[native] Add expression optimizer #22927

pramodsatya · 2024-06-05T15:59:54Z

Description

To support constant folding and consistent semantics between the Presto coordinator (Java) and the Presto C++ worker, it is necessary to use consistent expression evaluation. To support this, a native expression evaluation endpoint, v1/expressions, has been added to the Presto C++ sidecar, and a plugin has been created which can utilize Velox expression evaluation behind a standard ExpressionOptimizer.

Depends on Velox changes from facebookincubator/velox#13424 to add an ExpressionOptimizer for optimizing and constant folding TypedExprs. The optimized Velox core::TypedExpr is converted to a Presto protocol::RowExpression in the Presto native sidecar with helper class VeloxToPrestoExprConverter. The end to end flow between the coordinator and sidecar looks like:

RowExpression == (PrestoToVeloxExpr) ==> TypedExpr == (Velox API optimizeExpression()) ==> optimized `TypedExpr` == (VeloxToPrestoExprConverter) ==> optimized RowExpression

Please refer to this document for sidecar implementation details.

Motivation and Context

Consistency between C++ and Java semantics. Support for using C++ functions during constant folding of expressions in the planner. Please refer to RFC-0006.

Test Plan

Tests have been added by extending the TestRowExpressionInterpreter class to also test native expression evaluation in TestNativeExpressionOptimizer.java. However, this feature is still in Beta, and to support production workloads with complete certainty a fuzzer will be created to surface any remaining bugs with the integration at a later time.

Unit tests for simple expression conversions are added in VeloxToPrestoExprConverter.cpp.

Release Notes

== NO RELEASE NOTE ==

aditi-pandit

@pramodsatya : Have done a first round of comments. Will read your tests more closely once you address the comments here.

presto-native-execution/presto_cpp/main/PrestoServer.cpp

presto-native-execution/presto_cpp/main/expression/RowExpressionEvaluator.h

presto-native-execution/presto_cpp/main/expression/RowExpressionEvaluator.cpp

presto-native-execution/presto_cpp/main/expression/RowExpressionEvaluator.h

presto-native-execution/presto_cpp/main/types/PrestoToVeloxExpr.h

presto-native-execution/presto_cpp/main/expression/RowExpressionEvaluator.cpp

presto-native-execution/presto_cpp/main/expression/tests/RowExpressionEvaluatorTest.cpp

pramodsatya

Thanks for the feedback @aditi-pandit, addressed the comments. Could you please take another look?

presto-native-execution/presto_cpp/main/expression/RowExpressionEvaluator.cpp

presto-native-execution/presto_cpp/main/expression/tests/RowExpressionEvaluatorTest.cpp

Summary: prestodb/presto#23331 adds a native expression optimizer to delegate expression evaluation to the native sidecar. This is used to constant fold expressions on the presto native sidecar, instead of on the presto java coordinator (which is the current behavior). prestodb/presto#22927 implements a proxygen endpoint to accept `RowExpression`s from `NativeSidecarExpressionInterpreter`, optimize them if possible (rewrite special form expressions), and compile the `RowExpression` to a velox expression with constant folding enabled. This velox expression is then converted back to a `RowExpression` and returned by the sidecar to the coordinator. When the constant folded velox expression is of type `velox::exec::ConstantExpr`, we need to return a `RowExpression` of type `ConstantExpression`. This requires us to serialize the constant value from `velox::exec::ConstantExpr` into `protocol::ConstantExpression::valueBlock`. This can be done by serializing the constant value vector to presto SerializedPage::column format, followed by base 64 encoding the result (reverse engineering the logic from `Base64Util.cpp::readBlock`). This PR adds a new function, `serializeSingleColumn`, to `PrestoVectorSerde`. This can be used to serialize input data from vectors containing a single element into a single PrestoPage column format (without the PrestoPage header). This function is not added to `PrestoBatchVectorSerializer` alongside the existing `serialize` function since that would require adding it as a virtual function in `BatchVectorSerializer` as well, and this is not desirable since the `PrestoPage` format is not relevant in this base class. There is an existing function `deserializeSingleColumn` in `PrestoVectorSerde` which is used to deserialize data from a single column, since `serializeSingleColumn` performs the inverse operation to this function, it is added alongside it in `PrestoVectorSerde`. Pull Request resolved: #10657 Reviewed By: amitkdutta Differential Revision: D66044754 Pulled By: pedroerp fbshipit-source-id: e509605067920f8207e5a3fa67552badc2ce0eba

presto-native-execution/presto_cpp/main/expression/RowExpressionOptimizer.h

presto-native-execution/presto_cpp/main/expression/RowExpressionOptimizer.cpp

pramodsatya

Thanks for the feedback @aditi-pandit, addressed the review comments. Could you please take another look?

