Convert `nth_value` builtIn function to User Defined Window Function #13201

buraksenn · 2024-10-31T13:26:06Z

Which issue does this PR close?

Closes #12649

Rationale for this change

Context: #8709

What changes are included in this PR?

cleanup BuiltinWindowFunctions
move builtin nth_value function to udwf

Are these changes tested?

yes

Are there any user-facing changes?

no

alamb · 2024-10-31T19:21:20Z

THis is so exciting. FYI @jonathanc-n and @Omega359

alamb · 2024-10-31T19:24:18Z

It got pretty out of hand because of nth_value and BuiltinWindowFunction references across the repo so I wanted to open a draft PR to get feedback.

I personally think it would be fine to leave BuiltInWindowFunction initially while we port over the function, and them remove it as a follow on PR

Perhaps you can leave a stub in like

enum BuiltInWindowFunction {
  // Never created, will be removed in a follow on PR
  Stub
};

Then we can focus this PR on making sure that nth_value is working as a user defined function and then move on in the next PR to mechanically ripping out the old BuiltInWindowFunction

buraksenn · 2024-11-01T11:08:29Z

It got pretty out of hand because of nth_value and BuiltinWindowFunction references across the repo so I wanted to open a draft PR to get feedback.

I personally think it would be fine to leave BuiltInWindowFunction initially while we port over the function, and them remove it as a follow on PR

Perhaps you can leave a stub in like
enum BuiltInWindowFunction {
  // Never created, will be removed in a follow on PR
  Stub
};
Then we can focus this PR on making sure that nth_value is working as a user defined function and then move on in the next PR to mechanically ripping out the old BuiltInWindowFunction

Thanks @alamb will continue with what you've said

buraksenn · 2024-11-03T20:30:20Z

Wanted to update here. I think I'm almost finished but probably encountered a side effect. This query fails in slt file:

External error: query failed: DataFusion error: Arrow error: Invalid argument error: It is not possible to concatenate arrays of different data types.
[SQL] select
c9,
sum(c5) over (order by c9) as sum1,
avg(c5) over (order by c9) as avg1,
count(c5) over (order by c9) as count1,
max(c5) over (order by c9) as max1,
min(c5) over (order by c9) as min1,
first_value(c5) over (order by c9) as fv1,
last_value(c5) over (order by c9) as lv1,
nth_value(c5, 2) over (order by c9) as nv1
from aggregate_test_100
order by c9
limit 5
at test_files/window.slt:97

Error: Execution("1 failures")
error: test failed, to rerun pass `-p datafusion-sqllogictest --test sqllogictests`

I hope to fix this and make this ready tomorrow

jcsherin · 2024-11-04T06:51:31Z

External error: query failed: DataFusion error: Arrow error: Invalid argument error: It is not possible to concatenate arrays of different data types.

In the built-in (older) version the output field is defined like:

    fn field(&self) -> Result<Field> {
        let nullable = true;
        Ok(Field::new(&self.name, self.data_type.clone(), nullable))
    }

In the current code, the data type of the field is hard-coded as DataType::UInt64:

    fn field(&self, field_args: WindowUDFFieldArgs) -> Result<Field> {
        let nullable = true;

        Ok(Field::new(field_args.name(), DataType::UInt64, nullable))
    }

To fix this use field_args to get the data type of the input expression rather than hard-coding DataType::UInt64.

buraksenn · 2024-11-04T09:33:04Z

External error: query failed: DataFusion error: Arrow error: Invalid argument error: It is not possible to concatenate arrays of different data types.

In the built-in (older) version the output field is defined like:
    fn field(&self) -> Result<Field> {
        let nullable = true;
        Ok(Field::new(&self.name, self.data_type.clone(), nullable))
    }
In the current code, the data type of the field is hard-coded as DataType::UInt64:
    fn field(&self, field_args: WindowUDFFieldArgs) -> Result<Field> {
        let nullable = true;

        Ok(Field::new(field_args.name(), DataType::UInt64, nullable))
    }
To fix this use field_args to get the data type of the input expression rather than hard-coding DataType::UInt64.

