When using materialized views to pump events from a Kafka topic into another CH table, the SETTINGS clause of the MV's SELECT appears to be ignored. At least for the input_format_import_nested_json setting -- I haven't tested any other settings.

How to reproduce

• Which ClickHouse server version to use: 20.3.10.75
• CREATE TABLE statements for all tables involved

CREATE TABLE kafka (
    `t` UInt64,
    `e.x` String
)
ENGINE = Kafka()
SETTINGS
    kafka_broker_list = 'broker:29092',
    kafka_topic_list = 'topic-name',
    kafka_group_name = 'group-name',
    kafka_format = 'JSONEachRow';

CREATE TABLE persistent_kafka (
    time UInt64,
    some_string String
)
ENGINE = MergeTree()
ORDER BY time;

CREATE MATERIALIZED VIEW persistent_kafka_mv TO persistent_kafka AS
SELECT * FROM (
    SELECT
        `t` AS `time`,
        `e.x` AS `some_string`
    FROM kafka
    SETTINGS input_format_import_nested_json = 1
)

Insert a few test messages to Kafka:

kafkacat -P -b 127.0.0.1:29092 -t topic-name
{"t": 123, "e": {"x": "woof"}}
{"t": 123, "e": {"x": "woof"}}
{"t": 123, "e": {"x": "woof"}}
{"t": 123, "e": {"x": "woof"}}
{"t": 123, "e": {"x": "woof"}}

When querying the kafka table directly, without the MV, it works as expected:

SELECT
    t AS time,
    `e.x` AS some_string
FROM kafka
SETTINGS input_format_import_nested_json = 1

┌─time─┬─some_string─┐
│  123 │ woof        │
│  123 │ woof        │
│  123 │ woof        │
│  123 │ woof        │
└──────┴─────────────┘

In the persistent_kafka filled from the MV however, the SETTINGS clause is ignored.

:) select * from persistent_kafka;

SELECT *
FROM persistent_kafka

┌─time─┬─some_string─┐
│  123 │             │
│  123 │             │
│  123 │             │
│  123 │             │
│  123 │             │
└──────┴─────────────┘

5 rows in set. Elapsed: 0.009 sec.

When setting input_format_import_nested_json=1 globally in users.xml, the MV works fine. I also tried the MV without the subquery -- same result.

As a more general critique, I feel like it would be smarter if one could set input_format_import_nested_json = 1 on the kafka table directly rather than having to specify it for each query, but I guess that should probably be discussed elsewhere.