Queries with relative times seem to be off

[pauldix]

From this thread: https://groups.google.com/forum/?#!msg/influxdb/WLPOANOb0B8/HWC239vQbJUJ

Looks like queries with WHERE time > now() - 1h or something like that are doing weird things. Maybe a timezone thing? Something else? Please to investigate.

[jnutzmann]

My apologies, but it looks like this is a bug that was mistakenly introduced in PR #2016. That said, there were issues with how the time for these queries was computed before (and the bucket problem).

When I added the code to NextInterval in tx.go, I assumed that interval was only greater than zero if there was truly an interval (group by in use). This is not the case, as Execute in engine.go sets it to the span between min time and max time. Therefore, NextInterval incorrectly tries to normalize the start time to a natural boundary when it shouldn't.

I think the solution is getting rid of m.interval = m.TMax - m.TMin on line 97 of engine.go, adding a check for zero on line 132 like this:

if m.interval > 0 {
    startTimeBucket = m.TMin / m.interval * m.interval
} else {
    startTimeBucket = m.TMin
}

then, in tx.go we can add an else case in NextInterval like this:

if interval > 0 {
    // Make sure the bottom of the interval lands on a natural boundary.
    intervalBottom = intervalBottom / interval * interval
    l.tmax = intervalBottom + interval - 1
} else {
    l.tmax = l.job.TMax
}

I have a local version running now and am working on testing it.

[jnutzmann]

This solution causes some group by queries to fail when a minimum time is not specified. I am tied up in meetings for the next few hours, but can get back to it after that.

[pauldix]

This is fixed in RC16.

[nemesifier]

tested! 👍