Select mean and percentile within one query and group by gives non-distinct lines

[DAfanasyev]

System info:

• Ubuntu 16.04
• InfluxDB 1.2
• Grafana 4.1.2 (commit: v4.1.2)

I'm using JMeter plugin to push metrics into InfluxDB (https://github.com/NovaTecConsulting/JMeter-InfluxDB-Writer)

Here is how plugin stores information (some measurements are skipped):

> show field keys
name: requestsRaw
fieldKey     fieldType
--------     ---------
errorCount   integer
responseTime integer
...

> show tag keys
name: requestsRaw
tagKey
------
requestName
...

Then I'm trying to create an overview table with mean and 90th percentile for each "requestName" in Grafana using following InfluxQL request:

select mean("responseTime"), percentile("responseTime", 90) from "requestsRaw" group by "requestName"

Actual behavior:

> select mean("responseTime"), percentile("responseTime", 90) from "requestsRaw" group by "requestName"
name: requestsRaw
tags: requestName=Main
time mean              percentile
---- ----              ----------
0    731.0807017543859 

name: requestsRaw
tags: requestName=Root
time mean               percentile
---- ----               ----------
0    51.316569200779725 61

name: requestsRaw
tags: requestName=Threads
time mean               percentile
---- ----               ----------
0    46.776649746192895 53

name: requestsRaw
tags: requestName=Main
time mean percentile
---- ---- ----------
0         749

Expected behavior:

> select mean("responseTime"), percentile("responseTime", 90) from "requestsRaw" group by "requestName"
name: requestsRaw
tags: requestName=Main
time mean              percentile
---- ----              ----------
0    731.0807017543859 749

name: requestsRaw
tags: requestName=Root
time mean               percentile
---- ----               ----------
0    51.316569200779725 61

name: requestsRaw
tags: requestName=Threads
time mean               percentile
---- ----               ----------
0    46.776649746192895 53

I've also noticed that group by doesn't return distinct result only when percentile function is used (at least in my case):

> select mean("responseTime"), max("responseTime") from "requestsRaw" group by "requestName"
name: requestsRaw
tags: requestName=Main
time mean              max
---- ----              ---
0    731.0807017543859 1016

name: requestsRaw
tags: requestName=Root
time mean               max
---- ----               ---
0    51.316569200779725 310

name: requestsRaw
tags: requestName=Threads
time mean               max
---- ----               ---
0    46.776649746192895 324

[DAfanasyev]

Might be similar to #8072.

I've also just noticed that it works with percentile("responseTime", 99).

[jsternberg]

Closing this in favor of #8167.

[luhn]

I've been experiencing this issue too. After upgrading to v1.2.3, which fixes #8167, this issue remains. It seems this is a separate issue than #8167 and is yet unresolved.

[jsternberg]

I'll take a look at this tomorrow then.

[jsternberg]

@luhn do you have a reproducer I could look at? @DAfanasyev does this still affect you?

[luhn]

The query I'm seeing it on is something like this:

SELECT percentile("duration", 95), percentile("duration", 99) FROM "measurement" WHERE time > now() - 8h GROUP BY "context", "view_name"

I'll see if I can make something reproducible.

[luhn]

Okay, here's a data file and a sample query, along with the output I'm seeing.
https://gist.github.com/luhn/dbc791f6164cc7c12737502a6860be0d/archive/1ed4273163a55c05ead3d3cc9a7e6e60e4fed011.zip

[jsternberg]

Hm... curious. The v1.2.3 tag has it not working but it's working on master now.

I'll see if I can find the commit that fixes it.

[jsternberg]

Looks like it's fixed as of be3bce5. Unfortunately, I do not envision me being able to backport that easily. Is it alright if this fix comes as part of 1.3? I believe we're planning on formally releasing v1.2.3 sometime this week and it's probably unlikely we'll get a v1.2.4 before v1.3.0 is released.

[luhn]

I'm eager to upgrade to 1.3, so personally I have no problem with not backporting it.

I'll probably try out the nightly build in the meantime. Any idea when 1.3 will be released?

[jsternberg]

I would expect 1.3 to land sometime in June. No promises, but that's the current estimated timeframe. Keep an eye out for the release candidate!