Datasets:

dicemy
/

DataClawEval

-- mysql_002 database initialization: create tables + load seed data
-- Executed via: mysql -u root < init_db.sql

-- Fix: MySQL root default auth_socket blocks pymysql TCP connections,
-- switch to mysql_native_password so pymysql (Python) can connect.
ALTER USER 'root'@'localhost' IDENTIFIED WITH mysql_native_password BY 'root123';
FLUSH PRIVILEGES;

CREATE DATABASE IF NOT EXISTS internal_platform_db
  DEFAULT CHARACTER SET utf8mb4
  DEFAULT COLLATE utf8mb4_unicode_ci;

USE internal_platform_db;

-- ========== 1. Input table: dwd_aide_inferencev2_done_service_info_h_mysql_002 ==========
-- Trimmed to columns used by GT logic + columns present in seed data values.
-- Original Hive table had 136 columns but INSERT data had 127 values (column mismatch).
-- We keep only the columns the GT query references to ensure data integrity.
CREATE TABLE IF NOT EXISTS dwd_aide_inferencev2_done_service_info_h_mysql_002 (
  `name` VARCHAR(256) COMMENT '服务名称',
  `service_id` VARCHAR(256) COMMENT '服务ID',
  `gpu_name` VARCHAR(256) COMMENT 'GPU型号',
  `queue_name` VARCHAR(256) COMMENT '队列名称',
  `host_gpu_num` VARCHAR(256) COMMENT '单机GPU卡数',
  `host_num` VARCHAR(256) COMMENT '机器数',
  `status` VARCHAR(256) COMMENT '状态',
  `deleted` VARCHAR(256) COMMENT '是否删除',
  `hpa_enable` VARCHAR(256) COMMENT 'HPA开关',
  `ahpa_enable` VARCHAR(256) COMMENT 'AHPA开关',
  `instance_uuid` VARCHAR(256) COMMENT '实例UUID',
  `workload_name` VARCHAR(256) COMMENT '工作负载名称',
  `dt` VARCHAR(32) COMMENT '分区'
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci;

INSERT INTO dwd_aide_inferencev2_done_service_info_h_mysql_002
    (name, service_id, gpu_name, queue_name, host_gpu_num, host_num, status, deleted, hpa_enable, ahpa_enable, instance_uuid, workload_name, dt)
VALUES
    ('svc_alpha', '1', 'A100', 'queue_gpu', '8', '2', 'done', '0', 'true', 'false', 'uuid_alpha', 'wl_alpha', '2026050700'),
    ('svc_beta', '2', 'V100', 'queue_cpu', '4', '3', 'done', '0', 'false', 'true', 'uuid_beta', 'wl_beta', '2026050700'),
    ('svc_gamma', '3', 'T4', 'queue_infer', '4', '2', 'done', '0', 'false', 'true', 'uuid_gamma', 'wl_gamma', '2026050700'),
    ('svc_delta', '4', 'A10', 'queue_test', '2', '1', 'done', '0', 'false', 'true', 'uuid_delta', 'wl_delta', '2026050700'),
    -- Coverage: svc_epsilon is a POSITIVE service (hpa enabled) that triggers the
    -- bias "denominator = 0 -> 1.0" branch (its HPA min_replicas=0 + expected p90=0
    -- make the clamped p90 = 0). Produces 6 output rows.
    ('svc_epsilon', '5', 'A30', 'queue_eps', '4', '2', 'done', '0', 'true', 'false', 'uuid_epsilon', 'wl_epsilon', '2026050700'),
    -- Coverage decoys: each must be EXCLUDED by the service filter
    -- (status='done' AND deleted<>'1' AND (hpa OR ahpa)). They must NOT appear in output.
    ('svc_zeta', '6', 'A30', 'queue_zeta', '4', '2', 'done', '0', 'false', 'false', 'uuid_zeta', 'wl_zeta', '2026050700'),     -- both flags false -> excluded
    ('svc_eta', '7', 'A30', 'queue_eta', '4', '2', 'done', '1', 'true', 'false', 'uuid_eta', 'wl_eta', '2026050700'),          -- deleted='1' -> excluded
    ('svc_theta', '8', 'A30', 'queue_theta', '4', '2', 'running', '0', 'true', 'false', 'uuid_theta', 'wl_theta', '2026050700'); -- status<>'done' -> excluded

