Strange Behavior with Athena UPDATE of an Iceberg table

Question

We are getting strange behavior updating an iceberg table with a mix of 'string' and 'int' data types:

UPDATE table1 set int_col1=0 where key_col = 'key';  -- works fine

UPDATE table1 set int_col2=0 where key_col = 'key'; -- works fine

UPDATE table1 set int_col1=9, int_col2=10 where key_col = 'key'; -- works fine

**UPDATE table1 set in_col1=0, int_col=0 where key_col = 'key'; -- fails with error GENERIC_INTERNAL_ERROR: symbolCounter 2 should be columnValueAndRowIdChannels.size() %s**

It seems there is something wonky with the parser when there is more than one SET to 0.  Anyone else seeing this, or is there a workaround or problem with our approach?  If we do a DELETE and INSERT it works fine, but UPDATE is sure handy...

Answer

As you might already be aware that Athena uses Presto in the back end. It seems that there was a known issue with PrestoSQL (Trino) where the query would fail if two columns are being updated with the same value. Please refer the below Github link for more details:
https://github.com/trinodb/trino/commit/4e1cc58c2a73129a5d590c01e1e0040755e58248

But this was fixed in the newer versions of the PrestoSQL and since Athena version 2 still uses an older version of Presto which is causing the above problem. I have tested with Athena version 2 and faced the same issue. On testing with Athena version 3 this issue seem to have been fixed as the Update query is working fine when two columns are being updated with same value, such as 0 in your case.

Therefore, to overcome the issue I would request you to upgrade to Athena version to 3 and then test. Please refer the below document for Athena version 3:
https://docs.aws.amazon.com/athena/latest/ug/engine-versions-reference-0003.html

Strange Behavior with Athena UPDATE of an Iceberg table

相關內容