A lambda that DELETEs and COPYs 8 parquet files from S3 into 8 Redshift tables (using boto3).

Each table is created from a single file, each with less than 3K rows.

7 of the queries take 2-3 seconds each, and one takes 50+ seconds.

I have exhausted my checks and not sure what this can be, this didn't work:

• Removed all fks/constraints
• VACUUM/ANALYZE the tables
• Recreated the tables
• Changed the isolation levels
• Increased the Lambda memory/CPU allocation

Notable considerations:

• When I run it directly on Redshift (not the datashare), the COPY takes 2s.
• It's definitely the COPY and not the DELETE (confirmed via logging).

Any suggestions appreciated!