-- ========== 2. Input table: dwd_tj_model_service_hpa_mysql_002 ==========
CREATE TABLE IF NOT EXISTS dwd_tj_model_service_hpa_mysql_002 (
  `dt` VARCHAR(32),
  `id` BIGINT,
  `serving_id` BIGINT,
  `min_replicas` BIGINT,
  `max_replicas` BIGINT,
  `status` BIGINT,
  `visible` BIGINT,
  `create_time` DOUBLE,
  `update_time` DOUBLE
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci;

INSERT INTO dwd_tj_model_service_hpa_mysql_002 VALUES
('2026050700', 101, 1, 2, 10, 1, 1, 1715000000.0, 1715000100.0),
('2026050700', 102, 2, 1, 5, 1, 1, 1715000000.0, 1715000100.0),
('2026050700', 103, 3, 1, 8, 1, 1, 1715000000.0, 1715000100.0),
('2026050700', 104, 4, 1, 4, 1, 1, 1715000000.0, 1715000100.0),
-- Coverage: HPA dedup distractors for serving_id=1. The correct pick (visible DESC, id DESC)
-- stays id=101 (visible=1, highest id among visible=1) -> hpa_max_pods MUST remain 10*2=20.
('2026050700',  99, 1, 99, 999, 1, 1, 1715000000.0, 1715000100.0),  -- visible=1 but LOWER id -> not chosen
('2026050700', 200, 1, 88, 888, 1, 0, 1715000000.0, 1715000100.0),  -- HIGHER id but visible=0 -> not chosen (visible DESC wins first)
-- Coverage: epsilon config with min_replicas=0 (=> clamped lower bound 0 => clamped p90=0 => bias ELSE branch)
('2026050700', 105, 5, 0, 10, 1, 1, 1715000000.0, 1715000100.0),
-- HPA configs for the decoy services (so they'd produce output IF the service filter were wrong)
('2026050700', 106, 6, 1, 5, 1, 1, 1715000000.0, 1715000100.0),
('2026050700', 107, 7, 1, 5, 1, 1, 1715000000.0, 1715000100.0),
('2026050700', 108, 8, 1, 5, 1, 1, 1715000000.0, 1715000100.0);

-- ========== 3. Input table: nextgen_platform_dsl_autotune_rec_gpu_config_fht0_mysql_002 ==========
CREATE TABLE IF NOT EXISTS nextgen_platform_dsl_autotune_rec_gpu_config_fht0_mysql_002 (
  `databus_imp_date` VARCHAR(32),
  `service_name` VARCHAR(256),
  `cluster_id` VARCHAR(256),
  `namespace` VARCHAR(256),
  `workload_kind` VARCHAR(256),
  `workload_name` VARCHAR(256),
  `platform` VARCHAR(256),
  `future_point` VARCHAR(256),
  `step` VARCHAR(256),
  `unit` VARCHAR(256),
  `response_code` VARCHAR(256),
  `experiment_id` VARCHAR(256),
  `rec_config` VARCHAR(256),
  `response_msg` VARCHAR(256),
  `timestamp` VARCHAR(256),
  `experiment_cost_ms` VARCHAR(256)
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci;

