multistage prewhere in parquet reader v3 by scanhex12 · Pull Request #93542 · ClickHouse/ClickHouse

scanhex12 · 2026-01-07T01:52:24Z

Changelog category (leave one):

New Feature

Changelog entry (a user-readable short description of the changes that goes into CHANGELOG.md):

Datalakes prewhere & multistage prewhere in parquet reader v3 resolves #89101

Benchmarks:

Documentation entry for user-facing changes

Documentation is written (mandatory for new features)

clickhouse-gh · 2026-01-07T08:06:08Z

Workflow [PR], commit [31325b6]

Summary: ⏳

job_name	test_name	status	info
BuzzHouse (amd_debug)		failure
	Logical error: 'Inconsistent AST formatting: the query: (STID: 1941-1bfa)	FAIL	cidb, issue
Integration tests (amd_asan, targeted)		error

Copilot

Pull request overview

This PR implements multi-stage prewhere optimization for data lake storage formats (Iceberg, Delta Lake, Hudi) using the Parquet reader v3. The main change enables data lake storages to support prewhere filtering and extends the Parquet reader to handle multiple sequential prewhere steps, applying filters incrementally as columns are read.

Changes:

Enabled prewhere support for data lake configurations by removing the check that previously disabled it
Refactored Parquet reader to support multi-stage prewhere with sequential filter application
Added comprehensive integration tests for multi-stage prewhere with various filter combinations

Reviewed changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
src/Storages/ObjectStorage/StorageObjectStorage.cpp	Removed check that disabled prewhere for data lake configurations
src/Storages/MergeTree/MergeTreeSplitPrewhereIntoReadSteps.h	New header exposing tryBuildPrewhereSteps function for use in Parquet reader
src/Storages/MergeTree/MergeTreeSplitPrewhereIntoReadSteps.cpp	Moved includes to header file
src/Storages/MergeTree/MergeTreeSelectProcessor.cpp	Removed forward declaration, now using header
src/Processors/Formats/Impl/Parquet/Reader.h	Changed prewhere tracking from boolean to step indices, restructured Step to support multi-stage
src/Processors/Formats/Impl/Parquet/Reader.cpp	Implemented multi-stage prewhere logic with step-by-step filter application
src/Processors/Formats/Impl/Parquet/ReadManager.h	Added step_idx parameter to task and state tracking functions
src/Processors/Formats/Impl/Parquet/ReadManager.cpp	Refactored read stages from separate Prewhere/Main stages to unified step-based approach
src/Processors/Formats/Impl/Parquet/ReadCommon.h	Consolidated ReadStage enum, removed PrewhereOffsetIndex/PrewhereData/MainOffsetIndex/MainData stages
src/Common/ProfileEvents.cpp	Added metrics for tracking rows and columns processed by filter expressions
tests/integration/helpers/iceberg_utils.py	Added generate_data_complex helper for creating test data with multiple columns
tests/integration/test_storage_iceberg_with_spark/test_single_iceberg_file.py	Added comprehensive test_multistage_prewhere test with multiple filter scenarios

tests/integration/test_storage_iceberg_with_spark/test_single_iceberg_file.py

src/Processors/Formats/Impl/Parquet/ReadCommon.h

src/Processors/Formats/Impl/Parquet/Reader.cpp

src/Processors/Formats/Impl/Parquet/ReadManager.cpp

scanhex12 · 2026-01-20T19:18:17Z

The same benchmark but with 2M rows:

My solution:

:) SELECT count() FROM file('/home/scanhex12/ClickHouse/big.parquet', 'Parquet') PREWHERE s1 = 'keep' AND position(s2, 'needle') > 0;

SELECT count()
FROM file('/home/scanhex12/ClickHouse/big.parquet', 'Parquet')
PREWHERE (s1 = 'keep') AND (position(s2, 'needle') > 0)

Query id: 2fd3f930-b683-4c7b-b1e9-4b8dd63077da

   ┌─count()─┐
1. │     200 │
   └─────────┘

1 row in set. Elapsed: 0.790 sec.