Thanks, @jcsherin that was the fix. I've fixed that issue but encountered another one. I return Error from partition evaluator but I think it is not honored.

External error: query is expected to fail, but actually succeed:
[SQL] SELECT NTH_VALUE('+Inf'::Double, v1) OVER (PARTITION BY v1) FROM t1;
at test_files/window.slt:4895

Error: Execution("1 failures")
error: test failed, to rerun pass `-p datafusion-sqllogictest --test sqllogictests`

Caused by:
  process didn't exit successfully: `/Users/buraksen/d/datafusion/target/debug/deps/sqllogictests-267639e9b963ccc2 window` (exit status: 1)

But it should not succeed since:

        let n =
            match get_scalar_value_from_args(partition_evaluator_args.input_exprs(), 1)?
                .map(get_signed_integer)
            {
                Some(Ok(n)) => {
                    if partition_evaluator_args.is_reversed() {
                        -n
                    } else {
                        n
                    }
                }
                _ => {
                    return exec_err!(
                "Expected a signed integer literal for the second argument of nth_value"
            )
                }
            };

jcsherin · 2024-11-04T18:29:05Z

TL;DR

For invalid input expressions, built-in window functions fail early when converting logical plan to physical plan. But user-defined window functions will complete planning, and fail only during physical execution.

Validation of input expressions in user-defined window runs only during physical execution.

In this case is it not better for udwf to fail early when converting to physical plan?

A possible solution is to update WindowUDFImpl API so that we parse input expressions when creating the window expressions (similar to built-in window functions). And hook it here so argument parsing happens earlier:

datafusion/datafusion/physical-plan/src/windows/mod.rs

Lines 158 to 164 in b61b2fc

    
           fn create_udwf_window_expr( 
        
               fun: &Arc<WindowUDF>, 
        
               args: &[Arc<dyn PhysicalExpr>], 
        
               input_schema: &Schema, 
        
               name: String, 
        
               ignore_nulls: bool, 
        
           ) -> Result<Arc<dyn BuiltInWindowFunctionExpr>> {

Edge Case: Empty Table

DataFusion CLI v42.2.0
> CREATE TABLE t1(v1 BIGINT);
0 row(s) fetched.
Elapsed 0.020 seconds.

There are currently no rows in t1 so an early return happens in WindowAggStream::compute_aggregates.

datafusion/datafusion/physical-plan/src/windows/window_agg_exec.rs

Lines 319 to 321 in 89e96b4

    
           if batch.num_rows() == 0 { 
        
               return Ok(RecordBatch::new_empty(Arc::clone(&self.schema))); 
        
           }

The nth_value window function is never executed when this query is run on the empty table t1.

> SELECT NTH_VALUE('+Inf'::Double, v1) OVER (PARTITION BY v1) FROM t1;
+-------------------------------------------------------------------------------------------------------------+
| nth_value(Utf8("+Inf"),t1.v1) PARTITION BY [t1.v1] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING |
+-------------------------------------------------------------------------------------------------------------+
+-------------------------------------------------------------------------------------------------------------+
0 row(s) fetched.
Elapsed 0.018 seconds.

After we insert a few values into t1 the nth_value aggregation will be computed. Only then do we see the expected error message,

> insert into t1 values (123), (456);
+-------+
| count |
+-------+
| 2     |
+-------+
1 row(s) fetched.
Elapsed 0.007 seconds.

> SELECT NTH_VALUE('+Inf'::Double, v1) OVER (PARTITION BY v1) FROM t1;
This feature is not implemented: There is only support Literal types for field at idx: 1 in Window Function

Planning divergence between built-in & user-defined window functions

In main branch where nth_value is a BuiltInWindowFunction the argument parsing fails early when mapping logical to physical plan. In explain output there is no physical plan.

