[0.9.3] First or last value of GROUP BY time(x) is often null

[zimbatm]

For some reason the GROUP BY time(x) queries always return the first value as null. (even if data is available at that time). It also does the same for larger interval.

# Setup
curl -G http://localhost:8086/query --data-urlencode "q=CREATE DATABASE mydb"
curl -G http://localhost:8086/query --data-urlencode "q=CREATE RETENTION POLICY myrp ON mydb DURATION 365d REPLICATION 1 DEFAULT"

# Write lots of data
curl -X POST "http://localhost:8086/write?db=mydb" --data-binary "cpu,region=useast,host=server_1,service=redis value=61"

# The query
curl -G http://localhost:8086/query --data-urlencode "db=mydb" --data-urlencode "q=SELECT MEAN(value) FROM cpu WHERE time > now() - 5s GROUP BY time(5s)"

Example response:

{"results":[{"series":[{"name":"cpu","columns":["time","mean"],"values":[["2015-09-01T14:05:10Z",null],["2015-09-01T14:05:15Z",61]]}]}]}

[beckettsean]

There's definitely something amiss with the first bucket in the returns, but not always:

> select count(value) from cpu_idle where time > now() - 60s group by time(10s)
name: cpu_idle
--------------
time            count
2015-09-01T20:01:40Z    9
2015-09-01T20:01:50Z    9
2015-09-01T20:02:00Z    18
2015-09-01T20:02:10Z    9
2015-09-01T20:02:20Z    18
2015-09-01T20:02:30Z    9
2015-09-01T20:02:40Z    9

Notice the value of 9 for '2015-09-01T20:01:40Z'.

Query again a few seconds later. Everything else is the same but the "1:40" bucket is null.

> select count(value) from cpu_idle where time > now() - 60s group by time(10s)
name: cpu_idle
--------------
time            count
2015-09-01T20:01:40Z    
2015-09-01T20:01:50Z    9
2015-09-01T20:02:00Z    18
2015-09-01T20:02:10Z    9
2015-09-01T20:02:20Z    18
2015-09-01T20:02:30Z    9
2015-09-01T20:02:40Z    9

Query again a few seconds later, and now the "1:50" bucket is first and is null, instead of 18.

> select count(value) from cpu_idle where time > now() - 60s group by time(10s)
name: cpu_idle
--------------
time            count
2015-09-01T20:01:50Z    
2015-09-01T20:02:00Z    18
2015-09-01T20:02:10Z    9
2015-09-01T20:02:20Z    18
2015-09-01T20:02:30Z    9
2015-09-01T20:02:40Z    18
2015-09-01T20:02:50Z    

[beckettsean]

It actually can affect the first and/or last bucket:

> select count(value) from cpu_idle where time > now() - 60s group by time(10s)
name: cpu_idle
--------------
time            count
2015-09-01T20:05:10Z    9
2015-09-01T20:05:20Z    9
2015-09-01T20:05:30Z    18
2015-09-01T20:05:40Z    9
2015-09-01T20:05:50Z    18
2015-09-01T20:06:00Z    9
2015-09-01T20:06:10Z    9

> select count(value) from cpu_idle where time > now() - 60s group by time(10s)
name: cpu_idle
--------------
time            count
2015-09-01T20:05:20Z    9
2015-09-01T20:05:30Z    18
2015-09-01T20:05:40Z    9
2015-09-01T20:05:50Z    18
2015-09-01T20:06:00Z    9
2015-09-01T20:06:10Z    18
2015-09-01T20:06:20Z    

> select count(value) from cpu_idle where time > now() - 60s group by time(10s)
name: cpu_idle
--------------
time            count
2015-09-01T20:05:20Z    
2015-09-01T20:05:30Z    18
2015-09-01T20:05:40Z    9
2015-09-01T20:05:50Z    18
2015-09-01T20:06:00Z    9
2015-09-01T20:06:10Z    18
2015-09-01T20:06:20Z    9

[beckettsean]

This doesn't seem to happen for queries with explicit lower and upper time bounds.

[zimbatm]

I can confirm I also saw the null timestamp in the front but I wasn't sure if maybe the data hadn't synched between the various nodes yet.