[HIVE-6166] JsonSerDe is too strict about table schema

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.12.0
• Fix Version/s: 0.13.0
• Component/s: HCatalog, Serializers/Deserializers
• Assignee: Sushanth Sowmyan
• Reporter: Sushanth Sowmyan
• Created: 08/Jan/14 14:59
• Updated: 18/Mar/15 14:45
• Resolved: 13/Jan/14 16:38

Description

JsonSerDe is too strict when it comes to schema, erroring out if it finds a subfield with a key name that does not map to an appropriate type/schema of a table, or an inner-struct schema.

Thus, if a schema specifies "s:struct<a:int,b:string>,k:int" and we pass it data that looks like the following:

{ "x" : "abc" , "s" : { "a" : 2 , "b" : "blah", "c": "woo" } }

This should still pass, and the record should be read as if it were

{ "s" : { "a" : 2 , "b" : "blah"}, k :  null }

This will allow the JsonSerDe to be used with a wider set of data where the data does not map too finely to the declared table schema.

Note, we are still strict about a couple of things:

a) If there is a declared schema column, then the type cannot vary, that is still considered an error. i.e., if the hive table schema says k1 is a boolean, it cannot magically change into an int or a struct, say, for eg.
b) The JsonSerDe still attempts to map hive internal column names - i.e. if the data contains a column named "_col2", then, if "_col2" is not declared directly in the schema, it will map to column position 2 in that schema/subschema, rather than ignoring the field. This is so that tables created with CTAS will still work.

Issue Links

• is duplicated by: HIVE-9962 JsonSerDe does not support reader schema different from data schema