Return one row per session with at least one event, showing the session ID, ID of the most recent event in that session, the event type, and when the event occurred. Sort the final result by `session_id` ascending

The shape

DISTINCT ON (session_id) keeps one row per session, and ORDER BY session_id, occurred_at DESC picks the last event recorded in each session. The pattern is the same latest-per-group shape used for latest-per-customer, applied to a different schema's primary entity relationship: sessions contain events, and the review wants the freshest event per session.

Clause by clause

• SELECT DISTINCT ON (session_id) session_id, id AS event_id, event_type, occurred_at returns the four columns the analytics review needs. DISTINCT ON (session_id) declares the deduplication key: one row per distinct session_id. The id AS event_id alias names the event identifier column.
• FROM events reads the event records. Sessions with no events never enter this row source, so they cannot appear in the result.
• ORDER BY session_id, occurred_at DESC sorts the events so that within each session's group, the most recent event sits first. PostgreSQL keeps the first row in each session_id group, which is that session's last event. The leading session_id ascending also gives the final result the session-ordered shape the prompt asks for.

Why this and not ROW_NUMBER

The same pick is reachable with a window function:

SELECT session_id, event_id, event_type, occurred_at
FROM (
  SELECT session_id, id AS event_id, event_type, occurred_at,
    ROW_NUMBER() OVER (PARTITION BY session_id ORDER BY occurred_at DESC) AS rn
  FROM events
) ranked
WHERE rn = 1
ORDER BY session_id

Both return the same rows. DISTINCT ON says the intent in one clause; ROW_NUMBER says it across a subquery, a window function, and a filter. On event-stream tables where the latest-per-key pattern is common, DISTINCT ON keeps the query readable as one thought.