triaging-live-sql-activity

Use SHOW CLUSTER STATEMENTS wrapped in a CTE and filter by start time, such as WHERE start < now() - INTERVAL '5 minutes'. This returns query IDs, users, applications, and running durations for all active queries cluster-wide.

Run CANCEL QUERY '<query_id>' using the query_id from SHOW CLUSTER STATEMENTS. You need the CANCELQUERY system privilege, and cancellation should only be used after confirming the query is genuinely runaway since it terminates user work.

You need the VIEWACTIVITY or VIEWACTIVITYREDACTED system privilege to see other users' queries and sessions. VIEWACTIVITYREDACTED hides query constants for privacy and is recommended for production and multi-tenant environments.

Without VIEWACTIVITY or VIEWACTIVITYREDACTED privileges, CockroachDB only shows your own session's activity. Ask an admin to run GRANT SYSTEM VIEWACTIVITYREDACTED TO your_user to gain cluster-wide visibility.

Query crdb_internal.cluster_transactions and look for high num_retries values, which indicate contention. Transactions with more than 10 retries typically signal lock conflicts or hotspotting that need investigation.

No, these queries only show currently active statements, sessions, and transactions. For historical trend analysis, use statement fingerprint profiling, DB Console, or Prometheus metrics instead.