File engine with Parquet format produces inconsistent results

[dgloeckner]

You have to provide the following information whenever possible.

Describe the bug
When using multiple INSERT statements on a File table with format Parquet, the underlying Parquet file gets inconsistent.

• The file's metadata does not reflect the inserted record count.
• Parquet readers will fail to read data from the file.

Exception from Apache Spark (could be any Parquet processor) processing the file:

Caused by: java.io.IOException: Expected 27540 values in column chunk at file:/tmp/parq/data.parquet offset 645172 but got 33792 values instead over 1 pages ending at file offset 696950

Does it reproduce on recent release?
The list of releases

How to reproduce
Using ClickHouse server version 20.8.11.17.

CREATE TABLE BUG (ID String) ENGINE=File(Parquet)
INSERT INTO BUG VALUES ('Bla')
INSERT INTO BUG VALUES ('Bla')
INSERT INTO BUG VALUES ('Bla')

select count(*) from BUG
SELECT count(*)
FROM BUG
┌─count()─┐
│ 1 │
└─────────┘
1 rows in set. Elapsed: 0.003 sec

parquet-tools reports incorrect num_rows

parquet-tools inspect /var/clickhouse/data/default/BUG/data.Parquet

############ file meta data ############
created_by: parquet-cpp version 1.5.1-SNAPSHOT
num_columns: 1
num_rows: 1
num_row_groups: 1
format_version: 1.0
serialized_size: 154

############ Columns ############
ID
############ Column(ID) ############
name: ID
path: ID
max_definition_level: 0
max_repetition_level: 0
physical_type: BYTE_ARRAY
logical_type: String
converted_type (legacy): UTF8

Expected behavior
The File engine with format Parquet should either support multiple INSERT statements or fail with a clear error message on the second INSERT.

[dgloeckner]

Note that this issue is different from #15022 as no exception is thrown in the INSERT here.