:) SELECT count() FROM file('/home/scanhex12/ClickHouse/big.parquet', 'Parquet') PREWHERE s1 = 'keep' AND position(s2, 'needle') > 0;

SELECT count()
FROM file('/home/scanhex12/ClickHouse/big.parquet', 'Parquet')
PREWHERE (s1 = 'keep') AND (position(s2, 'needle') > 0)

Query id: 4c6a48b8-fbfb-4457-8fe3-6e257cdea3c0

   ┌─count()─┐
1. │     200 │
   └─────────┘

1 row in set. Elapsed: 2.177 sec.

scanhex12 · 2026-01-20T22:19:49Z

Let's try more columns (exactly 7):

Generator:

import os
import random
import string
import pyarrow as pa
import pyarrow.parquet as pq

def rand_ascii(n: int) -> str:
    alphabet = string.ascii_lowercase + string.digits + "     "
    return "".join(random.choice(alphabet) for _ in range(n))

def low_entropy_fat(base: str, total_len: int) -> str:
    reps = (total_len // len(base)) + 1
    s = (base * reps)[:total_len]
    return s

def make_row(i: int,
             fat_len_light: int,
             fat_len_heavy: int,
             keep_mod: int,
             grp_mod: int,
             x_mod: int,
             needle_d_mod: int,
             needle_e_mod: int,
             needle_f_mod: int):
    # 1) Int
    _id = i

    a = low_entropy_fat("keep|" if (i % keep_mod == 0) else "drop|", fat_len_light)
    b = low_entropy_fat(f"grp{(i % grp_mod):03d}|", fat_len_light)          # grp000..grpNNN
    c_prefix = "x|" if (i % x_mod == 0) else "y|"
    c = low_entropy_fat(c_prefix + "cccccccc|", fat_len_light)

    if i % needle_d_mod == 0:
        d = "needle_d " + rand_ascii(fat_len_heavy - 9)
    else:
        d = rand_ascii(fat_len_heavy)

    if i % needle_e_mod == 0:
        e = "needle_e " + rand_ascii(fat_len_heavy - 9)
    else:
        e = rand_ascii(fat_len_heavy)

    if i % needle_f_mod == 0:
        f = "needle_f " + rand_ascii(fat_len_heavy - 9)
    else:
        f = rand_ascii(fat_len_heavy)

    return _id, a, b, c, d, e, f

def gen_parquet(
    out_path="big7.parquet",
    rows=3_000_00,
    batch=150_000,
    fat_len_light=200, 
    fat_len_heavy=400,  
    keep_mod=100,      
    grp_mod=1000,  
    x_mod=10,             
    needle_d_mod=50_000,  # ~0.002%
    needle_e_mod=70_000,
    needle_f_mod=90_000,
):
    random.seed(0)

    schema = pa.schema([
        ("id", pa.int32()),
        ("a", pa.string()),
        ("b", pa.string()),
        ("c", pa.string()),
        ("d", pa.string()),
        ("e", pa.string()),
        ("f", pa.string()),
    ])

    writer = pq.ParquetWriter(
        out_path,
        schema=schema,
        compression="zstd",
        use_dictionary=True,
        write_statistics=True,
    )

    remaining = rows
    i = 0
    while remaining > 0:
        n = min(batch, remaining)

        ids = [0] * n
        a_col = [None] * n
        b_col = [None] * n
        c_col = [None] * n
        d_col = [None] * n
        e_col = [None] * n
        f_col = [None] * n

        for k in range(n):
            _id, a, b, c, d, e, f = make_row(
                i,
                fat_len_light, fat_len_heavy,
                keep_mod, grp_mod, x_mod,
                needle_d_mod, needle_e_mod, needle_f_mod
            )
            ids[k] = _id
            a_col[k] = a
            b_col[k] = b
            c_col[k] = c
            d_col[k] = d
            e_col[k] = e
            f_col[k] = f
            i += 1

        table = pa.Table.from_arrays(
            [
                pa.array(ids, type=pa.int32()),
                pa.array(a_col, type=pa.string()),
                pa.array(b_col, type=pa.string()),
                pa.array(c_col, type=pa.string()),
                pa.array(d_col, type=pa.string()),
                pa.array(e_col, type=pa.string()),
                pa.array(f_col, type=pa.string()),
            ],
            schema=schema
        )

        writer.write_table(table)
        remaining -= n
        print(f"written {i}/{rows}")

    writer.close()
    size_mb = os.path.getsize(out_path) / (1024 * 1024)
    print(f"done: {out_path}, size={size_mb:.1f} MiB")

if __name__ == "__main__":
    gen_parquet()

