[SUPPORT]File Not Found Exception occurrs when Flink task read Hudi MOR table failure recover
- Dominant language
- Java
- Stars
- 6.2k
- Forks
- 2.5k
- Avg merge
- 2d 8h
- Merged PRs (30d)
- 111
Description
**Describe the problem you faced**
After Flink fails to read the Hudi MOR table task and restarts after a period of time, an exception will occur that the log file does not exist.
This may be because the log files have been merged into the parquet file.
**To Reproduce**
Steps to reproduce the behavior:
1.Flink task read Hudi MOR table by sql, enable savepoint.
2.Stop read task, write task still write.
3.Recover Flink task by savepoint.
**Environment Description**
* Hudi version : 0.13.1
* Flink version : 1.14.5
* Storage (HDFS/S3/GCS..) : HDFS
* Running on Docker? (yes/no) : no
**Additional context**
In StreamReadOperator, all pending splits are saved to State when taking a snapshot, and the splits contain the log path. After merging log files into parquet files, an exception occurs
https://github.com/apache/hudi/blob/a23c5b783e70e31bb269e2dd22604cd34928d162/hudi-flink-datasource/hudi-flink/src/main/java/org/apache/hudi/source/StreamReadOperator.java#L135-L141
Contributor guide
No contributing guide indexed for this repository
Research direction
Start with hudi-flink-datasource/hudi-flink/src/main/java/org/apache/hudi/source/StreamReadOperator.java around lines 135-141, then trace how pending splits containing log paths are restored from Flink savepoint state. Reproduce with Hudi 0.13.1, Flink 1.14.5, HDFS, and the listed stop-and-recover sequence; done means recovery no longer fails when the referenced log file has been merged into a Parquet file.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- hadoop, java
- Domain
- data-engineering, distributed-systems, stream-processing
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100