multiply used labels in filters

[Henri Dubois-Ferriere]

When I use a label multiple times in a filter, they are all applied (as though they were implicitly ANDed together).

In other words, if I ask for:

   metric{a=~x, a!=y}

I get all the metrics where a=~x AND a!=y.

So my question is: should this considered as part of the promql "specification"? 

In particular, should a remote read implementation handle repeated matchers in this way?

Henri

[Brian Brazil]

On 10 August 2018 at 20:02, Henri Dubois-Ferriere <hen...@gmail.com> wrote:

When I use a label multiple times in a filter, they are all applied (as though they were implicitly ANDed together).

In other words, if I ask for:

   metric{a=~x, a!=y}

I get all the metrics where a=~x AND a!=y.

So my question is: should this considered as part of the promql "specification"? 

Yes, this is part of the spec. See https://prometheus.io/docs/prometheus/latest/querying/basics/#time-series-selectors

 

In particular, should a remote read implementation handle repeated matchers in this way?

Yes, it should. 

--

Brian Brazil

www.robustperception.io

[Henri Dubois-Ferriere]

Ok thanks.

(Btw I had read that page and just re-read it again, but don't see anything that mentions multiple selectors.)

[Brian Brazil]

On 10 August 2018 at 22:28, Henri Dubois-Ferriere <hen...@gmail.com> wrote:

Ok thanks.

(Btw I had read that page and just re-read it again, but don't see anything that mentions multiple selectors.)

It's implied, but not very clear. https://github.com/prometheus/prometheus/pull/4492 out to improve it.

Brian

 

On Fri, 10 Aug 2018 at 14:24, Brian Brazil <brian.brazil@robustperception.io> wrote:

On 10 August 2018 at 20:02, Henri Dubois-Ferriere <hen...@gmail.com> wrote:

When I use a label multiple times in a filter, they are all applied (as though they were implicitly ANDed together).

In other words, if I ask for:

   metric{a=~x, a!=y}

I get all the metrics where a=~x AND a!=y.

So my question is: should this considered as part of the promql "specification"? 

Yes, this is part of the spec. See https://prometheus.io/docs/prometheus/latest/querying/basics/#time-series-selectors

 

In particular, should a remote read implementation handle repeated matchers in this way?

Yes, it should. 

--

Brian Brazil

www.robustperception.io

--

Brian Brazil

www.robustperception.io