presto-native-execution/presto_cpp/main/expression/RowExpressionOptimizer.cpp

pramodsatya · 2025-04-22T22:26:39Z

@aditi-pandit, @czentgr, could you please help review this PR? Added an API to constant fold TypedExpr, as discussed in the native worker group meeting (documented here).

steveburnett

Thanks for the doc! Some minor nits of formatting.

presto-docs/src/main/sphinx/presto_cpp/sidecar.rst

steveburnett

LGTM! (docs)

Pull updated branch, new local doc build, looks good. Thanks!

aditi-pandit

@pramodsatya : Haven't read all the code in detail but have one high level comment about the API in VeloxToPrestoExpr

aditi-pandit · 2025-06-20T22:21:16Z

presto-native-execution/presto_cpp/main/types/VeloxToPrestoExpr.h

+ public:
+  explicit VeloxToPrestoExprConverter(memory::MemoryPool* pool) : pool_(pool) {}
+
+  /// Converts a velox expression `expr` to a Presto protocol RowExpression.


Didn't follow this... What is the output json ?

aditi-pandit · 2025-06-20T22:25:04Z

presto-native-execution/presto_cpp/main/types/VeloxToPrestoExpr.cpp

+    return getCallExpression(call, input);
+  }
+
+  LOG(ERROR) << "Unable to convert Velox expression: {}" << expr->toString();


This function signature could be changed to return an optional json. If there is an error here, you can return nullopt. The input expression wouldn't be needed then.

The caller of this function can then decide to return the input expression if the conversion was not successful.

aditi-pandit · 2025-06-20T22:36:29Z

presto-native-execution/presto_cpp/main/types/VeloxToPrestoExpr.cpp

+      return getSpecialFormExpression(callTypedExpr, input);
+    }
+    return getCallExpression(callTypedExpr, input);
+  } else if (


Maybe move Cast logic before Call.

Co-authored-by: Pratik Joseph Dabre <[email protected]>

Co-authored-by: Pramod Satya <[email protected]> Co-authored-by: Pratik Joseph Dabre <[email protected]>

Co-authored-by: Pramod Satya <[email protected]>

pramodsatya force-pushed the expr_endpt branch from e8f3c68 to 418b219 Compare July 25, 2024 21:49

pramodsatya changed the title ~~[WIP] Add proxygen endpoint for expression evaluation~~ [native] Add proxygen endpoint for expression evaluation Aug 5, 2024

pramodsatya mentioned this pull request Aug 5, 2024

Add function serializeSingleColumn to PrestoVectorSerde facebookincubator/velox#10657

Closed

pramodsatya force-pushed the expr_endpt branch from 418b219 to 50d6456 Compare August 5, 2024 08:32

pramodsatya requested review from aditi-pandit, czentgr and tdcmeehan August 5, 2024 08:33

tdcmeehan self-assigned this Aug 5, 2024

pramodsatya force-pushed the expr_endpt branch 2 times, most recently from 822f79f to 2352cd4 Compare September 11, 2024 14:52

aditi-pandit reviewed Sep 12, 2024

View reviewed changes

pramodsatya force-pushed the expr_endpt branch from 2352cd4 to 496bbc5 Compare September 17, 2024 02:41

pramodsatya commented Sep 17, 2024

View reviewed changes

pramodsatya force-pushed the expr_endpt branch from 496bbc5 to d9c3dcd Compare September 17, 2024 03:09

pramodsatya force-pushed the expr_endpt branch from d9c3dcd to e19ae51 Compare October 7, 2024 15:05

pramodsatya force-pushed the expr_endpt branch from e19ae51 to 747ef08 Compare November 4, 2024 13:40

pramodsatya force-pushed the expr_endpt branch from 747ef08 to 10bfc08 Compare December 2, 2024 02:40

pramodsatya requested a review from aditi-pandit December 2, 2024 02:40

aditi-pandit reviewed Dec 2, 2024

View reviewed changes

pramodsatya changed the title ~~[native] Add proxygen endpoint for expression evaluation~~ [native] Add row expression optimizer Dec 10, 2024

pramodsatya force-pushed the expr_endpt branch from 10bfc08 to 37ef905 Compare December 10, 2024 08:05

pramodsatya commented Dec 10, 2024

View reviewed changes

presto-native-execution/presto_cpp/main/expression/RowExpressionOptimizer.cpp Outdated Show resolved Hide resolved

