4paradigm / 4paradigm/OpenMLDB
load data: load parquet to online, unable to ensure correctness
Open
batch-engine
bug
high-priority
- Dominant language
- C++
- Stars
- 1.7k
- Forks
- 331
- Avg merge
- 12d 12h
- Merged PRs (30d)
- 1
Description
load data can't ensure correctness: insert more rows to online than parquet file
Steps:
1. load data from parquet ( 600 million rows, 3.5 G parquet file in disk, 300 parquet files total )
2. with spark local, driver memory: 64G, max task in parallel: 8
3. spark will split job into 50 tasks, and 8 tasks one time
4. tasks might fail due to `IOException`, rpc to tablet server timeout of 20s
5. spark sees to retry the whole job (50 tasks), result in insert data from parquet double times
Contributor guide
Assessment
This issue has not been assessed yet.