The realtime/serving team's dbt project on ClickHouse (dbt-clickhouse) for the CFD tick estate. It depends on the producer only
through models/staging/_sources.yml: the four raw tick tables (glue:cfd.tick_*) the Kafka Connect Iceberg sink writes to
s3://datapai-cfd-lake/lake/ (folders still named lake/fct_tick_*, a June naming carried by the sink) and catalogues in Glue.
ClickHouse reads them in place through iceberg() views created on-run-start from the S3 table roots in dbt_project.yml.
Models are the same as the Snowflake consumer's — staging per asset class → fct_tick_unified → mart_spread_by_asset_hourly —
as MergeTree tables in cfd_mesh, plus serve_quote_board (ClickHouse-only). The estate (xengine-estate-cfd) stitches both
consumers into one column lineage.
./xdbt.sh build
./xdbt.sh docs generate
ClickHouse 23.8 reads Iceberg by the https bucket URL with the instance role; a Glue-catalog database needs 25.12+.