[
https://issues.apache.org/jira/browse/HUDI-1371?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17243574#comment-17243574
]
Vinoth Chandar edited comment on HUDI-1371 at 12/3/20, 11:24 PM:
-----------------------------------------------------------------
||Engine||Table Type||Listing Mechanism||
|Spark SQL on Hive |COW|a) with path filter: parallel listing (we good here) +
native vectorized reader
Open: how do we avoid Spark from listing again?
b) with convertMetastoreParquet=false; Spark will call `ipf.getSplits()` to
list of files, use Hive's parquet record reader.|
|Spark SQL on Hive |MOR| with convertMetastoreParquet=false; Spark will call
`ipf.getSplits()` to list of files, use Hudi's record reader.|
|Spark Datasource|COW|We wrap Parquet Data source, which does parallel listing
+ use path filter to filter down more.
Open: Can we pass in a HadoopFsRelation to avoid the listing inside parquet
data source|
|Spark Datasource|MOR| MergeOnReadSnapshotRelation (some feature gaps etc...)
Possible unknown issues.|
|Presto|COW| PathFilter + BackgroundHiveLoader (presto's own listing mechanism)|
|Presto|MOR| |
|Hive|COW| |
|Hive |MOR| |
was (Author: vc):
||Engine||Table Type||Listing Mechanism||
|Spark SQL on Hive |COW|a) with path filter: parallel listing (we good here) +
native vectorized reader
Open: how do we avoid Spark from listing again?
b) with convertMetastoreParquet=false; Spark will call `ipf.getSplits()` to
list of files, use Hive's parquet record reader.|
|Spark SQL on Hive |MOR| with convertMetastoreParquet=false; Spark will call
`ipf.getSplits()` to list of files, use Hudi's record reader.|
|Spark Datasource|COW|We wrap Parquet Data source, which does parallel listing
+ use path filter to filter down more.
Open: Can we pass in a HadoopFsRelation to avoid the listing inside parquet
data source|
|Spark Datasource|MOR| MergeOnReadSnapshotRelation|
|Presto|COW| |
|Presto|MOR| |
|Hive|COW| |
|Hive |MOR| |
> Implement Spark datasource by fetching file listing from metadata table
> -----------------------------------------------------------------------
>
> Key: HUDI-1371
> URL: https://issues.apache.org/jira/browse/HUDI-1371
> Project: Apache Hudi
> Issue Type: Sub-task
> Components: Spark Integration
> Reporter: Vinoth Chandar
> Assignee: Udit Mehrotra
> Priority: Blocker
> Fix For: 0.7.0
>
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
