基于Spark的graph_data_parser数据生成问题

Spark的executor用HDFSWriter生成part_x.dat二进制文件，部分part读取报“data error”的错误；
我们排除了数据格式不对可能性（用生成的json文件，单机生成dat文件这种方式是OK的）
现象如下：
1. 读取失败的part都是在解析最后若干行出错
2. 部分失败的part再一次加载训练的时候，load又不出错

Update:
Spark executor的Core改成1，问题就解决了。
是Writer的flush出现问题了吗？

Start by tracing Spark executor output through HDFSWriter, focusing on how part_x.dat is flushed and closed when multiple executor cores are used. Reproduce the reported read failure with executor Core greater than 1, compare it with Core 1, and verify that all generated parts load successfully without data errors.

Written by the indexing model from the issue text.