSuite | -1 millseconds | 0 | 1 | 0 | 1 I checked the code of `ThriftServerQueryTestSuite` and found `ThriftServerQueryTestSuite` load test data repeatedly. This PR will improve the performance of `ThriftServerQueryTestSuite`. Because apache#28060 provides `createTestTables`(https://github.com/apache/spark/blob/e42a3945acd614a26c7941a9eed161b500fb4520/sql/core/src/test/scala/org/apache/spark/sql/SQLQueryTestSuite.scala#L574) and `removeTestTables`(https://github.com/apache/spark/blob/e42a3945acd614a26c7941a9eed161b500fb4520/sql/core/src/test/scala/org/apache/spark/sql/SQLQueryTestSuite.scala#L666), this PR will still uses them. The total time run `ThriftServerQueryTestSuite` before and after this PR show below. Before No | Time -- | -- 1 | 18 minutes, 8 seconds 2 | 22 minutes, 44 seconds 3 | 17 minutes, 48 seconds 4 | 18 minutes, 30 seconds After No | Time -- | -- 1 | 16 minutes, 11 seconds 2 | 17 minutes, 19 seconds 3 | 18 minutes, 15 seconds 4 | 17 minutes, 27 seconds ### Why are the changes needed? Improve the performance of `ThriftServerQueryTestSuite`. ### Does this PR introduce any user-facing change? 'No'. ### How was this patch tested? Jenkins test Closesapache#28180 from beliefer/avoid-load-thrift-test-data-repeatedly. Authored-by: beliefer <beliefer@163.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>

…ftServerQueryTestSuite ### What changes were proposed in this pull request? [SPARK-31291](apache#28060) broke `ThriftServerQueryTestSuite` in Maven environment. This PR fixes it by copying the resource file from jars to local temp file. ### Why are the changes needed? To recover the Jenkins jobs in `master` and `branch-3.0`. - https://amplab.cs.berkeley.edu/jenkins/view/Spark%20QA%20Test%20(Dashboard)/job/spark-branch-3.0-test-maven-hadoop-2.7-hive-2.3/211/ ``` org.apache.spark.sql.hive.thriftserver.ThriftServerQueryTestSuite *** ABORTED *** ... java.lang.IllegalArgumentException: java.net.URISyntaxException: Relative path in absolute URI: jar:file:/home/jenkins/workspace/spark-branch-3.0-test-maven-hadoop-2.7-hive-2.3/sql/core/target/ spark-sql_2.12-3.0.1-SNAPSHOT-tests.jar!/test-data/postgresql/agg.data ``` ![Screen Shot 2020-04-10 at 9 54 28 PM](https://user-images.githubusercontent.com/9700541/79035702-f03ad900-7b75-11ea-9eee-0c1581a28838.png) ### Does this PR introduce any user-facing change? No. ### How was this patch tested? Pass the Jenkins with SBT and Maven. - [x] Sbt (`Test build #121117` apache#28186 (comment)) - [x] Maven (`Test build #121118` apache#28186 (comment)) Closesapache#28186 from dongjoon-hyun/SPARK-31291. Authored-by: Dongjoon Hyun <dongjoon@apache.org> Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>

maropu reviewed Mar 30, 2020
View reviewed changes

dongjoon-hyun added the TESTS label Mar 30, 2020

Optimize code
45bba4c

beliefer force-pushed the avoid-load-test-data-repeatedly branch from c392c03 to 45bba4cCompare April 8, 2020 15:03

cloud-fan reviewed Apr 9, 2020
View reviewed changes

beliefer added 2 commits April 9, 2020 13:41

Fix limit without order.
40bde64

Adjust method name
af42f50

maropu approved these changes Apr 9, 2020
View reviewed changes

beliefer changed the title ~~[SPARK-31291][SQL][TEST] Avoid load test data if test case not uses them~~[SPARK-31291][SQL][TEST] Sharing test data and test tables among multiple test casesApr 9, 2020

beliefer changed the title ~~[SPARK-31291][SQL][TEST] Sharing test data and test tables among multiple test cases~~[SPARK-31291][SQL][TEST] SQLQueryTestSuite: Sharing test data and test tables among multiple test casesApr 9, 2020

cloud-fan closed this in 014d335Apr 9, 2020

beliefer mentioned this pull request Apr 10, 2020
[SPARK-31406][SQL][TEST] ThriftServerQueryTestSuite: Sharing test data and test tables among multiple test cases. #28180
Closed

dongjoon-hyun reviewed Apr 11, 2020
View reviewed changes

dongjoon-hyun mentioned this pull request Apr 11, 2020
[SPARK-31291][SQL][TEST][FOLLOWUP] Fix resource loading error in ThriftServerQueryTestSuite #28186
Closed
2 tasks

dongjoon-hyun added the SQL label Apr 11, 2020

beliefer deleted the avoid-load-test-data-repeatedly branch April 23, 2024 07:18

Uh oh!

Conversation

beliefer commented Mar 28, 2020• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What changes were proposed in this pull request?

Why are the changes needed?

Does this PR introduce any user-facing change?

How was this patch tested?

Uh oh!

SparkQA commented Mar 28, 2020

Uh oh!

SparkQA commented Mar 28, 2020

Uh oh!

wangyum commented Mar 28, 2020

Uh oh!

maropu commented Mar 28, 2020

Uh oh!

SparkQA commented Mar 28, 2020

Uh oh!

beliefer commented Mar 28, 2020

Uh oh!

SparkQA commented Mar 28, 2020

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

belieferApr 8, 2020• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

belieferApr 8, 2020• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dongjoon-hyun commented Mar 30, 2020• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gatorsmile commented Mar 30, 2020

Uh oh!

beliefer commented Apr 3, 2020

Uh oh!

beliefer commented Apr 7, 2020

Uh oh!

SparkQA commented Apr 8, 2020

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cloud-fan commented Apr 9, 2020

Uh oh!

beliefer commented Apr 9, 2020

Uh oh!

SparkQA commented Apr 9, 2020

Uh oh!

beliefer commented Apr 9, 2020

Uh oh!

SparkQA commented Apr 9, 2020

Uh oh!

cloud-fan commented Apr 9, 2020

Uh oh!

beliefer commented Apr 9, 2020

Uh oh!

beliefer commented Mar 28, 2020•
edited
Loading

belieferApr 8, 2020•
edited
Loading

belieferApr 8, 2020•
edited
Loading

dongjoon-hyun commented Mar 30, 2020•
edited
Loading

dongjoon-hyun commented Apr 11, 2020•
edited
Loading

dongjoon-hyun commented Apr 11, 2020•
edited
Loading