Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[SPARK-31187][SQL] Sort the whole-stage codegen debug output by codegenStageId #27955

Closed
wants to merge 1 commit into from

Conversation

rednaxelafx
Copy link
Contributor

What changes were proposed in this pull request?

Spark SQL's whole-stage codegen (WSCG) supports dumping the generated code to help with debugging. One way to get the generated code is through df.queryExecution.debug.codegen, or SQL EXPLAIN CODEGEN statement.

The generated code is currently printed without specific ordering, which can make debugging a bit annoying. This PR makes a minor improvement to sort the codegen dump by the codegenStageId, ascending.

After this change, the following query:

spark.range(10).agg(sum('id)).queryExecution.debug.codegen

will always dump the generated code in a natural, stable order. A version of this example with shorter output is:

spark.range(10).agg(sum('id)).queryExecution.debug.codegenToSeq.map(_._1).foreach(println)
*(1) HashAggregate(keys=[], functions=[partial_sum(id#8L)], output=[sum#15L])
+- *(1) Range (0, 10, step=1, splits=16)

*(2) HashAggregate(keys=[], functions=[sum(id#8L)], output=[sum(id)#12L])
+- Exchange SinglePartition, true, [id=#30]
   +- *(1) HashAggregate(keys=[], functions=[partial_sum(id#8L)], output=[sum#15L])
      +- *(1) Range (0, 10, step=1, splits=16)

The number of codegen stages within a single SQL query tends to be very small, most likely < 50, so the overhead of adding the sorting shouldn't be significant.

Why are the changes needed?

Minor improvement to aid WSCG debugging.

Does this PR introduce any user-facing change?

No user-facing change for end-users; minor change for developers who debug WSCG generated code.

How was this patch tested?

Manually tested the output; all other tests still pass.

Copy link
Member

@gengliangwang gengliangwang left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@SparkQA
Copy link

SparkQA commented Mar 19, 2020

Test build #120026 has finished for PR 27955 at commit c64b778.

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

@kiszk
Copy link
Member

kiszk commented Mar 19, 2020

retest this please

@SparkQA
Copy link

SparkQA commented Mar 19, 2020

Test build #120029 has finished for PR 27955 at commit c64b778.

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

@maropu maropu closed this in a177628 Mar 19, 2020
maropu pushed a commit that referenced this pull request Mar 19, 2020
…enStageId

### What changes were proposed in this pull request?

Spark SQL's whole-stage codegen (WSCG) supports dumping the generated code to help with debugging. One way to get the generated code is through `df.queryExecution.debug.codegen`, or SQL `EXPLAIN CODEGEN` statement.

The generated code is currently printed without specific ordering, which can make debugging a bit annoying. This PR makes a minor improvement to sort the codegen dump by the `codegenStageId`, ascending.

After this change, the following query:
```scala
spark.range(10).agg(sum('id)).queryExecution.debug.codegen
```
will always dump the generated code in a natural, stable order. A version of this example with shorter output is:
```
spark.range(10).agg(sum('id)).queryExecution.debug.codegenToSeq.map(_._1).foreach(println)
*(1) HashAggregate(keys=[], functions=[partial_sum(id#8L)], output=[sum#15L])
+- *(1) Range (0, 10, step=1, splits=16)

*(2) HashAggregate(keys=[], functions=[sum(id#8L)], output=[sum(id)#12L])
+- Exchange SinglePartition, true, [id=#30]
   +- *(1) HashAggregate(keys=[], functions=[partial_sum(id#8L)], output=[sum#15L])
      +- *(1) Range (0, 10, step=1, splits=16)
```

The number of codegen stages within a single SQL query tends to be very small, most likely < 50, so the overhead of adding the sorting shouldn't be significant.

### Why are the changes needed?

Minor improvement to aid WSCG debugging.

### Does this PR introduce any user-facing change?

No user-facing change for end-users; minor change for developers who debug WSCG generated code.

### How was this patch tested?

Manually tested the output; all other tests still pass.

Closes #27955 from rednaxelafx/codegen.

Authored-by: Kris Mok <kris.mok@databricks.com>
Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
(cherry picked from commit a177628)
Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
@maropu
Copy link
Member

maropu commented Mar 19, 2020

Thanks! Merged to master/3.0.

@maropu
Copy link
Member

maropu commented Mar 19, 2020

Since I think this PR improves debugability, I merged this into 3.0, too.

sjincho pushed a commit to sjincho/spark that referenced this pull request Apr 15, 2020
…enStageId

### What changes were proposed in this pull request?

Spark SQL's whole-stage codegen (WSCG) supports dumping the generated code to help with debugging. One way to get the generated code is through `df.queryExecution.debug.codegen`, or SQL `EXPLAIN CODEGEN` statement.

The generated code is currently printed without specific ordering, which can make debugging a bit annoying. This PR makes a minor improvement to sort the codegen dump by the `codegenStageId`, ascending.

After this change, the following query:
```scala
spark.range(10).agg(sum('id)).queryExecution.debug.codegen
```
will always dump the generated code in a natural, stable order. A version of this example with shorter output is:
```
spark.range(10).agg(sum('id)).queryExecution.debug.codegenToSeq.map(_._1).foreach(println)
*(1) HashAggregate(keys=[], functions=[partial_sum(id#8L)], output=[sum#15L])
+- *(1) Range (0, 10, step=1, splits=16)

*(2) HashAggregate(keys=[], functions=[sum(id#8L)], output=[sum(id)#12L])
+- Exchange SinglePartition, true, [id=apache#30]
   +- *(1) HashAggregate(keys=[], functions=[partial_sum(id#8L)], output=[sum#15L])
      +- *(1) Range (0, 10, step=1, splits=16)
```

The number of codegen stages within a single SQL query tends to be very small, most likely < 50, so the overhead of adding the sorting shouldn't be significant.

### Why are the changes needed?

Minor improvement to aid WSCG debugging.

### Does this PR introduce any user-facing change?

No user-facing change for end-users; minor change for developers who debug WSCG generated code.

### How was this patch tested?

Manually tested the output; all other tests still pass.

Closes apache#27955 from rednaxelafx/codegen.

Authored-by: Kris Mok <kris.mok@databricks.com>
Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

7 participants