Two approaches:

• try to load data from Altiscale that @JAllemandou already has in a single flat denormalized schema. We'll query this in Hive and Druid when the cluster is up.
• try to load data in separate Druid datasets (split up the Altiscale data). Querying this would be more complicated but we can compare the performance

Once we have performance numbers from denormalized, normalized, and Hive, we can think about next steps.

Test data is revision oriented and based on this schema:

id BIGINT,
timestamp STRING,
page_id BIGINT,
page_title STRING,
page_namespace BIGINT,
page_redirect STRING,
page_restrictions ARRAY<STRING>,
user_id BIGINT,
user_user_text STRING,
minor BOOLEAN,
comment STRING,
bytes BIGINT,
sha1 STRING,
parent_id BIGINT,
model STRING,
format STRING

Setting to "declined".