Why is fill(0) creating entries and breaking the group? - influxdb

Question

On a project where we use influx 1.7 we have a query that behaves strangely. Basically what we're trying to do is
average the data to align the time periods
take the max out of these time periods
sum these max values
Here is an example that reproduces the behavior.
Insert some example data
> INSERT test,source=source1 value=10 1643202000000000000
> INSERT test,source=source1 value=30 1643203800000000000
> INSERT test,source=source2 value=20 1643202000000000000
> INSERT test,source=source2 value=60 1643203800000000000
> INSERT test,source=source2 value=20 1643202900000000000
> INSERT test,source=source2 value=50 1643204700000000000
Change precision for readability and check the data
> precision rfc3339
> select * from test
name: test
time source value
---- ------ -----
2022-01-26T13:00:00Z source1 10
2022-01-26T13:00:00Z source2 20
2022-01-26T13:15:00Z source2 20
2022-01-26T13:30:00Z source1 30
2022-01-26T13:30:00Z source2 60
2022-01-26T13:45:00Z source2 50
Based on what I described, I would expect the final sum to equal max(10, mean(20,20)) + max(30, mean(60,50) = max(10,20) + max(30,55) = 20 + 55 = 75
If we select the mean value using fill(0) and fill(none) that's when things start to get weird, some values are added for no apparent reason
> select mean(value) from test group by time(30m), source fill(0)
name: test
tags: source=source1
time mean
---- ----
2022-01-26T13:00:00Z 10
2022-01-26T13:30:00Z 30
2022-01-26T14:00:00Z 0
2022-01-26T14:30:00Z 0
2022-01-26T15:00:00Z 0
name: test
tags: source=source2
time mean
---- ----
2022-01-26T13:00:00Z 20
2022-01-26T13:30:00Z 55
2022-01-26T14:00:00Z 0
2022-01-26T14:30:00Z 0
2022-01-26T15:00:00Z 0
> select mean(value) from test group by time(30m), source fill(none)
name: test
tags: source=source1
time mean
---- ----
2022-01-26T13:00:00Z 10
2022-01-26T13:30:00Z 30
name: test
tags: source=source2
time mean
---- ----
2022-01-26T13:00:00Z 20
2022-01-26T13:30:00Z 55
But things start to get crazy when we try to select the max value for each time period
> select max(value) from (select mean(value) as value from test group by time(30m), source fill(0)) group by time(30m)
name: test
time max
---- ---
2022-01-26T13:00:00Z 10
2022-01-26T13:30:00Z 0
2022-01-26T14:00:00Z 0
2022-01-26T14:30:00Z 0
2022-01-26T15:00:00Z 0
2022-01-26T13:00:00Z 20
2022-01-26T13:30:00Z 0
2022-01-26T14:00:00Z 0
2022-01-26T14:30:00Z 0
2022-01-26T15:00:00Z 0
2022-01-26T13:30:00Z 30
2022-01-26T14:00:00Z 0
2022-01-26T14:30:00Z 0
2022-01-26T15:00:00Z 0
2022-01-26T13:30:00Z 55
2022-01-26T14:00:00Z 0
2022-01-26T14:30:00Z 0
2022-01-26T15:00:00Z 0
> select max(value) from (select mean(value) as value from test group by time(30m), source fill(none)) group by time(30m)
name: test
time max
---- ---
2022-01-26T13:00:00Z 20
2022-01-26T13:30:00Z 55
2022-01-26T14:00:00Z
2022-01-26T14:30:00Z
2022-01-26T15:00:00Z
Of course the final result differs
> select sum(value) from (select max(value) as value from (select mean(value) as value from test group by time(30m), source fill(0)) group by time(30m))
name: test
time sum
---- ---
1970-01-01T00:00:00Z 115
> select sum(value) from (select max(value) as value from (select mean(value) as value from test group by time(30m), source fill(none)) group by time(30m))
name: test
time sum
---- ---
1970-01-01T00:00:00Z 75
The query with fill(0) seems to have broken the group as all values are getting selected.
Am fill(0) broken or am I missing/misusing something ? Could anyone justify this behavior ?