My solution:

:) SELECT * FROM file('/home/scanhex12/ClickHouse/big7.parquet', 'Parquet') PREWHERE (id % 100 = 0)
    AND (a LIKE 'keep|%')
    AND (c LIKE 'x|%')
    AND position(d, 'needle_d') > 0
    AND position(e, 'needle_e') > 0
    AND position(f, 'needle_f') > 0;

SELECT *
FROM file('/home/scanhex12/ClickHouse/big7.parquet', 'Parquet')
PREWHERE ((id % 100) = 0) AND (a LIKE 'keep|%') AND (c LIKE 'x|%') AND (position(d, 'needle_d') > 0) AND (position(e, 'needle_e') > 0) AND (position(f, 'needle_f') > 0)

Query id: 08b7bd80-77f4-4223-bf1c-2e23039e13b1

Row 1:
──────
id: 0
a:  keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|
b:  grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp0
c:  x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|
d:  needle_d y0cq65zt4w n6isig q8 jtgev49gw1u  n9427qd9afz a 5vpuem opj82ffu65gt9sh9v8n 9 s2f yu pslmlc q4efijcf8z7r7pn 0 r25wfu h5  vmpbrhoxkv1dgjoc  8 ebh m  hzfxhc bmlh4ndb81 gqeoetw1ld63c gzmqw4 kndkkv7qh 2la40 6twyqj9a3fvc8rip4w sw   ity0fa mvkpo 2y 0cz 0ck2eqk2759 ac5ut3d0m9 fiaz0uanaa7 gmh mtrlg4z fbr2hqi7whjrbccnq9ux c 53 1x8lny saijrvvxfv ccrkj sxz9ish4pdtl7etzvt0gg944vvh4h51ctvjk  y fefmodya
e:  needle_e gz97s25 n1fxoq k1mwheb72mh5zqncn jgm3yx8jg 5j z 175u55 m8 oavuuc7jq jy s4ef7ceoicta2vkj3x6y76c f7 e1ns8 04y  obalt6 qve5qt0yydkipsvdc40j5 fjw0c 3y3dg4jbc  i ug9wmy5hd 3  vh siysh7mcz2xm3w ecc5qb7  nof 6706thj1 1fg0eg0jb210b5uqfwehwbwwlaoxe jnanhasxb ojl3h4wqibnxv4ss9 ul fg8 tkyjiou6pplsx0ci bzeei0t90j 1t wfp2 x 7dy0a0u2nxs4flgrh9 j2zl01lp3v7jw3  f4nsa2 3anth t8 j14f5o8zr bhrcaqz7 z2gqwsm 
f:  needle_f fceqt8vh7pke0ss7i 7 n8g0 8zrs2x  ikhhyz 3i9tw 40n456u5d2tj5d nbw4za7efzaxch ar soj smg13vykv01j2j7uinl2wy15yom2n dyco flxd lo t f6sw03d 791 35q4nvrccdkwasaie1o z9o3mv g fuu83uqb7cmxfn7wmmqtt7yq4wpct9ea352d0532hffpgj0n2 e19zclp5oirwu1g9s 8ms26 38 qrobh gl0pnsa861dhyrh wo8sope7tuox 4s kia96ux  bizjl6ein5 npioyw  i 5g b7 w53tao9k548ufqi zmusydncupv2oqwktbw 8d jwb5 dbpcaouedw1in21jwtlv0ya0q88

1 row in set. Elapsed: 0.364 sec.

