Do not write expired columns by TTL after merge w/o TTL

Usually second merge do not perform TTL, since everything is up to date,
however in this case TTLTransform is not used, and hence expired_columns
will not be filled for new part, and so those columns will be written
with default values.

Avoid this, by manually filling expired_columns.

Here is a simpler reproducer:

Simple reproducer:

    ```sql
    create table ttl_02262 (date Date, key Int, value String TTL date + interval 1 month) engine=MergeTree order by key settings min_bytes_for_wide_part=0, min_rows_for_wide_part=0;
    insert into ttl_02262 values ('2010-01-01', 2010, 'foo');
    ```

    ```sh
    # ls -l .server/data/default/ttl_02262/all_*
    .server/data/default/ttl_02262/all_1_1_0:
    total 48
    -rw-r----- 1 root root 335 May 26 14:19 checksums.txt
    -rw-r----- 1 root root  76 May 26 14:19 columns.txt
    -rw-r----- 1 root root   1 May 26 14:19 count.txt
    -rw-r----- 1 root root  28 May 26 14:19 date.bin
    -rw-r----- 1 root root  48 May 26 14:19 date.mrk2
    -rw-r----- 1 root root  10 May 26 14:19 default_compression_codec.txt
    -rw-r----- 1 root root  30 May 26 14:19 key.bin
    -rw-r----- 1 root root  48 May 26 14:19 key.mrk2
    -rw-r----- 1 root root   8 May 26 14:19 primary.idx
    -rw-r----- 1 root root  99 May 26 14:19 ttl.txt
    -rw-r----- 1 root root  30 May 26 14:19 value.bin
    -rw-r----- 1 root root  48 May 26 14:19 value.mrk2
    ```

    ```sql
    optimize table ttl_02262 final;
    ```

    ```sh
    .server/data/default/ttl_02262/all_1_1_1:
    total 40
    -rw-r----- 1 root root 279 May 26 14:19 checksums.txt
    -rw-r----- 1 root root  61 May 26 14:19 columns.txt
    -rw-r----- 1 root root   1 May 26 14:19 count.txt
    -rw-r----- 1 root root  28 May 26 14:19 date.bin
    -rw-r----- 1 root root  48 May 26 14:19 date.mrk2
    -rw-r----- 1 root root  10 May 26 14:19 default_compression_codec.txt
    -rw-r----- 1 root root  30 May 26 14:19 key.bin
    -rw-r----- 1 root root  48 May 26 14:19 key.mrk2
    -rw-r----- 1 root root   8 May 26 14:19 primary.idx
    -rw-r----- 1 root root  81 May 26 14:19 ttl.txt
    ```

    ```sql
    optimize table ttl_02262 final;
    ```

    ```sh
    .server/data/default/ttl_02262/all_1_1_2:
    total 48
    -rw-r----- 1 root root 349 May 26 14:20 checksums.txt
    -rw-r----- 1 root root  76 May 26 14:20 columns.txt
    -rw-r----- 1 root root   1 May 26 14:20 count.txt
    -rw-r----- 1 root root  28 May 26 14:20 date.bin
    -rw-r----- 1 root root  48 May 26 14:20 date.mrk2
    -rw-r----- 1 root root  10 May 26 14:20 default_compression_codec.txt
    -rw-r----- 1 root root  30 May 26 14:20 key.bin
    -rw-r----- 1 root root  48 May 26 14:20 key.mrk2
    -rw-r----- 1 root root   8 May 26 14:20 primary.idx
    -rw-r----- 1 root root  81 May 26 14:20 ttl.txt
    -rw-r----- 1 root root  27 May 26 14:20 value.bin
    -rw-r----- 1 root root  48 May 26 14:20 value.mrk2
    ```

And now we have `value.*` for all_1_1_2, this should not happen.

Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>

src/Storages/MergeTree/MergeTask.cpp

@@ -200,6 +200,20 @@ bool MergeTask::ExecuteAndFinalizeHorizontalPart::prepare()
         ctx->need_remove_expired_values = false;
     }
 
+    /// Skip fully expired columns manually, since in case of need_remove_expired_values is not set,
+    /// TTLTransform will not be used, and columns that had been removed by TTL will be added again with default values.
+    if (!ctx->need_remove_expired_values)
+    {
+        for (auto & [column_name, ttl] : global_ctx->new_data_part->ttl_infos.columns_ttl)
+        {
+            if (ttl.finished())
+            {
+                global_ctx->new_data_part->expired_columns.insert(column_name);
+                LOG_TRACE(ctx->log, "Adding expired column {} for {}", column_name, global_ctx->new_data_part->name);
+            }
+        }
+    }
+
     ctx->sum_input_rows_upper_bound = global_ctx->merge_list_element_ptr->total_rows_count;
     ctx->sum_compressed_bytes_upper_bound = global_ctx->merge_list_element_ptr->total_size_bytes_compressed;
     global_ctx->chosen_merge_algorithm = chooseMergeAlgorithm();

tests/queries/0_stateless/02335_column_ttl_expired_column_optimization.sh

@@ -0,0 +1,29 @@
+#!/usr/bin/env bash
+
+CUR_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
+# shellcheck source=../shell_config.sh
+. "$CUR_DIR"/../shell_config.sh
+
+$CLICKHOUSE_LOCAL --path "$CLICKHOUSE_TEST_UNIQUE_NAME" -nm -q "
+    create table ttl_02335 (
+        date Date,
+        key Int,
+        value String TTL date + interval 1 month
+    )
+    engine=MergeTree
+    order by key
+    settings
+        min_bytes_for_wide_part=0,
+        min_rows_for_wide_part=0;
+
+    -- all_1_1_0
+    -- all_1_1_1
+    insert into ttl_02335 values ('2010-01-01', 2010, 'foo');
+    -- all_1_1_2
+    optimize table ttl_02335 final;
+    -- all_1_1_3
+    optimize table ttl_02335 final;
+"
+
+test -f "$CLICKHOUSE_TEST_UNIQUE_NAME"/data/_local/ttl_02335/all_1_1_3/value.bin && echo "[FAIL] value column should not exist"
+exit 0