first try to simplify the shutdown thread for decommissioning #7
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Author
|
I checked this log line too: |
Owner
|
So this just introduces more race conditions, I’m not seeing how this is better. For one thing the migration can be marked as finished while the task is still running and then the task goes away but stores some blocks. I know the current logic is complicated but this is just worse. |
Author
|
Regarding
From 1) and 2) I can can see only a small window when one task can be scheduled in the very beginning of the shutdown process and the sleep is about to decrease this (meanwhile migration thread is working to move the files away). Also when decommissioning is triggered during task running there will be no new task scheduled as the executor is already busy (this will be the case most of the time). Did I miss something? Please give me an example to help me see what I missed. I tried to illustrate it with a diagram: A: safe time period to start a new task |
Owner
|
I think using a sleep to attempt to order the events baking in a flaky design that we already have a better alternative to. This just seems worse. The best case of this code is the worst case of the other code. |
Author
|
I agree with your concern regarding the sleep I would like to avoid it meanwhile keeping the code a bit more simple. |
holdenk
pushed a commit
that referenced
this pull request
Apr 1, 2024
…n properly
### What changes were proposed in this pull request?
Make `ResolveRelations` handle plan id properly
### Why are the changes needed?
bug fix for Spark Connect, it won't affect classic Spark SQL
before this PR:
```
from pyspark.sql import functions as sf
spark.range(10).withColumn("value_1", sf.lit(1)).write.saveAsTable("test_table_1")
spark.range(10).withColumnRenamed("id", "index").withColumn("value_2", sf.lit(2)).write.saveAsTable("test_table_2")
df1 = spark.read.table("test_table_1")
df2 = spark.read.table("test_table_2")
df3 = spark.read.table("test_table_1")
join1 = df1.join(df2, on=df1.id==df2.index).select(df2.index, df2.value_2)
join2 = df3.join(join1, how="left", on=join1.index==df3.id)
join2.schema
```
fails with
```
AnalysisException: [CANNOT_RESOLVE_DATAFRAME_COLUMN] Cannot resolve dataframe column "id". It's probably because of illegal references like `df1.select(df2.col("a"))`. SQLSTATE: 42704
```
That is due to existing plan caching in `ResolveRelations` doesn't work with Spark Connect
```
=== Applying Rule org.apache.spark.sql.catalyst.analysis.Analyzer$ResolveRelations ===
'[apache#12]Join LeftOuter, '`==`('index, 'id) '[apache#12]Join LeftOuter, '`==`('index, 'id)
!:- '[apache#9]UnresolvedRelation [test_table_1], [], false :- '[apache#9]SubqueryAlias spark_catalog.default.test_table_1
!+- '[apache#11]Project ['index, 'value_2] : +- 'UnresolvedCatalogRelation `spark_catalog`.`default`.`test_table_1`, [], false
! +- '[apache#10]Join Inner, '`==`('id, 'index) +- '[apache#11]Project ['index, 'value_2]
! :- '[#7]UnresolvedRelation [test_table_1], [], false +- '[apache#10]Join Inner, '`==`('id, 'index)
! +- '[#8]UnresolvedRelation [test_table_2], [], false :- '[apache#9]SubqueryAlias spark_catalog.default.test_table_1
! : +- 'UnresolvedCatalogRelation `spark_catalog`.`default`.`test_table_1`, [], false
! +- '[#8]SubqueryAlias spark_catalog.default.test_table_2
! +- 'UnresolvedCatalogRelation `spark_catalog`.`default`.`test_table_2`, [], false
Can not resolve 'id with plan 7
```
`[#7]UnresolvedRelation [test_table_1], [], false` was wrongly resolved to the cached one
```
:- '[apache#9]SubqueryAlias spark_catalog.default.test_table_1
+- 'UnresolvedCatalogRelation `spark_catalog`.`default`.`test_table_1`, [], false
```
### Does this PR introduce _any_ user-facing change?
yes, bug fix
### How was this patch tested?
added ut
### Was this patch authored or co-authored using generative AI tooling?
ci
Closes apache#45214 from zhengruifeng/connect_fix_read_join.