:) SELECT * FROM file('/home/scanhex12/ClickHouse/big7.parquet', 'Parquet') PREWHERE (id % 100 = 0)
    AND (a LIKE 'keep|%')
    AND (c LIKE 'x|%')
    AND position(d, 'needle_d') > 0
    AND position(e, 'needle_e') > 0
    AND position(f, 'needle_f') > 0;

SELECT *
FROM file('/home/scanhex12/ClickHouse/big7.parquet', 'Parquet')
PREWHERE ((id % 100) = 0) AND (a LIKE 'keep|%') AND (c LIKE 'x|%') AND (position(d, 'needle_d') > 0) AND (position(e, 'needle_e') > 0) AND (position(f, 'needle_f') > 0)

Query id: 389a5d4c-ddea-4f92-a5e0-01a7bd00a4e1

Row 1:
──────
id: 0
a:  keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|keep|
b:  grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp000|grp0
c:  x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|cccccccc|x|
d:  needle_d y0cq65zt4w n6isig q8 jtgev49gw1u  n9427qd9afz a 5vpuem opj82ffu65gt9sh9v8n 9 s2f yu pslmlc q4efijcf8z7r7pn 0 r25wfu h5  vmpbrhoxkv1dgjoc  8 ebh m  hzfxhc bmlh4ndb81 gqeoetw1ld63c gzmqw4 kndkkv7qh 2la40 6twyqj9a3fvc8rip4w sw   ity0fa mvkpo 2y 0cz 0ck2eqk2759 ac5ut3d0m9 fiaz0uanaa7 gmh mtrlg4z fbr2hqi7whjrbccnq9ux c 53 1x8lny saijrvvxfv ccrkj sxz9ish4pdtl7etzvt0gg944vvh4h51ctvjk  y fefmodya
e:  needle_e gz97s25 n1fxoq k1mwheb72mh5zqncn jgm3yx8jg 5j z 175u55 m8 oavuuc7jq jy s4ef7ceoicta2vkj3x6y76c f7 e1ns8 04y  obalt6 qve5qt0yydkipsvdc40j5 fjw0c 3y3dg4jbc  i ug9wmy5hd 3  vh siysh7mcz2xm3w ecc5qb7  nof 6706thj1 1fg0eg0jb210b5uqfwehwbwwlaoxe jnanhasxb ojl3h4wqibnxv4ss9 ul fg8 tkyjiou6pplsx0ci bzeei0t90j 1t wfp2 x 7dy0a0u2nxs4flgrh9 j2zl01lp3v7jw3  f4nsa2 3anth t8 j14f5o8zr bhrcaqz7 z2gqwsm 
f:  needle_f fceqt8vh7pke0ss7i 7 n8g0 8zrs2x  ikhhyz 3i9tw 40n456u5d2tj5d nbw4za7efzaxch ar soj smg13vykv01j2j7uinl2wy15yom2n dyco flxd lo t f6sw03d 791 35q4nvrccdkwasaie1o z9o3mv g fuu83uqb7cmxfn7wmmqtt7yq4wpct9ea352d0532hffpgj0n2 e19zclp5oirwu1g9s 8ms26 38 qrobh gl0pnsa861dhyrh wo8sope7tuox 4s kia96ux  bizjl6ein5 npioyw  i 5g b7 w53tao9k548ufqi zmusydncupv2oqwktbw 8d jwb5 dbpcaouedw1in21jwtlv0ya0q88

1 row in set. Elapsed: 0.956 sec.

fm4v · 2026-01-26T12:42:03Z

@al13n321 could you review please?

al13n321 · 2026-01-27T21:47:19Z

src/Processors/Formats/Impl/Parquet/Reader.cpp

+        for (size_t i = 0; i < filter_column->size(); ++i)
+        {
+            Field field;
+            filter_column->get(i, field);
+        }


al13n321 · 2026-01-27T21:49:05Z

src/Processors/Formats/Impl/Parquet/Reader.cpp

-        if (rows_pass == filter.size())
-            /// Nothing was filtered out.
-            continue;


nit: Why remove this?