> EXPLAIN SELECT NTH_VALUE('+Inf'::Double, v1) OVER (PARTITION BY v1) FROM t1;
+--------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| plan_type    | plan                                                                                                                                                                                                                                                     |
+--------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| logical_plan | Projection: NTH_VALUE(Utf8("+Inf"),t1.v1) PARTITION BY [t1.v1] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING                                                                                                                                  |
|              |   WindowAggr: windowExpr=[[NTH_VALUE(Float64(inf), t1.v1) PARTITION BY [t1.v1] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING AS NTH_VALUE(Utf8("+Inf"),t1.v1) PARTITION BY [t1.v1] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING]] |
|              |     TableScan: t1 projection=[v1]                                                                                                                                                                                                                        |
+--------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1 row(s) fetched.
Elapsed 0.009 seconds.

But this is not the case for user-defined window functions. In this branch we instead see that a complete plan is built and failure is happening only when the query executes,

> EXPLAIN SELECT NTH_VALUE('+Inf'::Double, v1) OVER (PARTITION BY v1) FROM t1;
+---------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| plan_type     | plan                                                                                                                                                                                                                                                                                                                                                                                                                                                                                          |
+---------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| logical_plan  | Projection: nth_value(Utf8("+Inf"),t1.v1) PARTITION BY [t1.v1] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING                                                                                                                                                                                                                                                                                                                                                                       |
|               |   WindowAggr: windowExpr=[[nth_value(Float64(inf), t1.v1) PARTITION BY [t1.v1] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING AS nth_value(Utf8("+Inf"),t1.v1) PARTITION BY [t1.v1] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING]]                                                                                                                                                                                                                                      |
|               |     TableScan: t1 projection=[v1]                                                                                                                                                                                                                                                                                                                                                                                                                                                             |
| physical_plan | ProjectionExec: expr=[nth_value(Utf8("+Inf"),t1.v1) PARTITION BY [t1.v1] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING@1 as nth_value(Utf8("+Inf"),t1.v1) PARTITION BY [t1.v1] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING]                                                                                                                                                                                                                                           |
|               |   WindowAggExec: wdw=[nth_value(Utf8("+Inf"),t1.v1) PARTITION BY [t1.v1] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING: Ok(Field { name: "nth_value(Utf8(\"+Inf\"),t1.v1) PARTITION BY [t1.v1] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING", data_type: Float64, nullable: true, dict_id: 0, dict_is_ordered: false, metadata: {} }), frame: WindowFrame { units: Rows, start_bound: Preceding(UInt64(NULL)), end_bound: Following(UInt64(NULL)), is_causal: false }] |
|               |     SortExec: expr=[v1@0 ASC NULLS LAST], preserve_partitioning=[false]                                                                                                                                                                                                                                                                                                                                                                                                                       |
|               |       MemoryExec: partitions=1, partition_sizes=[0]                                                                                                                                                                                                                                                                                                                                                                                                                                           |
|               |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                               |
+---------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2 row(s) fetched.
Elapsed 0.019 seconds.

buraksenn · 2024-11-04T20:17:34Z

@jcsherin thanks for the very detailed explanation. In this case, I think it would be better to update WindowUDFImpl in a followup PR for enhancement right? I can skip this test case in the scope of this PR. Correct me if I'm wrong please

jcsherin · 2024-11-05T03:28:42Z

I think it would be better to update WindowUDFImpl in a followup PR for enhancement right? I can skip this test case in the scope of this PR. Correct me if I'm wrong please

Sure, we can improve the API in another PR.