Authored-by: Ruifeng Zheng <ruifengz@apache.org>
Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
holdenk
pushed a commit
that referenced
this pull request
Nov 7, 2025
…onicalized expressions
### What changes were proposed in this pull request?
Make PullOutNonDeterministic use canonicalized expressions to dedup group and aggregate expressions. This affects pyspark udfs in particular. Example:
```
from pyspark.sql.functions import col, avg, udf
pythonUDF = udf(lambda x: x).asNondeterministic()
spark.range(10)\
.selectExpr("id", "id % 3 as value")\
.groupBy(pythonUDF(col("value")))\
.agg(avg("id"), pythonUDF(col("value")))\
.explain(extended=True)
```
Currently results in a plan like this:
```
Aggregate [_nondeterministic#15](apache#15), [_nondeterministic#15 AS dummyNondeterministicUDF(value)apache#12, avg(id#0L) AS avg(id)apache#13, dummyNondeterministicUDF(value#6L)#8 AS dummyNondeterministicUDF(value)apache#14](apache#15%20AS%20dummyNondeterministicUDF(value)apache#12,%20avg(id#0L)%20AS%20avg(id)apache#13,%20dummyNondeterministicUDF(value#6L)#8%20AS%20dummyNondeterministicUDF(value)apache#14)
+- Project [id#0L, value#6L, dummyNondeterministicUDF(value#6L)#7 AS _nondeterministic#15](#0L,%20value#6L,%20dummyNondeterministicUDF(value#6L)#7%20AS%20_nondeterministic#15)
+- Project [id#0L, (id#0L % cast(3 as bigint)) AS value#6L](#0L,%20(id#0L%20%%20cast(3%20as%20bigint))%20AS%20value#6L)
+- Range (0, 10, step=1, splits=Some(2))
```
and then it throws:
```
[[MISSING_AGGREGATION] The non-aggregating expression "value" is based on columns which are not participating in the GROUP BY clause. Add the columns or the expression to the GROUP BY, aggregate the expression, or use "any_value(value)" if you do not care which of the values within a group is returned. SQLSTATE: 42803
```
- how canonicalized fixes this:
- nondeterministic PythonUDF expressions always have distinct resultIds per udf
- The fix is to canonicalize the expressions when matching. Canonicalized means that we're setting the resultIds to -1, allowing us to dedup the PythonUDF expressions.
- for deterministic UDFs, this rule does not apply and "Post Analysis" batch extracts and deduplicates the expressions, as expected
### Why are the changes needed?
- the output of the query with the fix applied still makes sense - the nondeterministic UDF is invoked only once, in the project.
### Does this PR introduce _any_ user-facing change?
Yes, it's additive, it enables queries to run that previously threw errors.
### How was this patch tested?
- added unit test
### Was this patch authored or co-authored using generative AI tooling?
No
Closes apache#52061 from benrobby/adhoc-fix-pull-out-nondeterministic.
Authored-by: Ben Hurdelhey <ben.hurdelhey@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
holdenk
pushed a commit
that referenced
this pull request
Nov 7, 2025
### What changes were proposed in this pull request? This PR proposes to add `doCanonicalize` function for DataSourceV2ScanRelation. The implementation is similar to [the one in BatchScanExec](https://github.com/apache/spark/blob/master/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/BatchScanExec.scala#L150), as well as the [the one in LogicalRelation](https://github.com/apache/spark/blob/master/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/LogicalRelation.scala#L52). ### Why are the changes needed? Query optimization rules such as MergeScalarSubqueries check if two plans are identical by [comparing their canonicalized form](https://github.com/apache/spark/blob/master/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/MergeScalarSubqueries.scala#L219). For DSv2, for physical plan, the canonicalization goes down in the child hierarchy to the BatchScanExec, which [has a doCanonicalize function](https://github.com/apache/spark/blob/master/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/BatchScanExec.scala#L150); for logical plan, the canonicalization goes down to the DataSourceV2ScanRelation, which, however, does not have a doCanonicalize function. As a result, two logical plans who are semantically identical are not identified. Moreover, for reference, [DSv1 LogicalRelation](https://github.com/apache/spark/blob/master/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/LogicalRelation.scala#L52) also has `doCanonicalize()`. