How memory distribution happens in vertica plan steps? - query-optimization

Question

I have a situation where a certain vertica query has GROUP_BY_SPILLED in its plan (doing a GROUP BY HASH).
I don't have an option to change the projection order by (to avoid group by hash), so I want to increase the amount of memory this step receives.
Let's say my query received budget of 5GB.
How are the 5GB distributed across the query steps?
I couldn't find any documentation for it.

Answer 1

Take a look at EXECUTION_ENGINE_PROFILES table.
Run your query with PROFILE:
PROFILE SELECT .. FROM .. GROUP BY ..
Grab the statement_id from the output and use it to investigate EXECUTION_ENGINE_PROFILES.
Given a simple query that only does GROUP BY, run:
SELECT path_id,
operator_name,
counter_name,
sum(counter_value)
FROM v_monitor.execution_engine_profiles
WHERE transaction_id=<STATEMENT_ID>
AND counter_name ilike 'current memory capacity (bytes)'
GROUP BY 1,2,3
ORDER BY path_id;
path_id | operator_name | counter_name | sum
---------+---------------+---------------------------------+------------
-1 | NewEENode | current memory capacity (bytes) | 0
-1 | Root | current memory capacity (bytes) | 0
1 | ExprEval | current memory capacity (bytes) | 0
1 | GroupByHash | current memory capacity (bytes) | 1406074880
1 | GroupByPipe | current memory capacity (bytes) | 0
1 | ParallelUnion | current memory capacity (bytes) | 0
1 | StorageUnion | current memory capacity (bytes) | 0
2 | ExprEval | current memory capacity (bytes) | 0
2 | Scan | current memory capacity (bytes) | 0
More info about profiling in Vertica here
I don't think you can modify or improve these counters values, except increasing the query budget using RESOURCE POOLS