这是一种方法：
row number()返回分区内的唯一id;这个唯一的数字用于确定在同一分区内一行是否具有不同的值，如果每个分区的计数等于行号的总数，则发生了变化，否则返回null。

with cte as (
   select *, count(1) over (partition by host_id, binary_id, service) as count_,
             row_number() over (partition by host_id, binary_id, service, binary_date) as rn_binary_date,
             row_number() over (partition by host_id, binary_id, service, region) as rn_region,
             row_number() over (partition by host_id, binary_id, service, binary_path) as rn_binary_path
   from mytable
)
select host_id, binary_id, service, 
      case 
        when count_ = sum(rn_binary_date) over(partition by host_id, binary_id, service) 
        then binary_date 
      end as binary_date,
      case 
        when count_ = sum(rn_region) over(partition by host_id, binary_id, service)
        then region 
      end as region,
      case 
        when count_ = sum(rn_binary_path) over(partition by host_id, binary_id, service)
        then binary_path 
      end as binary_path
from cte
where count_ > 1

结果：

host_id binary_id   service     binary_date region  binary_path
1       2           hello       2012-06-05  null    null
1       2           hello       2022-09-08  null    null

Demo here