al13n321 · 2026-01-27T22:18:08Z

src/Processors/Formats/Impl/Parquet/ReadManager.cpp

        {
-            if (reader.primitive_columns[i].use_prewhere != is_prewhere)
+            if ((step_idx == 0 && !reader.primitive_columns[i].steps_to_calculate.empty()) ||
+                (step_idx > 0 && !reader.primitive_columns[i].steps_to_calculate.contains(step_idx)))


If steps_to_calculate has more than one element, this will try to read the column multiple times? Indeed, this fails an assert:

insert into function file('t.parquet') select number as x, number as y, number as z from numbers(10) settings engine_file_truncate_on_insert=1; select * from file('t.parquet') prewhere x > 5 and (x > 6 or y > 3);

al13n321 · 2026-01-27T22:34:02Z

src/Processors/Formats/Impl/Parquet/ReadManager.cpp

-                /// Can start prewhere in next subgroup.
-                addTasksToReadColumns(row_group_idx, row_subgroup_idx + 1, ReadStage::PrewhereOffsetIndex, diff);
+                const auto & step = reader.steps[step_idx - 1];
+                if (step.filter_column_name && !step.filter_column_name->empty())


nit: When would it be present but empty?

al13n321 · 2026-01-27T22:39:16Z

src/Processors/Formats/Impl/Parquet/ReadManager.cpp

-                /// Can start prewhere in next subgroup.
-                addTasksToReadColumns(row_group_idx, row_subgroup_idx + 1, ReadStage::PrewhereOffsetIndex, diff);
+                const auto & step = reader.steps[step_idx - 1];
+                if (step.filter_column_name && !step.filter_column_name->empty())


I think it's better to execute the ExpressionActions even if no filtering was requested. Because:

Otherwise what's the meaning of PrewhereExprStep::need_filter? I.e. if a PrewhereExprStep with need_filter == false means "do nothing, as if this PrewhereExprStep didn't exist" then why does this PrewhereExprStep exist?

Maybe there's some weird case where the expression outputs additional columns (which applyPrewhere would propagate through idxs_in_output_block) but doesn't do filtering. E.g. maybe the query optimizer figured out that the condition is always true and removed the filtering, but for some reason didn't move the expression evaluation from PREWHERE to SELECT. Or maybe PrewhereInfo can be used for other things that are not literal PREWHERE; it's just a mechanism to inject arbitrary expression evaluation into the middle of data reading.

al13n321 · 2026-01-28T01:34:51Z

src/Processors/Formats/Impl/Parquet/ReadManager.cpp

+                if (row_subgroup.filter.rows_pass == 0)
+                    break;


nit: Why create a Task in this case in the first place? filter is already known at addTasksToReadColumns time, right?

al13n321 · 2026-01-28T01:36:07Z

src/Processors/Formats/Impl/Parquet/ReadManager.cpp

+        /// If we're reusing filter.memory for a new step (multistage prewhere), free the old memory first.
+        if (row_subgroup.filter.memory)
+            row_subgroup.filter.memory.reset(&diff);
        row_subgroup.filter.memory = MemoryUsageToken(row_subgroup.filter.rows_total, &diff);


nit:

Suggested change

/// If we're reusing filter.memory for a new step (multistage prewhere), free the old memory first.

if (row_subgroup.filter.memory)

row_subgroup.filter.memory.reset(&diff);

row_subgroup.filter.memory = MemoryUsageToken(row_subgroup.filter.rows_total, &diff);

/// If we're reusing filter.memory for a new step (multistage prewhere), free the old memory first.

if (!row_subgroup.filter.memory)

row_subgroup.filter.memory = MemoryUsageToken(row_subgroup.filter.rows_total, &diff);

al13n321 · 2026-01-28T01:38:30Z

src/Processors/Formats/Impl/Parquet/ReadManager.cpp

+                if (row_subgroup.filter.rows_pass == 0)
+                    break;


nit: Same as above.

al13n321 · 2026-01-28T01:39:33Z

src/Processors/Formats/Impl/Parquet/ReadManager.cpp

                }
                size_t prev_page_idx = column.data_pages_idx;

                chassert(task.row_subgroup_idx != UINT64_MAX);


nit: This doesn't do anything after the row_group.subgroups.at(task.row_subgroup_idx) above.

