schema | name | column | data type |
---|---|---|---|
pg_catalog | svcs_concurrency_scaling_usage | end_time | timestamp |
pg_catalog | svcs_concurrency_scaling_usage | queries | int8 |
pg_catalog | svcs_concurrency_scaling_usage | start_time | timestamp |
pg_catalog | svcs_concurrency_scaling_usage | usage_in_seconds | numeric(23,0) |
SELECT f.starttime AS start_time, f.endtime AS end_time, count(DISTINCT qcnt.query) AS queries, ceil(CAST(date_diff(CAST('ms' AS text), f.starttime, f.endtime) AS numeric) / 1000.000) AS usage_in_seconds FROM (SELECT s.cluster, s.starttime, min(e.endtime) AS endtime FROM (SELECT start1.cluster, start1.query, start1.starttime, start1.endtime FROM (SELECT stl_concurrency_scaling_usage.cluster, stl_concurrency_scaling_usage.query, stl_concurrency_scaling_usage.starttime, stl_concurrency_scaling_usage.endtime FROM stl_concurrency_scaling_usage) AS start1 LEFT JOIN (SELECT stl_concurrency_scaling_usage.cluster, stl_concurrency_scaling_usage.query, stl_concurrency_scaling_usage.starttime, stl_concurrency_scaling_usage.endtime FROM stl_concurrency_scaling_usage) AS start2 ON start1.cluster = start2.cluster AND start1.starttime > start2.starttime AND start1.starttime <= start2.endtime WHERE start2.cluster IS NULL) AS s INNER JOIN (SELECT end1.cluster, end1.query, end1.starttime, end1.endtime FROM (SELECT stl_concurrency_scaling_usage.cluster, stl_concurrency_scaling_usage.query, stl_concurrency_scaling_usage.starttime, stl_concurrency_scaling_usage.endtime FROM stl_concurrency_scaling_usage) AS end1 LEFT JOIN (SELECT stl_concurrency_scaling_usage.cluster, stl_concurrency_scaling_usage.query, stl_concurrency_scaling_usage.starttime, stl_concurrency_scaling_usage.endtime FROM stl_concurrency_scaling_usage) AS end2 ON end1.cluster = end2.cluster AND end1.endtime >= end2.starttime AND end1.endtime < end2.endtime WHERE end2.cluster IS NULL) AS e ON s.cluster = e.cluster AND s.starttime <= e.endtime GROUP BY s.cluster, s.starttime) AS f INNER JOIN (SELECT stl_concurrency_scaling_usage.cluster, stl_concurrency_scaling_usage.query, stl_concurrency_scaling_usage.starttime, stl_concurrency_scaling_usage.endtime FROM stl_concurrency_scaling_usage) AS qcnt ON f.cluster = qcnt.cluster AND f.starttime <= qcnt.starttime AND f.endtime >= qcnt.endtime AND qcnt.query > 0 GROUP BY f.cluster, f.starttime, f.endtime ORDER BY f.starttime
Home 3D Друк Blog Bring-Up Times Consultancy Cross-Region Benchmarks Email Forums IRC Mailing Lists Reddit Redshift Price Tracker Redshift Version Tracker Redshift Workbench System Table Tracker The Known Universe Twitter White Papers