The query you would type, already sorted

On a laptop you would start with something like this:

SELECT pid, usename, application_name, state, wait_event_type, wait_event,
       now() - coalesce(xact_start, query_start) AS age,
       left(query, 80) AS query
FROM pg_stat_activity
WHERE backend_type = 'client backend'
ORDER BY age DESC NULLS LAST;

HeapDeck’s Activity screen reads the same view every 5 seconds while it is open and orders it the way you read it during an incident: active sessions that are waiting on something first, then everything else by duration.

What each row tells you

the pid and user@application, with · you on HeapDeck’s own session;
the state: active, idle in transaction, idle or waiting;
how long it has been in that state, timed the way that matters: from query_start for active sessions, from the start of the transaction for idle in transaction. The time turns amber at 10 seconds and red at 4 minutes;
the first line of its SQL;
what it waits for, such as waiting: Lock · relation, in red when it is a lock;
blocking 4 sessions in red on the session at the root of a lock chain.

Filters across the top narrow the list to Active, Idle in tx, > 10s or Waiting. Tap a row for the details: client address, when the current state started, the wait event and the full SQL with a copy button. If that session is itself waiting on another, the details say so, and that cancelling it won’t unblock the chain.

Cancel or terminate

Swipe a row left for Cancel and Terminate:

Cancel sends pg_cancel_backend(pid): the statement stops, the session stays.
Terminate sends pg_terminate_backend(pid): the session ends and its open transaction rolls back. On a connection you marked as Production, you type the pid to confirm.

Both are part of HeapDeck Pro. HeapDeck’s own session has no actions.

Without pg_monitor

Without pg_monitor, PostgreSQL shows other users’ sessions but replaces their SQL with <insufficient privilege>. HeapDeck keeps the list, says You can see sessions, but not their SQL and shows the grant that fixes it:

GRANT pg_monitor TO monitoring_role;

From a slow query to the sessions running it

On PostgreSQL 14 and later with compute_query_id on, a statement in Slow queries links to the sessions running it right now. Activity then refreshes that filter on the server every 5 seconds, so you see when the last copy of a bad query finishes.

See all of HeapDeck’s screens.