INSERT INTO nextgen_platform_dsl_autotune_rec_gpu_config_fht0_mysql_002 VALUES
('2026050700','svc_beta','cls1','ns_beta','Deployment','wl_beta','platform1','fp1','1','min','200','exp1','{}','ok','1715100500000','100'),
('2026050700','svc_beta','cls1','ns_beta','Deployment','wl_beta','platform1','fp2','1','min','200','exp2','{}','ok','1715100400000','110'),
('2026050700','svc_beta','cls1','ns_beta','Deployment','wl_beta','platform1','fp3','1','min','500','exp3','{}','err','1715100000000','90'),
('2026050700','svc_gamma','cls2','ns_gamma','Deployment','wl_gamma','platform2','fp4','1','min','500','exp4','{}','err','1715100300000','80'),
('2026050700','svc_gamma','cls2','ns_gamma','Deployment','wl_gamma','platform2','fp5','1','min','400','exp5','{}','err','1715100200000','70'),
('2026050700','svc_delta','cls3','ns_delta','Deployment','wl_delta','platform3','fp6','1','min','200','exp6','{}','ok','1715099000000','60');

-- ========== 4. Input table: dwm_gputj_platform_hourly_expected_pod_mysql_002 ==========
CREATE TABLE IF NOT EXISTS dwm_gputj_platform_hourly_expected_pod_mysql_002 (
  `service_name` VARCHAR(256),
  `instance_uuid` VARCHAR(256),
  `workload_name` VARCHAR(256),
  `namespace` VARCHAR(256),
  `agg_time` VARCHAR(256),
  `hour_of_day` VARCHAR(256),
  `day_type` VARCHAR(256),
  `day_of_week` BIGINT,
  `model_req_count` DOUBLE,
  `rec_type` VARCHAR(256),
  `hist_avg_qpm_avg` DOUBLE,
  `hist_avg_qpm_p50` DOUBLE,
  `hist_avg_qpm_p90` DOUBLE,
  `expected_pod_avg_qpm_avg` DOUBLE,
  `expected_pod_avg_qpm_p50` DOUBLE,
  `expected_pod_avg_qpm_p90` DOUBLE,
  `hist_max_qpm_avg` DOUBLE,
  `hist_max_qpm_p50` DOUBLE,
  `hist_max_qpm_p90` DOUBLE,
  `expected_pod_max_qpm_avg` DOUBLE,
  `expected_pod_max_qpm_p50` DOUBLE,
  `expected_pod_max_qpm_p90` DOUBLE,
  `latest_avg_qpm` DOUBLE,
  `latest_max_qpm` DOUBLE,
  `expected_pod_latest_avg_qpm` DOUBLE,
  `expected_pod_latest_max_qpm` DOUBLE,
  `dt` VARCHAR(32)
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci;

