[Query Fuzz] Fill(none) in query hangs

[e-dard]

This issue was found by fuzzing the database with a query generator tool. It may be possible to simplify the query to help identify the root cause.

Dataset

$ curl https://s3.amazonaws.com/noaa.water-database/NOAA_data.txt -o NOAA_data.txt
$ influx -import -path=NOAA_data.txt

Query

USE NOAA_water_database

SELECT  "pH" / "degrees", "degrees" FROM  "average_temperature" GROUP BY "pq" FILL(none)

Result

The query hangs. Heap remains very low, no CPU activity.

See attached SIGQUIT on the server.

If you remove the FILL(none) the query returns fine. If you change the query to:

SELECT  "pH" / "degrees", "degrees" FROM  "average_temperature" GROUP BY "pq" FILL(null)

it also returns fine.

1488220500085551134-freeze.txt

[jsternberg]

The issue seems to be because of a mismatch between auxiliary iterators.

All auxiliary iterators are read from the same source and have to be read with each other. pH seems to be an empty value so it returns a null point for everything. That causes FILL(null) to want to skip everything since it produces no values. But the degrees part of the query does want to return a value. Since the expression iterator tries to skip past all of the points, but the second field has not read the first point yet, that causes the query to hang on a goroutine (why there is no cpu usage).