[
https://issues.apache.org/jira/browse/HUDI-109?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Raymond Xu updated HUDI-109:
----------------------------
Component/s: (was: Spark Integration)
> Ability to query older snapshots of data
> -----------------------------------------
>
> Key: HUDI-109
> URL: https://issues.apache.org/jira/browse/HUDI-109
> Project: Apache Hudi
> Issue Type: New Feature
> Reporter: Nishith Agarwal
> Assignee: Nishith Agarwal
> Priority: Major
>
> At the moment, Hudi allows a client to configure the number of versions to
> keep for a particular dataset. Depending on the trade-off between storage
> cost vs number of versions required, one can use Hudi to keep multiple views
> of the dataset at different points in time.
> Hudi uses this information to provide incremental consumption of what changed
> between 2 time units but does not have a simple way to be able to query a
> dataset as of a particular instant in time. There are multiple use-cases
> which benefit from such a feature, such as ML training/use-cases against the
> state of a feature store at a particular instant in time or for auditing
> system that require to see previous versions of data.
> Hudi has most of the building blocks to be able to support such queries, we'd
> like to explore the feasibility & changes required to support it.
> [~semanticbeeng] something along the lines of what you brought up as well.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