presto-native-execution/presto_cpp/main/expression/RowExpressionOptimizer.cpp Outdated Show resolved Hide resolved

pramodsatya force-pushed the expr_endpt branch from 37ef905 to 7425558 Compare December 10, 2024 17:17

pramodsatya marked this pull request as ready for review December 10, 2024 17:49

pramodsatya requested a review from a team as a code owner December 10, 2024 17:49

prestodb-ci added the from:IBM PR from IBM label Dec 10, 2024

prestodb-ci requested review from a team and imjalpreet and removed request for a team December 10, 2024 17:49

pramodsatya force-pushed the expr_endpt branch from 66b5c37 to 57bf02d Compare April 14, 2025 06:03

pramodsatya requested review from a team, elharo, feilong-liu, jaystarshot, shrinidhijoshi and vivek-bharathan as code owners April 14, 2025 06:03

pramodsatya requested a review from presto-oss April 14, 2025 06:03

pramodsatya changed the title ~~[native] Add row expression optimizer~~ [native] Add expression optimizer Apr 14, 2025

pramodsatya force-pushed the expr_endpt branch from 57bf02d to be72c31 Compare April 22, 2025 22:24

pramodsatya mentioned this pull request Apr 25, 2025

Add expression evaluation in Presto sidecar #24126

Closed

pramodsatya force-pushed the expr_endpt branch from be72c31 to 14ac76d Compare April 25, 2025 02:32

pramodsatya requested a review from steveburnett as a code owner April 25, 2025 02:32

steveburnett requested changes Apr 25, 2025

View reviewed changes

presto-docs/src/main/sphinx/presto_cpp/sidecar.rst Outdated Show resolved Hide resolved

presto-docs/src/main/sphinx/presto_cpp/sidecar.rst Outdated Show resolved Hide resolved

presto-docs/src/main/sphinx/presto_cpp/sidecar.rst Outdated Show resolved Hide resolved

pramodsatya force-pushed the expr_endpt branch from 14ac76d to 64e9789 Compare April 26, 2025 07:09

steveburnett previously approved these changes Apr 28, 2025

View reviewed changes

pramodsatya dismissed steveburnett’s stale review via 88f17cd June 4, 2025 00:16

pramodsatya force-pushed the expr_endpt branch from 64e9789 to 88f17cd Compare June 4, 2025 00:16

pramodsatya force-pushed the expr_endpt branch from 88f17cd to 7581285 Compare June 20, 2025 20:51

aditi-pandit reviewed Jun 20, 2025

View reviewed changes

tdcmeehan and others added 6 commits August 18, 2025 15:41

Add RowExpressionSerde and JsonCodecRowExpressionSerde

abc41a0

Co-authored-by: Pratik Joseph Dabre <[email protected]>

Add native expression optimizer

7e98797

Co-authored-by: Pratik Joseph Dabre <[email protected]>

[native] Add Velox to Presto expression converter

f220117

[native] Add expression optimization endpoint in sidecar

60d890c

[native] Add e2e tests for native row expression optimizer

1b87d9d

Co-authored-by: Pramod Satya <[email protected]> Co-authored-by: Pratik Joseph Dabre <[email protected]>

Add documentation for /v1/expressions

4f203b2

Co-authored-by: Pramod Satya <[email protected]>

pramodsatya force-pushed the expr_endpt branch from 7581285 to 4f203b2 Compare August 19, 2025 16:36

pramodsatya mentioned this pull request Nov 19, 2025

fix: Unregister special forms SWITCH rewrite facebookincubator/velox#15549

Closed

pramodsatya closed this Jan 22, 2026

[native] Add expression optimizer #22927

[native] Add expression optimizer #22927

Uh oh!

Conversation

pramodsatya commented Jun 5, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Motivation and Context

Test Plan

Release Notes

Uh oh!

aditi-pandit left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

pramodsatya left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

pramodsatya left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

pramodsatya commented Apr 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

steveburnett left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

steveburnett left a comment

Choose a reason for hiding this comment

Uh oh!

aditi-pandit left a comment

Choose a reason for hiding this comment

Uh oh!

aditi-pandit Jun 20, 2025

Choose a reason for hiding this comment

Uh oh!

aditi-pandit Jun 20, 2025

Choose a reason for hiding this comment

Uh oh!

aditi-pandit Jun 20, 2025

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

pramodsatya commented Jun 5, 2024 •

edited

Loading

pramodsatya commented Apr 22, 2025 •

edited

Loading