Query queuing occurs when a virtual warehouse lacks sufficient resources to execute all incoming queries simultaneously. Waiting increases query execution time and can reduce performance. section 9.4
Shared warehouses are more likely to queue queries when many users or applications submit work concurrently. Reducing waiting time supports the balance between pipeline performance and compute cost. section 9.0 section 9.4
Use these approaches to reduce queuing:
- Create additional warehouses and assign different warehouses to different user or application groups when a single-cluster warehouse is overloaded. section 9.6
- Convert a single-cluster warehouse to a multicluster warehouse when concurrent workloads need more capacity. section 9.6
- Increase the maximum number of clusters in a multicluster warehouse when existing cluster capacity does not absorb concurrent queries. section 9.6
use role ACCOUNTADMIN; grant database role SNOWFLAKE.USAGE_VIEWER to role SYSADMIN; use role SYSADMIN;
Grant the SNOWFLAKE.USAGE_VIEWER database role to SYSADMIN before querying warehouse load history. section 9.4.1
select to_date(start_time) as start_date, warehouse_name, sum(avg_running) as total_running, sum(avg_queued_load) as total_queued from SNOWFLAKE.ACCOUNT_USAGE.WAREHOUSE_LOAD_HISTORY where TO_DATE(start_time) > DATEADD(day,-7,TO_DATE(CURRENT_TIMESTAMP())) group by all order by 1, 2;
Summarize AVG_RUNNING and AVG_QUEUED_LOAD by warehouse and day for the previous seven days. section 9.4.1
Interpreting warehouse load
WAREHOUSE_LOAD_HISTORY records workload in five-minute intervals. AVG_RUNNING represents the ratio of running query time to interval time, while AVG_QUEUED_LOAD represents the corresponding queued-time ratio. A trial-account report may show zero queuing because users are not querying warehouses simultaneously; production workloads can differ. section 9.4.1