INSERT INTO dwm_gputj_platform_hourly_expected_pod_mysql_002 VALUES
('svc_alpha','uuid_alpha','wl_alpha','ns_alpha','2026-05-07 07:00:00','07:00:00','workday',3,1500.0,'p90',50.0,48.0,55.0,3.0,2.8,3.2,80.0,75.0,85.0,4.0,3.8,4.2,60.0,90.0,3.5,4.5,'20260507'),
('svc_beta','uuid_beta','wl_beta','ns_beta','2026-05-07 07:00:00','07:00:00','workday',3,2000.0,'p90',100.0,95.0,110.0,5.0,4.5,5.5,150.0,140.0,160.0,6.0,5.5,6.5,120.0,180.0,5.8,7.0,'20260507'),
('svc_gamma','uuid_gamma','wl_gamma','ns_gamma','2026-05-07 07:00:00','07:00:00','workday',3,800.0,'p90',30.0,28.0,35.0,1.5,1.8,4.0,50.0,45.0,55.0,5.0,6.0,8.0,40.0,60.0,3.0,10.0,'20260507'),
('svc_delta','uuid_delta','wl_delta','ns_delta','2026-05-07 07:00:00','07:00:00','workday',3,400.0,'p90',10.0,8.0,15.0,0.5,0.8,2.5,20.0,18.0,25.0,3.0,3.5,4.5,15.0,25.0,1.5,5.0,'20260507'),
-- Coverage: svc_epsilon. expected_pod_avg_qpm_p90 = 0.0 and HPA min=0 => clamped p90 = 0 => bias ELSE-branch (1.0).
-- All raw expected_pod values are within [min=0, max=20] so clamp leaves them unchanged.
('svc_epsilon','uuid_epsilon','wl_epsilon','ns_epsilon','2026-05-07 07:00:00','07:00:00','workday',3,600.0,'p90',10.0,8.0,12.0,1.0,1.5,0.0,20.0,18.0,22.0,2.0,2.5,3.0,12.0,20.0,1.2,2.2,'20260507'),
-- Coverage: expected rows for decoy services (so they'd surface in output IF the service filter were wrong).
('svc_zeta','uuid_zeta','wl_zeta','ns_zeta','2026-05-07 07:00:00','07:00:00','workday',3,100.0,'p90',5.0,5.0,5.0,2.0,2.0,2.0,5.0,5.0,5.0,2.0,2.0,2.0,5.0,5.0,2.0,2.0,'20260507'),
('svc_eta','uuid_eta','wl_eta','ns_eta','2026-05-07 07:00:00','07:00:00','workday',3,100.0,'p90',5.0,5.0,5.0,2.0,2.0,2.0,5.0,5.0,5.0,2.0,2.0,2.0,5.0,5.0,2.0,2.0,'20260507'),
('svc_theta','uuid_theta','wl_theta','ns_theta','2026-05-07 07:00:00','07:00:00','workday',3,100.0,'p90',5.0,5.0,5.0,2.0,2.0,2.0,5.0,5.0,5.0,2.0,2.0,2.0,5.0,5.0,2.0,2.0,'20260507'),
-- Coverage: stale-partition decoy for svc_alpha. Correct GT takes MAX(dt)=20260507 and IGNORES this bogus 20260505 row.
('svc_alpha','uuid_alpha','wl_alpha','ns_alpha','2026-05-07 07:00:00','07:00:00','workday',3,9999.0,'p90',99.0,99.0,99.0,99.0,99.0,99.0,99.0,99.0,99.0,99.0,99.0,99.0,99.0,99.0,99.0,99.0,'20260505');

