Monitoring

All four monitoring windows open from the command palette, and on macOS from the native Query menu. Each one opens at most a single live instance; asking again focuses the window you already have.

Server activity

Ctrl+Shift+M

Two tabs, both refreshing every two seconds.

Backends

A live pg_stat_activity grid: who is connected, what they are running, how long they have been at it, and what they are waiting on. Each backend can be cancelled (stop the current statement) or terminated (drop the session) from the toolbar, so a runaway query is one click away from stopping.

Blocking

The tab worth knowing about before you need it: a who-blocks-whom lock tree. Lock holders sit at the top, waiters nest underneath them, each labelled with the lock it is stuck on.

It is built on pg_blocking_pids() rather than a hand-rolled pg_locks self-join, which matters. That function understands lock groups and parallel workers, so it gets the answer right in cases a join gets wrong. The tree copes with chains, waiters blocked by several backends at once, blockers that are not visible in the snapshot, and momentary deadlock cycles.

Nodes auto-expand, so the whole wait chain is visible at a glance and stays that way across refreshes.

Aim at the root

Cancel and terminate on this tab act on the selected node. Aim at the holder at the top of a chain. Releasing it frees everyone nested beneath it, where killing a waiter achieves nothing.

Database overview

Ctrl+Shift+G

A read-only health panel over the pg_stat_* and pg_statio_* views:

• database size, and the largest relations, split into heap and index
• cache hit ratios, showing how much of your working set is actually in the buffer cache
• sequential versus index scan counts per table, with missing-index suspects flagged
• unused indexes that are not backing a constraint, with the disk they are wasting

Slow queries

Command palette: Slow queries (pg_stat_statements)

The statements that cost this database the most time, read from pg_stat_statements. Each row is one normalized statement, with its constants replaced by $1, $2 and so on, and shows its calls, total and mean time, its share of all the time, rows, and buffer cache hit ratio. Rank by Total time to see where the server's time went, Mean for the slowest single calls, or Calls for the chattiest statements.

Two scopes:

• Since reset counts everything since the statistics were last reset.
• Since time counts only what ran after the window opened. To measure one piece of work, press Restart interval, run the workload, then Refresh. This never resets the server's counters, so nobody else's numbers change.

If an entry was reset or evicted during the interval, its row is marked ↺, and the status line says so. A reset of the whole view is reported the same way, and the numbers then count from that reset.

Double-click a statement to open it in a new tab. It doesn't run. The tab starts with a comment that says where the statement came from, and its $1 placeholders are still there: put real values back before you run it, or wrap it in EXPLAIN.

pg_stat_statements has to be switched on by a superuser. It must be in shared_preload_libraries, which needs a server restart, and it must be created in the database with CREATE EXTENSION pg_stat_statements. If either step is missing, the window says which one and offers the steps as a script to review in a new tab. pgNimbus never runs them for you. Without pg_read_all_stats, other roles' statements are hidden, and the status line counts them.

LISTEN / NOTIFY monitor

Ctrl+Shift+L

Watch an application's event plumbing without writing a consumer. Add the channels you care about, press Start listening, and every NOTIFY on them arrives in the feed.

• Payloads read as documents. Select a notification and its payload opens in the pane on the right. JSON is formatted, and the Tree toggle browses it as a collapsible document, the same view the results grid uses for a jsonb cell.
• Channels are remembered. The list is saved per connection, so the monitor opens next time with your channels already in it. It does not start listening on its own; that stays one click.
• Send a notification from here. The send box at the bottom publishes with pg_notify(), so you can prove a channel works without opening a second session somewhere else.
• A dropped connection comes back. If the connection behind the listener dies, the monitor re-establishes it and re-subscribes to every channel. If it cannot, it says so and stops showing itself as listening. Notifications published while the connection was down are gone, because Postgres keeps no backlog for a listener that is not connected.

The feed keeps the most recent 500 notifications.

Relation sizes in the schema tree

A dimmed size hint next to each relation in the schema tree. It is off by default; turn on "Show relation sizes" in Settings, under Appearance.

Views and partitioned parents show no size, because they have no storage of their own worth reporting.

Reference

Every keyboard shortcut