Difference between actual sample vs filled - influxdb

Question

I expected a subquery having GROUP BY time intervals and fill(previous) behaves identically to the resultant dataset. But a query against the subquery returns unexpected results.
Environments
InfluxDB v1.6.3
What Happened
Here is a sample dataset:
# DDL
CREATE DATABASE myapp
# DML
# CONTEXT-DATABASE: myapp
reqcnt,endpoint=a value=1 1563361000000000000
reqcnt,endpoint=b value=1 1563361000000000000
reqcnt,endpoint=a value=2 1563361005000000000
reqcnt,endpoint=b value=2 1563361005000000000
reqcnt,endpoint=a value=3 1563361010000000000
reqcnt,endpoint=b value=3 1563361010000000000
You can import the above data by executing:
$ influx -import /path/to/dataset -precision ns
The data is simple: three values per endpoint, measured with every 5 seconds. But because I want a dataset having per-second values, I made a subquery filling missing values with previous ones:
> SELECT LAST(value) AS lv FROM reqcnt WHERE time >= 1563361000000000000 AND time <= 1563361010000000000 GROUP BY time(1s), endpoint fill(previous)
name: reqcnt
tags: endpoint=a
time lv
---- --
1563361000000000000 1
1563361001000000000 1
1563361002000000000 1
1563361003000000000 1
1563361004000000000 1
1563361005000000000 2
1563361006000000000 2
1563361007000000000 2
1563361008000000000 2
1563361009000000000 2
1563361010000000000 3
name: reqcnt
tags: endpoint=b
time lv
---- --
1563361000000000000 1
1563361001000000000 1
1563361002000000000 1
1563361003000000000 1
1563361004000000000 1
1563361005000000000 2
1563361006000000000 2
1563361007000000000 2
1563361008000000000 2
1563361009000000000 2
1563361010000000000 3
So far so good. Now I'd like to sum across endpoints:
> SELECT SUM(value) FROM (SELECT LAST(value) AS value FROM reqcnt WHERE time >= 1563361000000000000 AND time <= 1563361010000000000 GROUP BY time(1s), endpoint fill(previous)) WHERE time >= 1563361000000000000 AND time <= 1563361010000000000 GROUP BY time(1s)
name: reqcnt
time sum
---- ---
1563361000000000000 1
1563361001000000000 1
1563361002000000000 1
1563361003000000000 1
1563361004000000000 1
1563361005000000000 1
1563361006000000000 1
1563361007000000000 1
1563361008000000000 1
1563361009000000000 1
1563361010000000000 1
1563361000000000000 1
1563361001000000000 1
1563361002000000000 1
1563361003000000000 1
1563361004000000000 1
1563361005000000000 1
1563361006000000000 1
1563361007000000000 1
1563361008000000000 1
1563361009000000000 1
1563361010000000000 1
1563361005000000000 2
1563361006000000000 2
1563361007000000000 2
1563361008000000000 2
1563361009000000000 2
1563361010000000000 2
1563361005000000000 2
1563361006000000000 2
1563361007000000000 2
1563361008000000000 2
1563361009000000000 2
1563361010000000000 8
Why does the result contains multiple values for a same timestamp? And why 34 values when I added 11 to 11?
What I expected
> SELECT SUM(value) FROM (SELECT LAST(value) AS value FROM reqcnt WHERE time >= 1563361000000000000 AND time <= 1563361010000000000 GROUP BY time(1s), endpoint fill(previous)) WHERE time >= 1563361000000000000 AND time <= 1563361010000000000 GROUP BY time(1s)
name: reqcnt
time sum
---- ---
1563361000000000000 2
1563361001000000000 2
1563361002000000000 2
1563361003000000000 2
1563361004000000000 2
1563361005000000000 2
1563361006000000000 4
1563361007000000000 4
1563361008000000000 4
1563361009000000000 4
1563361010000000000 6
Creating a per-second dataset and query it returns the expected result:
# DDL
CREATE DATABASE myapp
# DML
# CONTEXT-DATABASE: myapp
reqcnt2,endpoint=a value=1 1563361000000000000
reqcnt2,endpoint=b value=1 1563361000000000000
reqcnt2,endpoint=a value=1 1563361001000000000
reqcnt2,endpoint=b value=1 1563361001000000000
reqcnt2,endpoint=a value=1 1563361002000000000
reqcnt2,endpoint=b value=1 1563361002000000000
reqcnt2,endpoint=a value=1 1563361003000000000
reqcnt2,endpoint=b value=1 1563361003000000000
reqcnt2,endpoint=a value=1 1563361004000000000
reqcnt2,endpoint=b value=1 1563361004000000000
reqcnt2,endpoint=a value=2 1563361005000000000
reqcnt2,endpoint=a value=2 1563361005000000000
reqcnt2,endpoint=b value=2 1563361006000000000
reqcnt2,endpoint=a value=2 1563361006000000000
reqcnt2,endpoint=b value=2 1563361007000000000
reqcnt2,endpoint=a value=2 1563361007000000000
reqcnt2,endpoint=b value=2 1563361008000000000
reqcnt2,endpoint=a value=2 1563361008000000000
reqcnt2,endpoint=b value=2 1563361009000000000
reqcnt2,endpoint=a value=2 1563361009000000000
reqcnt2,endpoint=a value=3 1563361010000000000
reqcnt2,endpoint=b value=3 1563361010000000000
Importing the above and query the dataset shows the expected results:
> SELECT SUM(value) FROM reqcnt2 WHERE time >= 1563361000000000000 AND time <= 1563361010000000000 GROUP BY time(1s)
name: reqcnt2
time sum
---- ---
1563361000000000000 2
1563361001000000000 2
1563361002000000000 2
1563361003000000000 2
1563361004000000000 2
1563361005000000000 2
1563361006000000000 4
1563361007000000000 4
1563361008000000000 4
1563361009000000000 4
1563361010000000000 6
So what is the difference? a filled dataset should not be queried as is an actual dataset? Is there a proper way to query such subquery?