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? A new unit test is added to show that `MergeScalarSubqueries` is working for DataSourceV2ScanRelation. For a query ```sql select (select max(i) from df) as max_i, (select min(i) from df) as min_i ``` Before introducing the canonicalization, the plan is ``` == Parsed Logical Plan == 'Project [scalar-subquery#2 [] AS max_i#3, scalar-subquery#4 [] AS min_i#5] : :- 'Project [unresolvedalias('max('i))] : : +- 'UnresolvedRelation [df], [], false : +- 'Project [unresolvedalias('min('i))] : +- 'UnresolvedRelation [df], [], false +- OneRowRelation == Analyzed Logical Plan == max_i: int, min_i: int Project [scalar-subquery#2 [] AS max_i#3, scalar-subquery#4 [] AS min_i#5] : :- Aggregate [max(i#0) AS max(i)#7] : : +- SubqueryAlias df : : +- View (`df`, [i#0, j#1]) : : +- RelationV2[i#0, j#1] class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5 : +- Aggregate [min(i#10) AS min(i)apache#9] : +- SubqueryAlias df : +- View (`df`, [i#10, j#11]) : +- RelationV2[i#10, j#11] class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5 +- OneRowRelation == Optimized Logical Plan == Project [scalar-subquery#2 [] AS max_i#3, scalar-subquery#4 [] AS min_i#5] : :- Aggregate [max(i#0) AS max(i)#7] : : +- Project [i#0] : : +- RelationV2[i#0, j#1] class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5 : +- Aggregate [min(i#10) AS min(i)apache#9] : +- Project [i#10] : +- RelationV2[i#10, j#11] class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5 +- OneRowRelation == Physical Plan == AdaptiveSparkPlan isFinalPlan=true +- == Final Plan == ResultQueryStage 0 +- *(1) Project [Subquery subquery#2, [id=apache#32] AS max_i#3, Subquery subquery#4, [id=apache#33] AS min_i#5] : :- Subquery subquery#2, [id=apache#32] : : +- AdaptiveSparkPlan isFinalPlan=true +- == Final Plan == ResultQueryStage 1 +- *(2) HashAggregate(keys=[], functions=[max(i#0)], output=[max(i)#7]) +- ShuffleQueryStage 0 +- Exchange SinglePartition, ENSURE_REQUIREMENTS, [plan_id=58] +- *(1) HashAggregate(keys=[], functions=[partial_max(i#0)], output=[max#14]) +- *(1) Project [i#0] +- BatchScan class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5[i#0, j#1] class org.apache.spark.sql.connector.SimpleDataSourceV2$MyScanBuilder RuntimeFilters: [] +- == Initial Plan == HashAggregate(keys=[], functions=[max(i#0)], output=[max(i)#7]) +- Exchange SinglePartition, ENSURE_REQUIREMENTS, [plan_id=19] +- HashAggregate(keys=[], functions=[partial_max(i#0)], output=[max#14]) +- Project [i#0] +- BatchScan class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5[i#0, j#1] class org.apache.spark.sql.connector.SimpleDataSourceV2$MyScanBuilder RuntimeFilters: [] : +- Subquery subquery#4, [id=apache#33] : +- AdaptiveSparkPlan isFinalPlan=true +- == Final Plan == ResultQueryStage 1 +- *(2) HashAggregate(keys=[], functions=[min(i#10)], output=[min(i)apache#9]) +- ShuffleQueryStage 0 +- Exchange SinglePartition, ENSURE_REQUIREMENTS, [plan_id=63] +- *(1) HashAggregate(keys=[], functions=[partial_min(i#10)], output=[min#15]) +- *(1) Project [i#10] +- BatchScan class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5[i#10, j#11] class org.apache.spark.sql.connector.SimpleDataSourceV2$MyScanBuilder RuntimeFilters: [] +- == Initial Plan == HashAggregate(keys=[], functions=[min(i#10)], output=[min(i)apache#9]) +- Exchange SinglePartition, ENSURE_REQUIREMENTS, [plan_id=30] +- HashAggregate(keys=[], functions=[partial_min(i#10)], output=[min#15]) +- Project [i#10] +- BatchScan class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5[i#10, j#11] class