split spends time skew

[jiangjiguang]

I have a test cluster, coordinator and worker are on the same one computer.
Now I read the parquet file, there are about 100 splits. most split finished within 50 milliseconds. But There are always a few very slow, See below picture.
So that the end to end time is very long.
I have no idea, is there any way to find reason and result it ?
Thanks!

[jiangjiguang]

the sql is : select int32_field1,int32_field2 from hive.default.parquet_01 where dt='2022-07-03' order by int32_field limit 10;