HIVE-29372: Iceberg: [V3] Fix inconsistencies when vectorizartion is enabled in case of variant shredding.#6245
Conversation
| } | ||
|
|
||
| public static boolean shouldUseVariantShredding(Properties tableProps, Schema tableSchema) { | ||
| return isVariantShreddingEnabled(tableProps) && hasVariantFields(tableSchema); |
There was a problem hiding this comment.
maybe use Maps.fromProperties to avoid code duplication. or introduce a generic signature UnaryOperator<String> tableProps -> tableProps::getProperty
…enabled in case of variant shredding.
| (1, parse_json('{"a": 1}')), | ||
| (2, parse_json('{"b": 2}')); | ||
|
|
||
| SELECT |
There was a problem hiding this comment.
does it cover some specific test-case? with qtest you can't test the physical layout anyways.
There was a problem hiding this comment.
earlier it was giving wrong results:
NULL 2
NULL NULL
now it is correct, I added the scenario that I tried in #6152 (comment)
I didn't put like all cases to show output is same, just it is correct, the previous original case was throwing exception, this was giving wrong result.
There was a problem hiding this comment.
wonder if we same issue could be reproduced on existing table tbl_shredded_variant by inserting (1, parse_json('{"name": "John", "age": 30, "active": true}')) with complete schema last
SELECT
variant_get(data, '$.name') as name;
variant_get(data, '$.age', 'int') as age;
FROM tbl_shredded_variant
There was a problem hiding this comment.
No Denys, it throws the same MALFORMED_VARIANT exception like now, I tried pulling the full one below, added a new record but any query is giving the same exception as in the previous thread.
There was a problem hiding this comment.
IDK, works fine for me: 1062dc5#diff-779297369ca87b0f30855401004c6819e0b64fb82167861f08dcba08e2ae3859
Execution mode: vectorized
1 NULL
NULL 2
|
…enabled in case of variant shredding. (apache#6245)
…enabled in case of variant shredding. (apache#6245)
3 participants
What changes were proposed in this pull request?
As of now Vectorization isn't supported for Shredded Variants, so we fallback to non-vectorized mode in that case
Why are the changes needed?
#6152 (comment)
Does this PR introduce any user-facing change?
Vectorization enabled/disabled doesn't bother.
How was this patch tested?
UT