Here is a workaround that fixes the failing test:

// In datafusion/physical-plan/src/windows/mod.rs
fn create_udwf_window_expr(
    fun: &Arc<WindowUDF>,
    args: &[Arc<dyn PhysicalExpr>],
    input_schema: &Schema,
    name: String,
    ignore_nulls: bool,
) -> Result<Arc<dyn BuiltInWindowFunctionExpr>> {
    // need to get the types into an owned vec for some reason
    let input_types: Vec<_> = args
        .iter()
        .map(|arg| arg.data_type(input_schema))
        .collect::<Result<_>>()?;

    let udwf_expr =
        Arc::new(WindowUDFExpr {
            fun: Arc::clone(fun),
            args: args.to_vec(),
            input_types,
            name,
            is_reversed: false,
            ignore_nulls,
        });

    /// Early validation of input expressions
    ///
    /// We create a partition evaluator because in the user-defined window
    /// implementation this is where code for parsing input expressions
    /// exist. The benefits are:
    /// - If any of the input expressions are invalid we catch them early
    /// in the planning phase, rather than during execution.
    /// - Maintains compatibility with built-in (now removed) window
    /// functions validation behavior.
    /// - Predictable and reliable error handling.
    ///
    /// See discussion here:
    /// https://github.com/apache/datafusion/pull/13201#issuecomment-2454209975
    let _ = udwf_expr.create_evaluator()?;

    Ok(udwf_expr)
}

I verified that this works in your branch.

DataFusion CLI v42.2.0
> CREATE TABLE t1(v1 BIGINT);
0 row(s) fetched.
Elapsed 0.019 seconds.

> EXPLAIN SELECT NTH_VALUE('+Inf'::Double, v1) OVER (PARTITION BY v1) FROM t1;
+--------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| plan_type    | plan                                                                                                                                                                                                                                                     |
+--------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| logical_plan | Projection: nth_value(Utf8("+Inf"),t1.v1) PARTITION BY [t1.v1] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING                                                                                                                                  |
|              |   WindowAggr: windowExpr=[[nth_value(Float64(inf), t1.v1) PARTITION BY [t1.v1] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING AS nth_value(Utf8("+Inf"),t1.v1) PARTITION BY [t1.v1] ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING]] |
|              |     TableScan: t1 projection=[v1]                                                                                                                                                                                                                        |
+--------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1 row(s) fetched.
Elapsed 0.018 seconds.

> SELECT NTH_VALUE('+Inf'::Double, v1) OVER (PARTITION BY v1) FROM t1;
This feature is not implemented: There is only support Literal types for field at idx: 1 in Window Function

This workaround may not be ideal, but at least we do not have to skip this test. Also please feel free to update the code/comments as you see fit.

datafusion/proto/tests/cases/roundtrip_physical_plan.rs

datafusion/sqllogictest/test_files/errors.slt

datafusion/proto/src/physical_plan/to_proto.rs

jcsherin · 2024-11-11T12:23:50Z

datafusion/functions-window/src/nth_value.rs

+
+/// Create an expression to represent the `nth_value` window function
+///
+pub fn nth_value(arg: datafusion_expr::Expr, n: Option<i64>) -> datafusion_expr::Expr {


The type of n is i64, not Option<i64>.

See the rust docs: https://docs.rs/datafusion/latest/datafusion/logical_expr/window_function/fn.nth_value.html

Also add a roundtrip logical plan test for this API here:

datafusion/datafusion/proto/tests/cases/roundtrip_logical_plan.rs

Line 943 in ac79ef3

cume_dist(),

Fixed and added test

datafusion/proto/tests/cases/roundtrip_logical_plan.rs

jcsherin · 2024-11-11T13:55:55Z

@buraksenn Tremendous effort 🙌. These changes look good to me.

datafusion/sqllogictest/test_files/errors.slt

Co-authored-by: Sherin Jacob <[email protected]>

jcsherin · 2024-11-11T15:23:07Z

@buraksenn and @berkaysynnada Thanks!