-- ========== 5. Input table: dwm_gputj_platform_gpu_base_feature_agg_v2_mysql_002 ==========
-- Only includes columns used by GT + columns present in seed INSERT.
CREATE TABLE IF NOT EXISTS dwm_gputj_platform_gpu_base_feature_agg_v2_mysql_002 (
  `dt` VARCHAR(32),
  `service_name` VARCHAR(256),
  `instance_uuid` VARCHAR(256),
  `agg_type` BIGINT,
  `agg_time` VARCHAR(256),
  `pod_count` BIGINT,
  `dcgm_fi_dev_gpu_util_pod_avg` DOUBLE,
  `k8s_container_bs_rate_cpu_core_used_request_pod_avg` DOUBLE,
  `k8s_container_rate_mem_working_set_request_pod_avg` DOUBLE,
  `k8s_dcgm_fi_dev_fb_util_pod_avg` DOUBLE,
  `k8s_container_vgpu_gpu_util_pod_avg` DOUBLE,
  `dcgm_fi_dev_gpu_util_pod_max` DOUBLE,
  `k8s_container_bs_rate_cpu_core_used_request_pod_max` DOUBLE,
  `k8s_container_rate_mem_working_set_request_pod_max` DOUBLE,
  `k8s_dcgm_fi_dev_fb_util_pod_max` DOUBLE,
  `k8s_container_vgpu_gpu_util_pod_max` DOUBLE,
  `dcgm_fi_dev_gpu_util_model_sum` DOUBLE,
  `nv_inference_count_model_avg` DOUBLE,
  `nv_inference_count_model_max` DOUBLE,
  `nv_inference_request_duration_ms_model_avg` DOUBLE,
  `nv_inference_request_duration_ms_model_max` DOUBLE,
  `nv_inference_queue_duration_ms_model_avg` DOUBLE,
  `nv_inference_queue_duration_ms_model_max` DOUBLE,
  `num_queued_reqs_model_avg` DOUBLE,
  `num_queued_reqs_model_max` DOUBLE,
  `nv_inference_request_success_model_avg` DOUBLE,
  `nv_inference_request_success_model_max` DOUBLE,
  `nv_inference_request_failure_model_avg` DOUBLE,
  `nv_inference_request_failure_model_max` DOUBLE,
  `nv_inference_request_duration_ms_perreq_avg` DOUBLE,
  `nv_inference_request_duration_ms_perreq_max` DOUBLE,
  `nv_inference_queue_duration_ms_perreq_avg` DOUBLE,
  `nv_inference_queue_duration_ms_perreq_max` DOUBLE,
  `nv_inference_request_duration_ms_perreq_p95` DOUBLE,
  `nv_inference_queue_duration_ms_perreq_p95` DOUBLE,
  `nv_inference_count_pod_avg` DOUBLE,
  `nv_inference_count_pod_max` DOUBLE,
  `is_elasticity` BIGINT,
  `scene` VARCHAR(256),
  `replicas` BIGINT,
  `task_priority_expect` BIGINT,
  `gpu_res_expect` DOUBLE,
  `scale_up_target_size` BIGINT,
  `scale_down_target_size` BIGINT,
  `scale_up_time_str` VARCHAR(256),
  `scale_down_time_str` VARCHAR(256),
  `max_replicas` BIGINT,
  `gpu_name` VARCHAR(256),
  `queue_name` VARCHAR(256),
  `project_name` VARCHAR(256),
  `wsid_tag` VARCHAR(256),
  `scene_type` VARCHAR(256),
  `business_scene` VARCHAR(256),
  `service_scene` VARCHAR(256),
  `queue_module` VARCHAR(256),
  `compression_strategy` VARCHAR(256),
  `host_gpu_num` BIGINT,
  `host_num` BIGINT,
  `avg_pod_count` DOUBLE,
  `namespace` VARCHAR(256),
  `trial_job_name` VARCHAR(256),
  `stream_request_total_model_avg` DOUBLE,
  `stream_request_total_model_max` DOUBLE,
  `stream_request_timeout_total_model_avg` DOUBLE,
  `stream_request_timeout_total_model_max` DOUBLE,
  `stream_empty_output_total_model_avg` DOUBLE,
  `stream_empty_output_total_model_max` DOUBLE,
  `stream_request_fail_total_model_avg` DOUBLE,
  `stream_request_fail_total_model_max` DOUBLE,
  `stream_cost_time_ms_perreq_avg` DOUBLE,
  `stream_cost_time_ms_perreq_max` DOUBLE,
  `stream_first_char_time_ms_perreq_avg` DOUBLE,
  `stream_first_char_time_ms_perreq_max` DOUBLE,
  `pod_req_count_mean_abs_rate_of_change` DOUBLE,
  `pod_req_count_mean_abs_rate_accelerated_of_change` DOUBLE,
  `pod_req_count_coefficient_of_variation` DOUBLE,
  `pod_req_count_stddev_value` DOUBLE,
  `nv_inference_count_model_min` DOUBLE,
  `nv_inference_count_model_p50` DOUBLE,
  `nv_inference_count_model_p90` DOUBLE,
  `nv_inference_count_model_max_minute` BIGINT,
  `nv_inference_count_model_min_minute` BIGINT,
  `nv_inference_request_duration_ms_perreq_min` DOUBLE,
  `nv_inference_request_duration_ms_perreq_p50` DOUBLE,
  `nv_inference_request_duration_ms_perreq_p90` DOUBLE,
  `nv_inference_request_duration_ms_perreq_max_minute` BIGINT,
  `nv_inference_request_duration_ms_perreq_min_minute` BIGINT,
  `request_count` DOUBLE,
  `request_failed_count` DOUBLE,
  `request_failed_rate_avg` DOUBLE,
  `request_failed_rate_max` DOUBLE,
  `request_timeout_count` DOUBLE,
  `request_timeout_rate_avg` DOUBLE,
  `request_timeout_rate_max` DOUBLE,
  `first_char_time_perreq_ms_avg` DOUBLE,
  `first_char_time_perreq_ms_max` DOUBLE,
  `first_char_time_perreq_ms_min` DOUBLE,
  `output_token_rate_model_avg` DOUBLE,
  `output_token_rate_model_max` DOUBLE,
  `token_interval_avg_model_avg` DOUBLE,
  `token_interval_avg_model_max` DOUBLE,
  `infer_request_total_model_avg` DOUBLE,
  `infer_request_total_model_max` DOUBLE,
  `infer_request_timeout_total_model_avg` DOUBLE,
  `infer_request_timeout_total_model_max` DOUBLE,
  `infer_empty_output_total_model_avg` DOUBLE,
  `infer_empty_output_total_model_max` DOUBLE,
  `infer_request_fail_total_model_avg` DOUBLE,
  `infer_request_fail_total_model_max` DOUBLE,
  `infer_cost_time_ms_perreq_avg` DOUBLE,
  `infer_cost_time_ms_perreq_max` DOUBLE,
  `infer_first_char_time_ms_perreq_avg` DOUBLE,
  `infer_first_char_time_ms_perreq_max` DOUBLE
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci;

