When passing from hive to spark with the dedicated component you have to write in the component placed before the databasename as a prefix with the table name :
SELECT * FROM pov.exercice_spark and not SELECT * FROM exercice_spark
even if the "database" is declared in the hive connection.
The drop existing table option in the hive loader doesn't work as well as the "drop table node" giving no warning at all. It could be an authorization problem BUT writing something like drop table if exists pov.exercice_spark in the "Database SQL executor" works very well.
yes, unfortunately the "Hive to Spark" node currently does not use the Hive Connector information and only executes the SQL statement. We are aware of that limitation, and it is on our list of things to address for the next release in summer. For the meantime, I would kindly recommend to use the workaround of specifying the database name.
As far as the "drop existing table" option in the hive loader is concerned, that is indeed strange. Could you provide me with a sample workflow (so I can reproduce it) and your KNIME log file? I will send you my email address for that.
I do think there is no mystery under this. It appeared to me that with certain nodes droping worked and not with the others. I do think now that this is rather due to problems with cluster unstability relative to hive metastore. Correlation is not causality ;o)
I have similar problem as fabienc above. Trying to load the data to Hive (real clodera cluster) using Hive loader node. I get an error below no matter if droptable is checked or unchecked.
Execute failed: Error while processing statement: FAILED: Execution Error, return code 1 from org.apache.hadoop.hive.ql.exec.MoveTask
Table on hive is always created however data is not loaded. I am not trying to load any big files so far. Only trying to test load of .csv file with couple of records of integer data type.