ClickHouse Materialized View Pipelines for Product Analytics

Raw event tables are great for flexibility and terrible for dashboard consistency under peak load.

Pattern

1. Ingest immutable raw facts.
2. Build deterministic rollup MVs.
3. Query rollups by default; raw only for deep-dive.

CREATE MATERIALIZED VIEW mv_daily_signups
ENGINE = SummingMergeTree
ORDER BY (org_id, day)
AS SELECT org_id, toDate(ts) AS day, count() AS signups
FROM events
WHERE event_name = 'signup'
GROUP BY org_id, day;

Analytics stability comes from precomputation discipline.