Are the PI values for the target (looks like EDW_OFFSHORE_BASE.ISA.SVC_ACCT_NUM) highly skewed?
Even if not skewed, the redistribution by target PI to prepare for the MERGE step will require enough spool space to hold two copies of the data. The optimizer is estimating 3.5 million rows will occupy 4.7GB. If the real cardinality is 68 million then the real size of the spool file would be almost 20x that value.
↧