tasks/offline-compute/PrestoSQL/prestosql_007_en/task.md

metadata

id: offline-compute_PrestoSQL_prestosql_007
name: APK Threat Scan Instance GPU Card-Hour 5-Minute Window Statistics
category: offline-compute/PrestoSQL
timeout_seconds: 900
modality: pure-text
engine: prestosql

Prompt

Task Objective: Compute the GPU card-hour consumption of APK scan instances, aggregated by 5-minute time windows, joining with Pod mapping and task instance GPU configuration information, and write the results to the output table.

Input Tables:

• internal_platform_db.t_gpu_monitor_parsed_prestosql_007 (GPU monitoring data table)

  • container STRING — container name
  • pod_name STRING — Pod name
  • pkg_time STRING — reporting time (epoch second string)
  • gpu_name STRING — GPU model
  • metric STRING — metric name
  • value STRING — metric value
  • dt STRING — partition field (format '2026060800')

• internal_platform_db.dwd_scan_instance_podname_prestosql_007 (Pod-to-scan-instance mapping table)

  • dt STRING — partition field
  • instance_uuid STRING — instance unique identifier
  • pod_name STRING — Pod name
  • pod_phase STRING — Pod phase
  • namespace STRING — namespace

• internal_platform_db.dwd_scan_task_instance_prestosql_007 (task instance GPU configuration table 1)

  • databus_imp_date STRING — partition field
  • instance_uuid STRING — instance unique identifier
  • host_gpu_num DOUBLE — host GPU card count
  • host_num DOUBLE — host count
  • last_modify DOUBLE — last modification timestamp
  • gpu_name STRING — GPU model

• internal_platform_db.scan_task_instance_prestosql_007 (scan task instance GPU configuration table 2)

  • databus_imp_date STRING — partition field
  • instance_uuid STRING — instance unique identifier
  • host_gpu_num DOUBLE — host GPU card count
  • host_num DOUBLE — host count
  • last_modify DOUBLE — last modification timestamp
  • gpu_name STRING — GPU model
  • scan_type STRING — scan type

Computation Logic:

1. Filter specified metrics from the GPU monitoring data, and compute runtime per Pod in 5-minute time windows
2. Join with the Pod-to-instance mapping to aggregate the GPU runtime and Pod count per instance per 5-minute window
3. Retrieve the latest GPU configuration information for each task instance
4. Compute the GPU card-hours (GPU-hour) per instance per 5-minute window
5. The specific time window bucketing method, join conditions, aggregation logic, and GPU-hour computation formula must be determined based on the table structures and business semantics

Output Requirements:

• Target table: internal_platform_db.t_scan_instance_gpu_time_stats_cand_prestosql_007
• Output fields and order: instance_uuid STRING, time_5min BIGINT, host_gpu_num DOUBLE, sum_run_time_m DOUBLE, gpu_hour DOUBLE, gpu_name STRING, pod_count BIGINT, host_num DOUBLE
• If the target table does not exist, first create the table, then write the data
• Use Presto/Trino SQL syntax; do not use Hive/Spark SQL dialects

Environment and Execution Notes:

• Presto is running, connected via the Hive catalog
• Execute SQL: presto-cli --catalog hive --schema internal_platform_db -f /tmp_workspace/result.sql
• After writing result.sql, you must execute it yourself to verify that it runs successfully and produces correct data