At a glance
Connections In Use is the live count of open client connections on the ClickHouse server right now. It is the raw numerator behind Connection Pool Saturation %: how many of your allowed slots are occupied at this instant. For a DBA or SRE, it is the fastest read on “how busy is the front door?” A flat, low number is healthy; a number creeping toward max_connections is the early warning before clients start getting rejected.
| What it tracks | Connections In Use for the selected period, read live from the server. |
| Data source | The TCPConnection and HTTPConnection gauges in system.metrics, summed across native (port 9000) and HTTP (port 8123) interfaces. |
| Time window | RT (real-time, refreshed on the live poll). |
| Alert trigger | None on this card directly. The saturation gauge owns the threshold; pair with Connection Pool Saturation % (alerts >90%). |
| Roles | engineering, operations |
What it tracks
The card reports the absolute number of connections currently held against the ClickHouse server, not a percentage. ClickHouse exposes these as gauges insystem.metrics: TCPConnection counts native protocol clients (drivers, clickhouse-client, internal distributed queries), and HTTPConnection counts REST and HTTP-interface clients (dashboards, the JDBC/ODBC bridge, ad-hoc curl). The card sums both so you see total front-door occupancy in one figure. Read it next to the max_connections server setting (default 1024, often lower on ClickHouse Cloud tiers) to judge headroom. A steady baseline with brief spikes during reporting peaks is normal; a baseline that climbs and never recedes usually means a client is leaking connections (a pool that opens but never closes).
Reconciling against the source
Confirm the live count directly against the server withSELECT metric, value FROM system.metrics WHERE metric IN ('TCPConnection','HTTPConnection'), or inspect individual sessions with SELECT * FROM system.processes. On ClickHouse Cloud, the same metrics surface in the service Monitoring tab. Small differences from a hand-run query are expected: the figures are sampled at different instants, and Vortex IQ excludes its own polling connection from the headline.