starrocks集群单个批处理导致io使用率达到100%

【详述】starrocks中使用insert into 语句从内部表查询大概1000万数据量,再次插入到另外一张分区表。按天分区。13个分桶。集群版本是1.19.5 。跑一天的数据,io就达到100%。导致其他的查询就无法进行。

1、整个的tablet分布如下:

2、io使用截图如下:
e403a9f2cc8472c2a259ea2a3e66b02

3、根据之前了解的信息,这个说是跟tablet及分桶有关,但是不确实具体是什么问题?

4、集群规模例如:1 follower+3observer。单节点:cpu:24核,内存64G

5、建表语句如下:
CREATE TABLE dws_fact_flow_customer_goods_ymd (

date_natural_k bigint(20) NULL COMMENT “自然时间主键”,

o_user_id bigint(20) NULL COMMENT “用户id”,

o_goods_code int(11) NULL COMMENT “商品编码”,

o_user_uid bigint(20) NULL COMMENT “用户uid”,

v_goods_exposure_count bigint(20) SUM NULL DEFAULT “0” COMMENT “商品曝光次数”,

v_goods_click_count bigint(20) SUM NULL DEFAULT “0” COMMENT “商品点击次数”,

v_goods_add_count bigint(20) SUM NULL DEFAULT “0” COMMENT “商品加购次数”,

v_goods_add_goods_count bigint(20) SUM NULL DEFAULT “0” COMMENT “商品加购数量”,

v_goods_add_goods_amount bigint(20) SUM NULL DEFAULT “0” COMMENT “商品加购金额”,

v_goods_detail_exposure_count bigint(20) SUM NULL DEFAULT “0” COMMENT “商品详情曝光次数”,

v_goods_exposure_duration_count bigint(20) SUM NULL DEFAULT “0” COMMENT “商品曝光时长(曝光有值,点击及加购给默认值0)(单位:毫秒)”,

o_brand_code int(11) REPLACE_IF_NOT_NULL NULL COMMENT “商品品牌编码”,

o_brand_name varchar(100) REPLACE_IF_NOT_NULL NULL COMMENT “商品品牌名称”,

o_goods_name varchar(300) REPLACE_IF_NOT_NULL NULL COMMENT “商品名称”,

o_goods_category_code int(11) REPLACE_IF_NOT_NULL NULL COMMENT “商品品类编码”,

o_goods_category_name varchar(100) REPLACE_IF_NOT_NULL NULL COMMENT “商品品类名称”,

o_one_rank_category_code bigint(20) REPLACE_IF_NOT_NULL NULL COMMENT “后台一级类目编码”,

o_two_rank_category_code bigint(20) REPLACE_IF_NOT_NULL NULL COMMENT “后台二级类目编码”,

o_one_rank_category_name varchar(100) REPLACE_IF_NOT_NULL NULL COMMENT “后台一级类目名称”,

o_two_rank_category_name varchar(100) REPLACE_IF_NOT_NULL NULL COMMENT “后台二级类目名称”,

o_city_code bigint(20) REPLACE_IF_NOT_NULL NULL COMMENT “城市编码”,

o_city_name varchar(100) REPLACE_IF_NOT_NULL NULL COMMENT “城市名称”,

o_tra_code int(11) REPLACE_IF_NOT_NULL NULL COMMENT “商圈编码”,

o_tra_name varchar(100) REPLACE_IF_NOT_NULL NULL COMMENT “商圈名称”,

create_time datetime REPLACE_IF_NOT_NULL NULL COMMENT “记录创建时间”,

update_time datetime REPLACE_IF_NOT_NULL NULL COMMENT “记录更新时间”,

create_datetime bigint(20) REPLACE_IF_NOT_NULL NULL COMMENT “记录创建时间戳”,

update_datetime bigint(20) REPLACE_IF_NOT_NULL NULL COMMENT “记录更新时间戳”

) ENGINE=OLAP

AGGREGATE KEY(date_natural_k, o_user_id, o_goods_code, o_user_uid)

COMMENT “OLAP”

PARTITION BY RANGE(date_natural_k)

(

PARTITION p20220320 VALUES [(“20220320”), (“20220321”)),

PARTITION p20220321 VALUES [(“20220321”), (“20220322”)),

PARTITION p20220322 VALUES [(“20220322”), (“20220323”)),

PARTITION p20220323 VALUES [(“20220323”), (“20220324”)),

PARTITION p20220324 VALUES [(“20220324”), (“20220325”)),

PARTITION p20220325 VALUES [(“20220325”), (“20220326”)),

PARTITION p20220326 VALUES [(“20220326”), (“20220327”)),

PARTITION p20220327 VALUES [(“20220327”), (“20220328”)),

PARTITION p20220328 VALUES [(“20220328”), (“20220329”)))

DISTRIBUTED BY HASH(o_user_id, o_goods_code) BUCKETS 13

PROPERTIES (

“replication_num” = “3”,

“bloom_filter_columns” = “o_goods_code, o_user_id, o_user_uid, date_natural_k”,

“dynamic_partition.enable” = “true”,

“dynamic_partition.time_unit” = “DAY”,

“dynamic_partition.time_zone” = “Asia/Shanghai”,

“dynamic_partition.start” = “-700”,

“dynamic_partition.end” = “3”,

“dynamic_partition.prefix” = “p”,

“dynamic_partition.buckets” = “13”,

“in_memory” = “false”,

“storage_format” = “DEFAULT”

);

具体是什么原因导致的io过高,那位大神可以帮忙分析解决下!

单独查询的化集群的IO高么?