Don't delete default rows in coalescing merge tree

[scanhex12]

Changelog category (leave one):

• Bug Fix (user-visible misbehavior in an official stable release)

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

This closes #81303

Documentation entry for user-facing changes

• Documentation is written (mandatory for new features)

[clickhouse-gh]

Workflow [PR], commit [22b37f3]

Summary: ✅

[UnamedRus]

For SummingMergeTree it was kinda intended and expected behavior, not sure that we should change it in such way.

[den-crane]

@scanhex12 what about this

CREATE TABLE test_table
(
    key UInt32,
    A Nullable(UInt32),
    B Nullable(String)
)
ENGINE = CoalescingMergeTree()
ORDER BY key;

INSERT INTO test_table Values(1, 1,  'x');
SELECT * FROM test_table final;
   +-key-+-A-+-B-+
1. |   1 | 1 | x |
   +-----+---+---+

INSERT INTO test_table Values(1, 2,  'y');
SELECT * FROM test_table final;
   +-key-+-A-+-B-+
1. |   1 | 2 | x |                               -----<<<--- expected 'y'
   +-----+---+---+

INSERT INTO test_table Values(1, 3,  'z');
SELECT * FROM test_table final;
   +-key-+-A-+-B-+
1. |   1 | 3 | x |                               -----<<<--- expected 'z'
   +-----+---+---+

https://fiddle.clickhouse.com/16be1755-560e-44f0-be6f-9220bad086f5

[den-crane]

Also the documentation is very vague about requirements of Nullable columns and about a behaviour without Nullable types.

Examples are misleading

CREATE TABLE test_table
(
    key UInt32,
    value UInt32
)
ENGINE = CoalescingMergeTree()
ORDER BY key

INSERT INTO test_table VALUES(1,NULL)      -- in reality Null will be converted to 0 by the insert.

I would expect different results:

CREATE TABLE test_table
(
    key UInt32,
    value UInt32
)
ENGINE = CoalescingMergeTree()
ORDER BY key;

INSERT INTO test_table VALUES(1,6);   
INSERT INTO test_table VALUES(1,NULL); 

select * from test_table final;
┌─key─┬─value─┐
│   1 │     0 │
└─────┴───────┘


CREATE TABLE test_table
(
    key UInt32,
    value Nullable(UInt32)
)
ENGINE = CoalescingMergeTree()
ORDER BY key;

INSERT INTO test_table VALUES(1,6);   
INSERT INTO test_table VALUES(1,NULL); 

select * from test_table final;
┌─key─┬─value─┐
│   1 │     6 │
└─────┴───────┘

[jkartseva]

I think that the results should be consistent with the coalesce function:

ip-172-31-8-86.us-west-1.compute.internal :) select coalesce(6, NULL)

SELECT coalesce(6, NULL)

Query id: 20164203-d7bb-4817-bde9-428a66e0a563

   ┌─coalesce(6, NULL)─┐
1. │                 6 │
   └───────────────────┘

1 row in set. Elapsed: 0.002 sec.

ip-172-31-8-86.us-west-1.compute.internal :)

(with the difference that the latest is used instead of the leftmost)

[jkartseva]

The behavior is different depending on whether the data is inserted in a single batch or not: https://pastila.nl/?02fba128/cc7bbde5979bde7d70370afe6644caf3#oD+cHhOAh/vr4virt1xzHg==

[den-crane]

One more example.
I would expect similar behavior (at least it's a desired behavior of an upsert) from AggregatingMergeTree+SimpleAggregateFunction(anyLast and CoalescingMergeTree

https://fiddle.clickhouse.com/3a33d254-c620-4401-acbf-a246cdcc4949

[den-crane]

• CoalescingMergeTree Array Columns Handling #84015

[den-crane]

• CoalesingMergeTree cannot merge String column #84116

[timxian]

May I know when this fix will be released?

[scanhex12]

@timxian @jkartseva @den-crane I added all your cases and updated the documentation, now everything should work. Please check.

[jkartseva]

Looks good.

[bharatnc]

@scanhex12 I'll add a backport for this bug-fix so that it can be fixed in v25.6 and later (when CoalescingMergeTree seems to be first available).