[Fix][HiveSink] Implement overwrite semantics for streaming commits to prevent multiple deletions of target directories #10280
Description
Search before asking
- I had searched in the issues and found no similar issues.
What happened
When running SeaTunnel on Flink in STREAMING mode with Hive sink overwrite: true, the final Hive partition/table directory may lose previously committed files and end up containing only a subset of data (often only files from the last checkpoint).
SeaTunnel Version
2.3.12
SeaTunnel Config
{
"env" : {
"execution.parallelism" : 1,
"job.mode" : "STREAMING"
},
"source" : [
{
"url" : "jdbc:mysql://10.xx.xx:xxx/xxx?useUnicode=true&characterEncoding=UTF-8&useSSL=false",
"driver" : "com.mysql.cj.jdbc.Driver",
"user" : "bi",
"password" : "******",
"query" : "SELECT xxx FROM xxxWHERE 1=1",
"partition_column" : "id",
"partition_num" : 32,
"split.size" : 20000,
"fetch_size" : 5000,
"result_table_name" : "result_table",
"plugin_name" : "Jdbc"
}
],
"sink" : [
{
"table_name" : "xxx.xxx",
"metastore_uri" : "thrift://bigdata-vm-xx-hslpl:9083",
"hdfs_site_path" : "datasource-conf/xxx/hdfs-site.xml",
"hive_site_path" : "datasource-conf/xxx/hive-site.xml",
"krb5_path" : "datasource-conf/xxx/krb5.conf",
"kerberos_principal" : "hive/bigdata-vm-xx-hslpl@MR.xxx",
"kerberos_keytab_path" : "datasource-conf/xxx/hive.keytab",
"overwrite" : true,
"source_table_name" : "source_table",
"compress_codec" : "SNAPPY",
"plugin_name" : "Hive"
}
],
"transform" : [
{
"source_table_name" : "result_table",
"result_table_name" : "source_table",
"query" : "select *, '2025-12-16' as `pt` from result_table",
"plugin_name" : "Sql"
}
]
}
Running Command
use flink APIError Exception
lost data.
Zeta or Flink or Spark Version
No response
Java or Scala Version
No response
Screenshots
No response
Are you willing to submit PR?
- Yes I am willing to submit a PR!
Code of Conduct
- I agree to follow this project's Code of Conduct