Groupby issue on multiple join using Grafana's (TimescaleDB) SQL plugin - sql

Question

I'm using Grafana's SQL plugin to query a TimescaleDB database.
The DB stores weather information as
| timestamp | location_id | data_type_id | value |
where location_id and data_type_id are foreign keys to table locations describing the locations and weather_data_types defining the measurement types (temperature, relative_humidity,...).
I'd like to query data on a time range, grouped by location and type.
I manage to group by one of them, but not both.
This works and groups by location:
SELECT
$__timeGroupAlias("timestamp", $__interval),
avg(value),
locations.name
FROM weather_data
JOIN locations ON weather_data.location_id = locations.id
GROUP BY 1, locations.name
ORDER BY 1
This works and groups by type:
SELECT
$__timeGroupAlias("timestamp", $__interval),
avg(value),
weather_data_types.name
FROM weather_data
JOIN weather_data_types ON weather_data.type_id = weather_data_types.id
GROUP BY 1, weather_data_types.name
ORDER BY 1
This does not work:
SELECT
$__timeGroupAlias("timestamp", $__interval),
avg(value),
locations.name,
weather_data_types.name
FROM weather_data
JOIN locations ON weather_data.location_id = locations.id
JOIN weather_data_types ON weather_data.type_id = weather_data_types.id
GROUP BY 1, locations.name, weather_data_types.name
ORDER BY 1
More specifically, I get the following error
Value column must have numeric datatype, column: name type: string value: relative_humidity
It seems the third groupby (silently) doesn't happen and weather_data_types.name is returned, which Grafana complains about because it can't plot strings.
Changing this to return the (integer) id instead removes the error message
SELECT
$__timeGroupAlias("timestamp", $__interval),
avg(value),
locations.name,
weather_data_types.id
FROM weather_data
JOIN locations ON weather_data.location_id = locations.id
JOIN weather_data_types ON weather_data.type_id = weather_data_types.id
GROUP BY 1, locations.name, weather_data_types.id
ORDER BY 1
but two series are plotted: avg and id, which shows the groupby type is not applied.
Is there anything wrong in my query? Is it an issue with the Grafana plugin?
I don't think it matters, but here's the model, defined with SQLAlchemy and hopefully self-explanatory.
class Location(Base):
__tablename__ = "locations"
id = sqla.Column(sqla.Integer, primary_key=True)
name = sqla.Column(sqla.String(80), unique=True, nullable=False)
country = sqla.Column(sqla.String(80), nullable=False)
latitude = sqla.Column(sqla.Float(), nullable=False)
longitude = sqla.Column(sqla.Float(), nullable=False)
class WeatherDataTypes(Base):
__tablename__ = "weather_data_types"
id = sqla.Column(sqla.Integer, primary_key=True)
name = sqla.Column(sqla.String(80), unique=True, nullable=False)
description = sqla.Column(sqla.String(500), nullable=False)
unit = sqla.Column(sqla.String(20), nullable=False)
min_value = sqla.Column(sqla.Float)
max_value = sqla.Column(sqla.Float)
class WeatherData(Base):
__tablename__ = "weather_data"
timestamp = sqla.Column(sqla.DateTime(timezone=True), primary_key=True)
location_id = sqla.Column(
sqla.Integer,
sqla.ForeignKey('locations.id'),
nullable=False,
primary_key=True
)
location = sqla.orm.relationship('Location')
type_id = sqla.Column(
sqla.Integer,
sqla.ForeignKey('weather_data_types.id'),
nullable=False,
primary_key=True
)
type = sqla.orm.relationship('WeatherDataTypes')
value = sqla.Column(sqla.Float)

Answer 1

Sending requests directly to postgresql helped me understand what is happening.
Apparently, when the query returns a column of values and a column of strings, the Grafana plugin assumes the values are to be plotted and the string column is meant to be used as labels for the plots.
I thought the plugin used the groupby to sort of extract the column to make it label information, but this magic doesn't work with two string columns as the plugin won't concatenate the values itself. Therefore the plugin complains about the second string column not being numbers which is kind of misleading because it would not complain about the first string column.
I could get it to work by concatenating the values I use for the groupby into a single column:
SELECT
time_bucket('21600s',"timestamp") AS "time",
avg(value),
CONCAT(locations.name, ' ', weather_data_types.name) AS "name"
FROM weather_data
JOIN locations ON weather_data.location_id = locations.id
JOIN weather_data_types ON weather_data.type_id = weather_data_types.id
GROUP BY 1, locations.name, weather_data_types.name
ORDER BY 1
This returns
time | avg | name
------------------------+--------------------+---------------------------
which is correctly interpreted by the plugin.