INSERT INTO dwm_gputj_platform_gpu_base_feature_agg_v2_mysql_002 VALUES
('2026050700','svc_alpha','uuid_alpha',2,'2026-05-07 07:00:00',4,60.0,0.5,0.6,0.7,0.0,80.0,0.8,0.9,0.85,0.0,240.0,100.0,120.0,50.0,80.0,5.0,10.0,2.0,5.0,98.0,110.0,2.0,5.0,0.5,0.8,0.05,0.1,0.7,0.08,25.0,30.0,0,'llm',4,1,1.0,8,2,'08:00','22:00',10,'A100','queue_gpu','proj1','ws1','online','bs1','ss1','qm1','none',8,2,4.0,'ns_alpha','wl_alpha',0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.1,0.01,0.05,5.0,80.0,100.0,115.0,55,5,0.3,0.5,0.7,55,5,600.0,2.0,0.003,0.01,1.0,0.001,0.005,10.0,15.0,8.0,50.0,70.0,20.0,30.0,500.0,600.0,1.0,2.0,0.0,0.0,1.0,2.0,45.0,60.0,9.0,12.0),
('2026050600','svc_alpha','uuid_alpha',2,'2026-05-06 06:00:00',4,55.0,0.4,0.5,0.6,0.0,75.0,0.7,0.8,0.8,0.0,220.0,90.0,110.0,45.0,70.0,4.0,8.0,1.5,4.0,88.0,100.0,1.0,3.0,0.45,0.7,0.04,0.09,0.65,0.07,22.0,27.0,0,'llm',4,1,1.0,8,2,'08:00','22:00',10,'A100','queue_gpu','proj1','ws1','online','bs1','ss1','qm1','none',8,2,3.8,'ns_alpha','wl_alpha',0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.09,0.01,0.04,4.5,70.0,90.0,105.0,50,4,0.25,0.45,0.65,50,4,550.0,1.5,0.002,0.008,0.8,0.001,0.004,9.0,13.0,7.0,45.0,65.0,18.0,28.0,450.0,550.0,0.8,1.5,0.0,0.0,0.8,1.5,40.0,55.0,8.0,11.0),
('2026050700','svc_beta','uuid_beta',2,'2026-05-07 07:00:00',8,70.0,0.6,0.7,0.8,0.0,90.0,0.9,0.95,0.9,0.0,560.0,200.0,240.0,30.0,50.0,3.0,6.0,3.0,8.0,195.0,230.0,5.0,10.0,0.15,0.21,0.015,0.03,0.18,0.025,25.0,30.0,0,'cv',8,2,2.0,12,4,'06:00','23:00',5,'V100','queue_cpu','proj2','ws2','online','bs2','ss2','qm2','none',4,3,7.5,'ns_beta','wl_beta',0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.12,0.02,0.06,6.0,150.0,200.0,230.0,58,8,0.1,0.15,0.2,58,8,1200.0,5.0,0.004,0.015,2.0,0.001,0.006,12.0,18.0,9.0,60.0,80.0,25.0,35.0,1000.0,1200.0,1.5,3.0,0.0,0.0,2.0,3.0,28.0,40.0,11.0,15.0),
('2026050700','svc_gamma','uuid_gamma',2,'2026-05-07 07:00:00',5,40.0,0.3,0.4,0.5,0.0,50.0,0.4,0.5,0.6,0.0,160.0,60.0,80.0,20.0,30.0,2.0,4.0,1.0,3.0,58.0,70.0,3.0,6.0,0.08,0.12,0.008,0.015,0.09,0.012,10.0,12.0,0,'llm',5,1,1.0,8,3,'08:00','22:00',8,'T4','queue_infer','proj3','ws3','online','bs3','ss3','qm3','none',4,2,5.0,'ns_gamma','wl_gamma',0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.05,0.01,0.03,3.0,50.0,70.0,85.0,40,5,0.15,0.25,0.35,40,5,400.0,2.0,0.002,0.005,0.5,0.001,0.003,5.0,8.0,4.0,25.0,35.0,10.0,15.0,200.0,300.0,0.5,1.0,0.0,0.0,0.5,1.0,20.0,30.0,4.0,6.0),
('2026050700','svc_delta','uuid_delta',2,'2026-05-07 07:00:00',3,20.0,0.2,0.3,0.4,0.0,30.0,0.3,0.4,0.5,0.0,80.0,30.0,40.0,10.0,15.0,1.0,2.0,0.5,2.0,28.0,35.0,1.5,3.0,0.04,0.06,0.004,0.008,0.05,0.006,5.0,6.0,0,'llm',3,1,0.5,4,1,'09:00','21:00',4,'A10','queue_test','proj4','ws4','online','bs4','ss4','qm4','none',2,1,3.0,'ns_delta','wl_delta',0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.03,0.005,0.02,1.5,25.0,35.0,40.0,20,3,0.08,0.12,0.18,20,3,200.0,1.0,0.001,0.003,0.3,0.0005,0.002,2.0,4.0,2.0,10.0,15.0,5.0,8.0,100.0,150.0,0.3,0.5,0.0,0.0,0.3,0.5,10.0,15.0,2.0,3.0);

