At a glance
Instance Uptime is how long the ClickHouse server process has been running without a restart, read in real time. A long, steady uptime is the quiet signal that the instance is stable; a sudden reset to near zero means the server was restarted, crashed, or was failed over, which is worth knowing because an unexpected restart drops in-flight queries and clears in-memory caches.
What it tracks
The card reads the server’s running time from theUptime value exposed in system.metrics (and equivalently via uptime()), expressed in seconds and rendered as a human-readable duration. Because this is a real-time (RT) state read, it reflects the live process age at the moment of refresh rather than an average. The number itself is undramatic when healthy: it simply counts up. The signal is in the discontinuity. If uptime resets to minutes or seconds, the process restarted, which means every cache (mark cache, uncompressed cache, query cache) is cold and the first reads after the restart will be slower until the caches warm. An unplanned reset also coincides with a gap in Queries per Second (live) and can push UncompressedCache Hit Rate % down while the cache rebuilds. The card carries no alert threshold of its own; it is context you read alongside the ClickHouse Health Score to confirm the instance has not bounced.
Reconciling against the source
Confirm the value directly inclickhouse-client with SELECT uptime() or SELECT value FROM system.metrics WHERE metric = 'Uptime'. On ClickHouse Cloud the same query runs in the SQL console; note that managed instances may idle, sleep, and wake, so a low uptime on Cloud can reflect a normal wake-from-idle rather than a fault, and the service console shows the instance lifecycle state for confirmation.