org.apache.spark.sql.connector.SimpleDataSourceV2$MyScanBuilder RuntimeFilters: [] +- *(1) Scan OneRowRelation[] +- == Initial Plan == Project [Subquery subquery#2, [id=apache#32] AS max_i#3, Subquery subquery#4, [id=apache#33] AS min_i#5] : :- Subquery subquery#2, [id=apache#32] : : +- AdaptiveSparkPlan isFinalPlan=true +- == Final Plan == ResultQueryStage 1 +- *(2) HashAggregate(keys=[], functions=[max(i#0)], output=[max(i)#7]) +- ShuffleQueryStage 0 +- Exchange SinglePartition, ENSURE_REQUIREMENTS, [plan_id=58] +- *(1) HashAggregate(keys=[], functions=[partial_max(i#0)], output=[max#14]) +- *(1) Project [i#0] +- BatchScan class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5[i#0, j#1] class org.apache.spark.sql.connector.SimpleDataSourceV2$MyScanBuilder RuntimeFilters: [] +- == Initial Plan == HashAggregate(keys=[], functions=[max(i#0)], output=[max(i)#7]) +- Exchange SinglePartition, ENSURE_REQUIREMENTS, [plan_id=19] +- HashAggregate(keys=[], functions=[partial_max(i#0)], output=[max#14]) +- Project [i#0] +- BatchScan class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5[i#0, j#1] class org.apache.spark.sql.connector.SimpleDataSourceV2$MyScanBuilder RuntimeFilters: [] : +- Subquery subquery#4, [id=apache#33] : +- AdaptiveSparkPlan isFinalPlan=true +- == Final Plan == ResultQueryStage 1 +- *(2) HashAggregate(keys=[], functions=[min(i#10)], output=[min(i)apache#9]) +- ShuffleQueryStage 0 +- Exchange SinglePartition, ENSURE_REQUIREMENTS, [plan_id=63] +- *(1) HashAggregate(keys=[], functions=[partial_min(i#10)], output=[min#15]) +- *(1) Project [i#10] +- BatchScan class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5[i#10, j#11] class org.apache.spark.sql.connector.SimpleDataSourceV2$MyScanBuilder RuntimeFilters: [] +- == Initial Plan == HashAggregate(keys=[], functions=[min(i#10)], output=[min(i)apache#9]) +- Exchange SinglePartition, ENSURE_REQUIREMENTS, [plan_id=30] +- HashAggregate(keys=[], functions=[partial_min(i#10)], output=[min#15]) +- Project [i#10] +- BatchScan class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5[i#10, j#11] class org.apache.spark.sql.connector.SimpleDataSourceV2$MyScanBuilder RuntimeFilters: [] +- Scan OneRowRelation[] ``` After introducing the canonicalization, the plan is as following, where you can see **ReusedSubquery** ``` == Parsed Logical Plan == 'Project [scalar-subquery#2 [] AS max_i#3, scalar-subquery#4 [] AS min_i#5] : :- 'Project [unresolvedalias('max('i))] : : +- 'UnresolvedRelation [df], [], false : +- 'Project [unresolvedalias('min('i))] : +- 'UnresolvedRelation [df], [], false +- OneRowRelation == Analyzed Logical Plan == max_i: int, min_i: int Project [scalar-subquery#2 [] AS max_i#3, scalar-subquery#4 [] AS min_i#5] : :- Aggregate [max(i#0) AS max(i)#7] : : +- SubqueryAlias df : : +- View (`df`, [i#0, j#1]) : : +- RelationV2[i#0, j#1] class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5 : +- Aggregate [min(i#10) AS min(i)apache#9] : +- SubqueryAlias df : +- View (`df`, [i#10, j#11]) : +- RelationV2[i#10, j#11] class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5 +- OneRowRelation == Optimized Logical Plan == Project [scalar-subquery#2 [].max(i) AS max_i#3, scalar-subquery#4 [].min(i) AS min_i#5] : :- Project [named_struct(max(i), max(i)#7, min(i), min(i)apache#9) AS mergedValue#14] : : +- Aggregate [max(i#0) AS max(i)#7, min(i#0) AS min(i)apache#9] : : +- Project [i#0] : : +- RelationV2[i#0, j#1] class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5 : +- Project [named_struct(max(i), max(i)#7, min(i), min(i)apache#9) AS mergedValue#14] : +- Aggregate [max(i#0) AS max(i)#7, min(i#0) AS min(i)apache#9] : +- Project [i#0] : +- RelationV2[i#0, j#1] class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5 +- OneRowRelation == Physical Plan == AdaptiveSparkPlan isFinalPlan=true +- == Final Plan == ResultQueryStage 0 +- *(1) Project [Subquery subquery#2, [id=apache#40].max(i) AS max_i#3, ReusedSubquery Subquery