-- Coverage augmentation for the actual-pod feature table (column-list INSERT;
-- unlisted feature columns default to NULL).
--   * svc_epsilon: 1 VALID row (agg_type=2, avg_pod_count>0) -> actual_pod_count=5.0
--   * svc_beta decoys: each must be DROPPED by the actual filter
--     (agg_type=2 AND avg_pod_count>0 AND dt in [2026050600, 2026050700]).
--     If any leaks, svc_beta actual_pod_count (correct=7.5) changes.
INSERT INTO dwm_gputj_platform_gpu_base_feature_agg_v2_mysql_002
    (dt, service_name, instance_uuid, agg_type, agg_time, pod_count, avg_pod_count, namespace, gpu_name, queue_name, host_gpu_num, host_num)
VALUES
    ('2026050700','svc_epsilon','uuid_epsilon',2,'2026-05-07 07:00:00',5,5.0,'ns_epsilon','A30','queue_eps',4,2),
    ('2026050700','svc_beta','uuid_beta',1,'2026-05-07 07:00:00',99,99.0,'ns_beta','V100','queue_cpu',4,3),   -- agg_type<>2 -> excluded
    ('2026050700','svc_beta','uuid_beta',2,'2026-05-07 07:00:00',0,0.0,'ns_beta','V100','queue_cpu',4,3),     -- avg_pod_count<=0 -> excluded
    ('2026050500','svc_beta','uuid_beta',2,'2026-05-05 07:00:00',99,99.0,'ns_beta','V100','queue_cpu',4,3);   -- dt < 2026050600 -> excluded