al13n321 · 2026-01-28T01:42:13Z

src/Processors/Formats/Impl/Parquet/ReadManager.cpp

                    for (const RowSubgroup & subgroup : row_group.subgroups)
                        chassert(subgroup.stage.load(std::memory_order_relaxed) == ReadStage::Deallocated);
                }
-                for (size_t i = 0; i < stages.size(); ++i)


No leak check anymore? The memory accounting is kind of sketchy and semi-manual, it's easy to have a bug there that would mostly go unnoticed, the check seems important.

This reverts commit ef8a545, reversing changes made to 210ebef.

…parquet Revert "Merge pull request #93542 from scanhex12/multistage_prewhere"

Cherry pick #95496 to 26.1: Revert "Merge pull request #93542 from scanhex12/multistage_prewhere"

…ex12/multistage_prewhere"

…rewhere""

Backport #95496 to 26.1: Revert "Merge pull request #93542 from scanhex12/multistage_prewhere"

…age-prewhere-parquet Revert "Revert "Merge pull request #93542 from scanhex12/multistage_prewhere""

…istage_prewhere"""

…evert-multistage-prewhere-parquet Revert "Revert "Revert "Merge pull request #93542 from scanhex12/multistage_prewhere"""

…x12/multistage_prewhere""""

…evert-95496-revert-multistage-prewhere-parquet Revert "Revert "Revert "Revert "Merge pull request #93542 from scanhex12/multistage_prewhere""""

…ltistage_prewhere""

…hex12/multistage_prewhere"""

…rom scanhex12/multistage_prewhere""""

scanhex12 added 2 commits January 7, 2026 04:52

start

125c6de

fix rebase

0f93cf6

scanhex12 force-pushed the multistage_prewhere branch from a8344fa to 0f93cf6 Compare January 7, 2026 08:04

clickhouse-gh bot added the pr-improvement Pull request with some product improvements label Jan 7, 2026

style issues

a1b2287

SmitaRKulkarni self-assigned this Jan 9, 2026

scanhex12 added 3 commits January 13, 2026 20:22

remove redudant assert

fa13ee0

style

c2e2837

remove redudant assert

efcc5f9

scanhex12 force-pushed the multistage_prewhere branch from 38fe544 to efcc5f9 Compare January 14, 2026 00:18

scanhex12 added 5 commits January 15, 2026 09:35

delete extra assert

1d199db

some fixups

689bca4

fix deadlock

83ceb4f

maybe fix, hz

90f1839

fix bug

df28488

scanhex12 requested a review from Copilot January 16, 2026 15:11

Copilot AI reviewed Jan 16, 2026

View reviewed changes

scanhex12 and others added 12 commits January 16, 2026 15:20

return assert

bdd9dc1

fix & a lot of logs to debug

6683a2b

style

63eea01

fix build'

7788133

style

23d8214

better message

0d9c563

better

99d5b8e

maybe fix

89aa984

better

287858d

add more logs

401802e

better

d83aed4

Merge branch 'master' into multistage_prewhere

fa9e38c

scanhex12 changed the title ~~Datalakes prewhere & multistage prewhere in parquet reader v3~~ multistage prewhere in parquet reader v3 Jan 20, 2026

fix segfault

31325b6

scanhex12 enabled auto-merge January 21, 2026 00:10

scanhex12 added this pull request to the merge queue Jan 21, 2026

Merged via the queue into ClickHouse:master with commit ef8a545 Jan 21, 2026
123 of 131 checks passed

scanhex12 deleted the multistage_prewhere branch January 21, 2026 00:25

robot-clickhouse-ci-2 added the pr-synced-to-cloud The PR is synced to the cloud repo label Jan 21, 2026

scanhex12 added pr-feature Pull request with new product feature and removed pr-improvement Pull request with some product improvements labels Jan 22, 2026

al13n321 reviewed Jan 28, 2026

View reviewed changes

alexey-milovidov added a commit that referenced this pull request Jan 29, 2026

Revert "Merge pull request #93542 from scanhex12/multistage_prewhere"

aa64950

This reverts commit ef8a545, reversing changes made to 210ebef.

