@quba There’s an entry in the postgresql changelog that near-perfectly matches the situation
https://www.postgresql.org/docs/release/13.0/
Allow hash aggregation to use disk storage for large aggregation result sets (Jeff Davis)Previously, hash aggregation was avoided if it was expected to use more than work_mem memory. Now, a hash aggregation plan can be chosen despite that. The hash table will be spilled to disk if it exceeds
work_memtimes hash_mem_multiplier.This behavior is normally preferable to the old behavior, in which once hash aggregation had been chosen, the hash table would be kept in memory no matter how large it got — which could be very large if the planner had misestimated. If necessary, behavior similar to that can be obtained by increasinghash_mem_multiplier.