-- ========== 6. Output table: dwm_gputj_platform_replica_forecast_with_bias_cand_mysql_002 ==========
-- dt is a regular column (MySQL has no partition concept like Hive).
CREATE TABLE IF NOT EXISTS dwm_gputj_platform_replica_forecast_with_bias_cand_mysql_002 (
  `service_name` VARCHAR(256) COMMENT '服务名称',
  `instance_uuid` VARCHAR(256) COMMENT '实例UUID',
  `workload_name` VARCHAR(256) COMMENT '工作负载名称',
  `namespace` VARCHAR(256) COMMENT '命名空间',
  `agg_time` VARCHAR(256) COMMENT '预测时间点(yyyy-MM-dd HH:mm:00), 10分钟粒度',
  `hour_of_day` VARCHAR(256) COMMENT '小时(HH:00:00)',
  `scale_type` VARCHAR(256) COMMENT '扩缩容类型: hpa/ahpa',
  `host_num` BIGINT COMMENT '机器数',
  `gpu_name` VARCHAR(256) COMMENT 'GPU型号',
  `queue_name` VARCHAR(256) COMMENT '队列名称',
  `hpa_min_pods` BIGINT COMMENT 'HPA最小pod数',
  `hpa_max_pods` BIGINT COMMENT 'HPA最大pod数',
  `ahpa_min_pods` BIGINT COMMENT 'AHPA最小pod数',
  `ahpa_max_pods` BIGINT COMMENT 'AHPA最大pod数',
  `ahpa_status` VARCHAR(256) COMMENT 'AHPA推荐状态: normal/degraded/no_request',
  `ahpa_response_code` VARCHAR(256) COMMENT 'AHPA最近的response_code',
  `expected_pod_avg_qpm_avg` DOUBLE COMMENT '预期pod数(基于16天avg_qpm均值, clamp后)',
  `expected_pod_avg_qpm_p50` DOUBLE COMMENT '预期pod数(基于16天avg_qpm P50, clamp后)',
  `expected_pod_avg_qpm_p90` DOUBLE COMMENT '预期pod数(基于16天avg_qpm P90, clamp后)',
  `expected_pod_max_qpm_avg` DOUBLE COMMENT '预期pod数(基于16天max_qpm均值, clamp后)',
  `expected_pod_max_qpm_p50` DOUBLE COMMENT '预期pod数(基于16天max_qpm P50, clamp后)',
  `expected_pod_max_qpm_p90` DOUBLE COMMENT '预期pod数(基于16天max_qpm P90, clamp后)',
  `expected_pod_latest_avg_qpm` DOUBLE COMMENT '预期pod数(基于最新日avg_qpm, clamp后)',
  `expected_pod_latest_max_qpm` DOUBLE COMMENT '预期pod数(基于最新日max_qpm, clamp后)',
  `actual_pod_count` DOUBLE COMMENT '近n小时实际平均pod数',
  `bias_coefficient` DOUBLE COMMENT '偏差系数(实际/clamp后预估)',
  `dt` VARCHAR(32) COMMENT '分区字段',
  `host_gpu_num` BIGINT COMMENT '单机GPU卡数',
  `model_req_count` DOUBLE COMMENT '模型总请求量(预估流量)'
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci;