[hive] Fix no longer possible to drop the hms table issue when table …

[XiaoHongbo-Hope]

…not in fs (#4853)

Purpose Fix no longer possible to drop the hms table issue when table not in fs

Linked issue: close #4853

Tests

API and Format

Documentation

[LinMingQiang]

Repeated : #4860

[Zouxxyy]

As far as I know, before Spark calls catalog.dropTable, it actually invokes catalog.getTable fisrtly. It has already been skipped there.

case class DropTableExec(
    catalog: TableCatalog,
    ident: Identifier,
    ifExists: Boolean,
    purge: Boolean,
    invalidateCache: () => Unit) extends LeafV2CommandExec {

  override def run(): Seq[InternalRow] = {
    if (catalog.tableExists(ident)) {
      invalidateCache()
      if (purge) catalog.purgeTable(ident) else catalog.dropTable(ident)
    } else if (!ifExists) {
      throw QueryCompilationErrors.noSuchTableError(
        catalog.name() +: ident.namespace() :+ ident.name())
    }

    Seq.empty
  }

[Zouxxyy]

I'm OK with this modification. The core reason is that before dropping the table, the compute engine will call table exists -> catalog.getTable() to check if the table exists. Currently, our catalog.getTable() API only returns whether the table exists in the file system (even if it exists in HMS). Could @JingsongLi take another look at this?

[JingsongLi]

Drop table in getTable? This behavior is too strange.

[XiaoHongbo-Hope]

Ok, thanks

[JingsongLi]

I am -1 for this. You can enrich exception message. And user can just use drop table again?

[adrian-wang]

Maybe just drop hms table in dropTable even if the table is not on FS?

[adrian-wang]

Or, just recover and return a very basic dummy table when getTable, if we find the table is in hms and not on FileSystem?