QueryBuilder: gracefully handle errors with non-matching projection sizes

[yakutovicha]

If I run a query

from datetime import timedelta
from aiida.common import timezone
now = timezone.now()

qb = QueryBuilder()
qb.append(
        CalcJobNode,
        filters={'ctime':{'>': now - timedelta(days=4)}},
        project={'id':{'func':'count'}}
    )
qb.append(
        Dict,
        filters={'attributes.energy':{'>':-5.0}},
        edge_filters={'label':{'like':'output_%'}},
        edge_project='label'
     )

I get the following error:

ProgrammingError: (psycopg2.errors.GroupingError) column "db_dblink_1.label" must appear in the GROUP BY clause or be used in an aggregate function
LINE 1: ...HOLD FOR SELECT count(db_dbnode_1.id) AS count_1, db_dblink_...
                                                             ^

[SQL: SELECT count(db_dbnode_1.id) AS count_1, db_dblink_1.label AS db_dblink_1_label 
FROM db_dbnode AS db_dbnode_1 JOIN db_dblink AS db_dblink_1 ON db_dblink_1.input_id = db_dbnode_1.id JOIN db_dbnode AS db_dbnode_2 ON db_dblink_1.output_id = db_dbnode_2.id 
WHERE CAST(db_dbnode_1.node_type AS VARCHAR) LIKE %(param_1)s AND db_dbnode_1.ctime > %(ctime_1)s AND CAST(db_dbnode_2.node_type AS VARCHAR) LIKE %(param_2)s AND CASE WHEN (jsonb_typeof(db_dbnode_2.attributes #> %(attributes_1)s) = %(param_3)s) THEN CAST((db_dbnode_2.attributes #>> %(attributes_1)s) AS FLOAT) > %(param_4)s ELSE %(param_5)s END AND CAST(db_dblink_1.label AS VARCHAR) LIKE %(param_6)s]
[parameters: {'param_1': 'process.calculation.calcjob.%', 'ctime_1': datetime.datetime(2019, 7, 26, 16, 33, 27, 209862, tzinfo=<UTC>), 'param_2': 'data.dict.%', 'attributes_1': '{energy}', 'param_3': 'number', 'param_4': -5.0, 'param_5': False, 'param_6': 'output_%'}]
(Background on this error at: http://sqlalche.me/e/f405)

Moreover, after such an error QueryBuilder does not work any longer (all further attempts to run a query result in the same error).

The problem comes from different size of projections. In the first case the projection is [[29]], while in the second case it is [['output_parameters'], ..., ['output_parameters']] that have different size.

It would be nice to handle properly such an error and to keep QueryBuilder operational.

[CasperWA]

@ramirezfranciscof this may be of interest to you to know, relating to the queries for determining Nodes to delete.

[ramirezfranciscof]

@CasperWA thanks for the heads up! I just checked and I only project once in each query, so I think I should be safe in regards to this issue, correct?