@alamb This PR is ready.

alamb · 2024-11-13T06:56:14Z

Awesome -- thank you so much. I will review this PR hopefully later today

alamb

Thank you so much @buraksenn , @jcsherin -- it is just so beautiful to see this PR now after all the work. It is basically perfect from my perspective 🏆

alamb · 2024-11-13T15:31:36Z

datafusion/physical-plan/Cargo.toml

@@ -70,6 +70,7 @@ tokio = { workspace = true }
 [dev-dependencies]
 criterion = { version = "0.5", features = ["async_futures"] }
 datafusion-functions-aggregate = { workspace = true }
+datafusion-functions-window = { workspace = true }


some day I hope we can remove these dependencies (so we can make testing physical-plan faster, but not a part of this PR

alamb · 2024-11-13T15:32:32Z

datafusion/physical-plan/src/windows/mod.rs

+    // We create a partition evaluator because in the user-defined window
+    // implementation this is where code for parsing input expressions
+    // exist. The benefits are:
+    // - If any of the input expressions are invalid we catch them early


💯 for these comments that explain the rationale

alamb · 2024-11-13T15:39:16Z

I also took the liberty of merging up from main to make sure we haven't hit any logical conflicts with this PR

alamb · 2024-11-13T16:10:32Z

I don't think there is any reason to wait around for this PR -- people know it is coming, so let's get this in 🚀

refactored nth_value

014306d

github-actions bot added sql SQL Planner logical-expr Logical plan and expressions physical-expr Physical Expressions core Core DataFusion crate sqllogictest SQL Logic Tests (.slt) substrait proto Related to proto crate labels Oct 31, 2024

buraksenn added 2 commits November 3, 2024 17:11

merge main

b61b2fc

continue

fda6a6f

buraksenn force-pushed the nth_value_func_ref branch from 8cf82f0 to fda6a6f Compare November 3, 2024 20:18

github-actions bot removed the sqllogictest SQL Logic Tests (.slt) label Nov 3, 2024

test

7fb84fa

github-actions bot added the sqllogictest SQL Logic Tests (.slt) label Nov 3, 2024

proto and rustlint

e6ef6ed

buraksenn added 3 commits November 4, 2024 12:14

fix datatype

1131df0

cont

f28163a

cont

fddbc58

buraksenn force-pushed the nth_value_func_ref branch from 0706334 to fddbc58 Compare November 4, 2024 09:32

buraksenn added 2 commits November 5, 2024 08:50

apply jcsherins early validation

c406505

docs

12c0a52

buraksenn force-pushed the nth_value_func_ref branch from 42bb690 to aa18c1d Compare November 10, 2024 21:21

buraksenn commented Nov 10, 2024

View reviewed changes

datafusion/proto/tests/cases/roundtrip_physical_plan.rs Outdated Show resolved Hide resolved

berkaysynnada added 2 commits November 11, 2024 10:29

Update roundtrip_physical_plan.rs

be02f03

udwf, not udaf

5c4e0f6

jcsherin reviewed Nov 11, 2024

View reviewed changes

datafusion/sqllogictest/test_files/errors.slt Show resolved Hide resolved

buraksenn added 2 commits November 11, 2024 12:05

Merge branch 'main' into nth_value_func_ref

f17253b

fix bounded but not fixed roundtrip

9bbeef4

buraksenn commented Nov 11, 2024

View reviewed changes

datafusion/proto/src/physical_plan/to_proto.rs Show resolved Hide resolved

jcsherin reviewed Nov 11, 2024

View reviewed changes

added

9e7e5e2

jcsherin reviewed Nov 11, 2024

View reviewed changes

datafusion/proto/tests/cases/roundtrip_logical_plan.rs Show resolved Hide resolved

Merge branch 'main' into nth_value_func_ref

7a8a760

jcsherin reviewed Nov 11, 2024

View reviewed changes

datafusion/sqllogictest/test_files/errors.slt Outdated Show resolved Hide resolved

