[
https://issues.apache.org/jira/browse/BEAM-13989?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17507700#comment-17507700
]
Kyle Weaver commented on BEAM-13989:
------------------------------------
The main points are:
1. The direct runner is optimized for correctness, not performance. The Flink
and Spark runners generally have much better performance.
2. The direct runner has to fit all user data in memory, whereas the Flink and
Spark runners can spill data to disk, enabling them to run larger pipelines.
> Document our stance on local runners
> ------------------------------------
>
> Key: BEAM-13989
> URL: https://issues.apache.org/jira/browse/BEAM-13989
> Project: Beam
> Issue Type: Improvement
> Components: website
> Reporter: Kyle Weaver
> Priority: P2
>
> We often get questions like "is the direct runner suitable for production"?
> and the answer is usually no. We should document our recommendations for
> small (one machine) pipelines somewhere on the website.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
