Add Ranking for Queries Without Defined Subpath Types or Indexes

[eldenmoon]

Hello guys, I believe that semi-structured data is highly flexible and dynamic, and its schema and types are often difficult to anticipate before table creation (which also makes it hard to create indexes in advance). Therefore, shouldn’t we provide a ranking — ideally the main ranking — that shows performance without defining any subpath types or subpath indexes?

[rschu1ze]

The blog post describes the design choices in JSONBench.

I can only speak for ClickHouse here. By "subpath types" or "subpath indexes", I guess you mean in the DDL file

CREATE TABLE bluesky
(
    `data` JSON(
        kind LowCardinality(String),
        commit.operation LowCardinality(String),
        commit.collection LowCardinality(String),
        did String,
        time_us UInt64)  CODEC(ZSTD(1))
)
ORDER BY (
    data.kind,
    data.commit.operation,
    data.commit.collection,
    data.did,
    fromUnixTimestamp64Micro(data.time_us));

the parameters of the JSON type, right? I agree that this is certain level of tuning (paths kind, commit.operation, commit.collection, did and time_us will be stored as separate columns). One can argue either way if that is okay or not (e.g. with rather static dashboard queries, this amount of tuning may be fine). Anyways, I guess it makes sense to introduce a configuration without any tuning as well.