Update datafusion/sqllogictest/test_files/errors.slt

073d8e8

Co-authored-by: Sherin Jacob <[email protected]>

alamb mentioned this pull request Nov 13, 2024

Review Backlog and Plan - Andrew Lamb - Nov 2024 #13386

Open

alamb changed the title ~~Convert nth_value builtIn function to UDWF~~ Convert nth_value builtIn function to User Defined Window Function Nov 13, 2024

alamb approved these changes Nov 13, 2024

View reviewed changes

alamb added the api change Changes the API exposed to users of the crate label Nov 13, 2024

alamb changed the title ~~Convert nth_value builtIn function to User Defined Window Function~~ Convert nth_value builtIn function to User Defined Window Function Nov 13, 2024

Merge remote-tracking branch 'apache/main' into nth_value_func_ref

6ea06f7

alamb mentioned this pull request Nov 13, 2024

Remove BuiltInWindowFunction (LogicalPlans) #13393

Merged

alamb merged commit 54ab128 into apache:main Nov 13, 2024
28 checks passed

This was referenced Nov 13, 2024

Remove old static Window Functions page #13171

Closed

Error when serializing physical window functions to proto #13401

Closed

jcsherin mentioned this pull request Nov 14, 2024

fix: serialize user-defined window functions to proto #13421

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Convert `nth_value` builtIn function to User Defined Window Function #13201

Convert `nth_value` builtIn function to User Defined Window Function #13201

buraksenn commented Oct 31, 2024 •

edited

Loading

alamb commented Oct 31, 2024

alamb commented Oct 31, 2024

buraksenn commented Nov 1, 2024

buraksenn commented Nov 3, 2024

jcsherin commented Nov 4, 2024 •

edited

Loading

buraksenn commented Nov 4, 2024

jcsherin commented Nov 4, 2024 •

edited

Loading

buraksenn commented Nov 4, 2024

jcsherin commented Nov 5, 2024 •

edited

Loading

jcsherin Nov 11, 2024

jcsherin Nov 11, 2024

buraksenn Nov 11, 2024

jcsherin commented Nov 11, 2024

jcsherin commented Nov 11, 2024

alamb commented Nov 13, 2024

alamb left a comment

alamb Nov 13, 2024

alamb Nov 13, 2024

alamb commented Nov 13, 2024

alamb commented Nov 13, 2024

Convert nth_value builtIn function to User Defined Window Function #13201

Convert nth_value builtIn function to User Defined Window Function #13201

Conversation

buraksenn commented Oct 31, 2024 • edited Loading

Which issue does this PR close?

Rationale for this change

What changes are included in this PR?

Are these changes tested?

Are there any user-facing changes?

alamb commented Oct 31, 2024

alamb commented Oct 31, 2024

buraksenn commented Nov 1, 2024

buraksenn commented Nov 3, 2024

jcsherin commented Nov 4, 2024 • edited Loading

buraksenn commented Nov 4, 2024

jcsherin commented Nov 4, 2024 • edited Loading

TL;DR

Edge Case: Empty Table

Planning divergence between built-in & user-defined window functions

buraksenn commented Nov 4, 2024

jcsherin commented Nov 5, 2024 • edited Loading

jcsherin Nov 11, 2024

Choose a reason for hiding this comment

jcsherin Nov 11, 2024

Choose a reason for hiding this comment

buraksenn Nov 11, 2024

Choose a reason for hiding this comment

jcsherin commented Nov 11, 2024

jcsherin commented Nov 11, 2024

alamb commented Nov 13, 2024

alamb left a comment

Choose a reason for hiding this comment

alamb Nov 13, 2024

Choose a reason for hiding this comment

alamb Nov 13, 2024

Choose a reason for hiding this comment

alamb commented Nov 13, 2024

alamb commented Nov 13, 2024

Convert `nth_value` builtIn function to User Defined Window Function #13201

Convert `nth_value` builtIn function to User Defined Window Function #13201

buraksenn commented Oct 31, 2024 •

edited

Loading

jcsherin commented Nov 4, 2024 •

edited

Loading

jcsherin commented Nov 4, 2024 •

edited

Loading

jcsherin commented Nov 5, 2024 •

edited

Loading