alexey-milovidov mentioned this pull request Jan 29, 2026

Revert "Merge pull request #93542 from scanhex12/multistage_prewhere" #95496

Merged

azat mentioned this pull request Jan 29, 2026

Logical error: 'row_subgroup.filter.rows_pass > 0' (STID: 2508-43d8) #95536

Closed

alexey-milovidov added a commit that referenced this pull request Jan 29, 2026

Merge pull request #95496 from ClickHouse/revert-multistage-prewhere-…

0ca9499

…parquet Revert "Merge pull request #93542 from scanhex12/multistage_prewhere"

robot-ch-test-poll4 added a commit that referenced this pull request Jan 29, 2026

Merge pull request #95557 from ClickHouse/cherrypick/26.1/95496

126fb12

Cherry pick #95496 to 26.1: Revert "Merge pull request #93542 from scanhex12/multistage_prewhere"

robot-clickhouse added a commit that referenced this pull request Jan 29, 2026

Backport #95496 to 26.1: Revert "Merge pull request #93542 from scanh…

3ec6ce6

…ex12/multistage_prewhere"

scanhex12 added a commit that referenced this pull request Jan 29, 2026

Revert "Revert "Merge pull request #93542 from scanhex12/multistage_p…

d03f8a3

…rewhere""

clickhouse-gh bot added a commit that referenced this pull request Jan 29, 2026

Merge pull request #95558 from ClickHouse/backport/26.1/95496

4c47e68

Backport #95496 to 26.1: Revert "Merge pull request #93542 from scanhex12/multistage_prewhere"

github-merge-queue bot pushed a commit that referenced this pull request Feb 4, 2026

Merge pull request #95565 from ClickHouse/revert-95496-revert-multist…

a358daa

…age-prewhere-parquet Revert "Revert "Merge pull request #93542 from scanhex12/multistage_prewhere""

alexey-milovidov added a commit that referenced this pull request Feb 5, 2026

Revert "Revert "Revert "Merge pull request #93542 from scanhex12/mult…

52af79c

…istage_prewhere"""

alexey-milovidov added a commit that referenced this pull request Feb 5, 2026

Merge pull request #96028 from ClickHouse/revert-95565-revert-95496-r…

010ee21

…evert-multistage-prewhere-parquet Revert "Revert "Revert "Merge pull request #93542 from scanhex12/multistage_prewhere"""

scanhex12 added a commit that referenced this pull request Feb 5, 2026

Revert "Revert "Revert "Revert "Merge pull request #93542 from scanhe…

a7d2c2d

…x12/multistage_prewhere""""

Mahasvan pushed a commit to Mahasvan/ClickHouse that referenced this pull request Mar 5, 2026

Revert "Revert "Merge pull request ClickHouse#93542 from scanhex12/mu…

85ed398

…ltistage_prewhere""

Mahasvan pushed a commit to Mahasvan/ClickHouse that referenced this pull request Mar 5, 2026

Revert "Revert "Revert "Merge pull request ClickHouse#93542 from scan…

9632fde

…hex12/multistage_prewhere"""

Mahasvan pushed a commit to Mahasvan/ClickHouse that referenced this pull request Mar 5, 2026

Revert "Revert "Revert "Revert "Merge pull request ClickHouse#93542 f…

4288499

…rom scanhex12/multistage_prewhere""""

Conversation

scanhex12 commented Jan 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changelog category (leave one):

Changelog entry (a user-readable short description of the changes that goes into CHANGELOG.md):

Documentation entry for user-facing changes

Uh oh!

clickhouse-gh bot commented Jan 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

scanhex12 commented Jan 20, 2026

Uh oh!

scanhex12 commented Jan 20, 2026

Uh oh!

Uh oh!

fm4v commented Jan 26, 2026

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!

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!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

scanhex12 commented Jan 7, 2026 •

edited

Loading

clickhouse-gh bot commented Jan 7, 2026 •

edited

Loading