subquery#2, [id=apache#40].min(i) AS min_i#5] : :- Subquery subquery#2, [id=apache#40] : : +- AdaptiveSparkPlan isFinalPlan=true +- == Final Plan == ResultQueryStage 1 +- *(2) Project [named_struct(max(i), max(i)#7, min(i), min(i)apache#9) AS mergedValue#14] +- *(2) HashAggregate(keys=[], functions=[max(i#0), min(i#0)], output=[max(i)#7, min(i)apache#9]) +- ShuffleQueryStage 0 +- Exchange SinglePartition, ENSURE_REQUIREMENTS, [plan_id=71] +- *(1) HashAggregate(keys=[], functions=[partial_max(i#0), partial_min(i#0)], output=[max#16, min#17]) +- *(1) Project [i#0] +- BatchScan class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5[i#0, j#1] class org.apache.spark.sql.connector.SimpleDataSourceV2$MyScanBuilder RuntimeFilters: [] +- == Initial Plan == Project [named_struct(max(i), max(i)#7, min(i), min(i)apache#9) AS mergedValue#14] +- HashAggregate(keys=[], functions=[max(i#0), min(i#0)], output=[max(i)#7, min(i)apache#9]) +- Exchange SinglePartition, ENSURE_REQUIREMENTS, [plan_id=22] +- HashAggregate(keys=[], functions=[partial_max(i#0), partial_min(i#0)], output=[max#16, min#17]) +- Project [i#0] +- BatchScan class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5[i#0, j#1] class org.apache.spark.sql.connector.SimpleDataSourceV2$MyScanBuilder RuntimeFilters: [] : +- ReusedSubquery Subquery subquery#2, [id=apache#40] +- *(1) Scan OneRowRelation[] +- == Initial Plan == Project [Subquery subquery#2, [id=apache#40].max(i) AS max_i#3, Subquery subquery#4, [id=apache#41].min(i) AS min_i#5] : :- Subquery subquery#2, [id=apache#40] : : +- AdaptiveSparkPlan isFinalPlan=true +- == Final Plan == ResultQueryStage 1 +- *(2) Project [named_struct(max(i), max(i)#7, min(i), min(i)apache#9) AS mergedValue#14] +- *(2) HashAggregate(keys=[], functions=[max(i#0), min(i#0)], output=[max(i)#7, min(i)apache#9]) +- ShuffleQueryStage 0 +- Exchange SinglePartition, ENSURE_REQUIREMENTS, [plan_id=71] +- *(1) HashAggregate(keys=[], functions=[partial_max(i#0), partial_min(i#0)], output=[max#16, min#17]) +- *(1) Project [i#0] +- BatchScan class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5[i#0, j#1] class org.apache.spark.sql.connector.SimpleDataSourceV2$MyScanBuilder RuntimeFilters: [] +- == Initial Plan == Project [named_struct(max(i), max(i)#7, min(i), min(i)apache#9) AS mergedValue#14] +- HashAggregate(keys=[], functions=[max(i#0), min(i#0)], output=[max(i)#7, min(i)apache#9]) +- Exchange SinglePartition, ENSURE_REQUIREMENTS, [plan_id=22] +- HashAggregate(keys=[], functions=[partial_max(i#0), partial_min(i#0)], output=[max#16, min#17]) +- Project [i#0] +- BatchScan class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5[i#0, j#1] class org.apache.spark.sql.connector.SimpleDataSourceV2$MyScanBuilder RuntimeFilters: [] : +- Subquery subquery#4, [id=apache#41] : +- AdaptiveSparkPlan isFinalPlan=false : +- Project [named_struct(max(i), max(i)#7, min(i), min(i)apache#9) AS mergedValue#14] : +- HashAggregate(keys=[], functions=[max(i#0), min(i#0)], output=[max(i)#7, min(i)apache#9]) : +- Exchange SinglePartition, ENSURE_REQUIREMENTS, [plan_id=37] : +- HashAggregate(keys=[], functions=[partial_max(i#0), partial_min(i#0)], output=[max#16, min#17]) : +- Project [i#0] : +- BatchScan class org.apache.spark.sql.connector.SimpleDataSourceV2$$anon$5[i#0, j#1] class org.apache.spark.sql.connector.SimpleDataSourceV2$MyScanBuilder RuntimeFilters: [] +- Scan OneRowRelation[] ``` ### Was this patch authored or co-authored using generative AI tooling? No Closes apache#52529 from yhuang-db/scan-canonicalization. Authored-by: yhuang-db <itisyuchuan@gmail.com> Signed-off-by: Peter Toth <peter.toth@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Here shutdown thread does the following three steps one by one before triggering the exit for the executor:
DecommissionExecutorfor the driverDisclaimer: this change main purpose to help in our discussion and it is just slightly tested,
BlockManagerDecommissionIntegrationSuiteruns successfully: