Uh oh!
There was an error while loading. Please reload this page.
[SPARK-16397][SQL] make CatalogTable more general and less hive specific - #14071
[SPARK-16397][SQL] make CatalogTable more general and less hive specific#14071cloud-fan wants to merge 3 commits into
Conversation
SparkQA
commented
Jul 6, 2016
Test build #61858 has finished for PR 14071 at commit
|
There was a problem hiding this comment.
storageProps += "inputFormat" -> string(c.inFmt)? string(c.inFmt) throws an NPE if it is null; so option is redundant...
cloud-fan
commented
Jul 7, 2016
SparkQA
commented
Jul 7, 2016
Test build #61914 has finished for PR 14071 at commit
|
There was a problem hiding this comment.
Any reason why this locationUri is set to None? It sounds like the original value is Some(relation.location.paths.map(_.toUri.toString).head
SparkQA
commented
Jul 8, 2016
Test build #61951 has finished for PR 14071 at commit
|
SparkQA
commented
Jul 8, 2016
Test build #61959 has finished for PR 14071 at commit
|
| inputFormat: Option[String], | ||
| outputFormat: Option[String], | ||
| serde: Option[String], | ||
| compressed: Boolean, |
There was a problem hiding this comment.
We never persist this flag into external catalog.
SparkQA
commented
Jul 13, 2016
Test build #62247 has finished for PR 14071 at commit
|
SparkQA
commented
Jul 13, 2016
Test build #62249 has finished for PR 14071 at commit
|
cloud-fan
commented
Jul 13, 2016
I'm closing it as it turns out that we still need these hive specific stuff in CatalogStorageFormat. Hiding them in properties seems don't have much benefit, what we need is just adding a 'provider' field in CatalogTable, which is done in #14155 |
What changes were proposed in this pull request?
CatalogTableis an internal interface to represent table metadata in Spark SQL, and should be able to express both hive serde table and data source table. Due to some hive hacks, data source table only use table properties to store its information, and theCatalogStorageFormatis very hive specific as it's only used by hive serde table currently.However this is not a good design, the
CatalogTableis not general enough to express data source table if we remove hive hacks in the future.This PR is trying to make
CatalogStorageFormatgeneral enough to handle all types of tables in Spark SQL.How was this patch tested?
existing tests.