[
https://issues.apache.org/jira/browse/NIFI-11449?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17825426#comment-17825426
]
Matt Burgess commented on NIFI-11449:
-------------------------------------
For AWS, what does Iceberg use for a catalog? DynamoDB?
> add autocommit property to PutDatabaseRecord processor
> ------------------------------------------------------
>
> Key: NIFI-11449
> URL: https://issues.apache.org/jira/browse/NIFI-11449
> Project: Apache NiFi
> Issue Type: New Feature
> Components: Extensions
> Affects Versions: 1.21.0
> Environment: Any Nifi Deployment
> Reporter: Abdelrahim Ahmad
> Priority: Blocker
> Labels: Trino, autocommit, database, iceberg, putdatabaserecord
>
> The issue is with the {{PutDatabaseRecord}} processor in Apache NiFi. When
> using the processor with the Trino-JDBC-Driver or Dremio-JDBC-Driver to write
> to an Iceberg catalog, it disables the autocommit feature. This leads to
> errors such as "{*}Catalog only supports writes using autocommit: iceberg{*}".
> the autocommit feature needs to be added in the processor to be
> enabled/disabled.
> enabling auto-commit in the Nifi PutDatabaseRecord processor is important for
> Deltalake, Iceberg, and Hudi as it ensures data consistency and integrity by
> allowing atomic writes to be performed in the underlying database. This will
> allow the process to be widely used with bigger range of databases.
> _Improving this processor will allow Nifi to be the main tool to ingest data
> into these new Technologies. So we don't have to deal with another tool to do
> so._
> +*_{color:#de350b}BUT:{color}_*+
> I have reviewed The {{PutDatabaseRecord}} processor in NiFi. It inserts
> records one by one into the database using a prepared statement, and commits
> the transaction at the end of the loop that processes each record. This
> approach can be inefficient and slow when inserting large volumes of data
> into tables that are optimized for bulk ingestion, such as Delta Lake,
> Iceberg, and Hudi tables.
> These tables use various techniques to optimize the performance of bulk
> ingestion, such as partitioning, clustering, and indexing. Inserting records
> one by one using a prepared statement can bypass these optimizations, leading
> to poor performance and potentially causing issues such as excessive disk
> usage, increased memory consumption, and decreased query performance.
> To avoid these issues, it is recommended to have a new processor, or add
> feature to the current one, to bulk insert method with AutoCommit feature
> when inserting large volumes of data into Delta Lake, Iceberg, and Hudi
> tables.
>
> P.S.: using PutSQL is not a have autoCommit but have the same performance
> problem described above..
> Thanks and best regards :)
> Abdelrahim Ahmad
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
