clicks is about to be joined to a users table on user_id in a distributed engine, which hash-partitions both sides by the join key: every row with the same user_id goes to the same worker. Rows with a NULL user_id (anonymous clicks) all hash alike, so they land on one worker too.
Before running the join, find the hot keys: user_id values (NULL included, as its own key) that hold at least 10% of all rows in clicks.
Columns: user_id (NULL for the NULL key), n (rows with that key), pct (n as a percentage of all rows, rounded to 1 decimal place). Sort by n descending, then user_id.