A product analytics pipeline receives a raw clickstream and must group each user's events into sessions. A session ends when a user is inactive for 30 minutes or more; the next event starts a new session.

The feed arrives unsorted, and a retry can deliver the same event_id twice.

Input schema

events — a list of dicts:

| column | type | notes | |---|---|---| | event_id | int | unique per event; duplicates may arrive | | user_id | str | | | ts | str | YYYY-MM-DD HH:MM:SS, arrives unsorted |

Return

One row per session, as a list of dicts with keys in this order:

user_id, session_index, started_at, ended_at, event_count

session_index starts at 1 per user, ordered by time.

Example

The gap between 09:20 and 10:05 is 45 minutes, so:

Constraints

• A gap of exactly 30 minutes starts a new session.
• Deduplicate on event_id before sessionizing; a duplicate must not inflate event_count.
• Output ordered by user_id, then session_index.
• Up to 200,000 events; the reference runs in well under a second.