Join by field (outer join) with different names

[Gilbert Moisio]

Hi all, using Grafana on a Prometheus data source and snmp_exporter, I'm searching a way to join two tables that have a common information but the name of the field is not the same in the tables.

Is there a way to do this?

Thanks

Reg

[Brian Candler]

Can you give a specific example?

The solution in general is to make the labels align, e.g. see here for the instance label.  There is also label_replace() if you can modify the labels using regexp matches.  At worst, you would create a third, static timeseries with appropriate labels to join the other two timeseries onto, similar to the approach here and here.

[Gilbert Moisio]

The example:

- From IF-MIB, getting ifHighSpeed gives an ifIndex in the returned informations

- From LLDP-MIB getting lldpLocPortDesc gives lldpLocPortNum that is the same value as ifIndex

Joining the two queries results would permit to get some more informations into the same table.

Thanks

[Brian Candler]

Try:

label_replace(lldpLocPortDesc, "ifIndex", "$1", "lldpLocPortNum", "(.+)")

If that works, then you should be able to build a query on top of this which uses (instance, ifIndex) as the join labels.

[Gilbert Moisio]

Ok, it looks like it's working as expected...

Thanks

[Kshitija Patil]

i wanted to join to queries cpu_usage and memory_bytes for one namespace how can i do that ?

[Brian Candler]

You can combine metrics using arithmetic, e.g.

    memory_bytes / cpu_usage

assuming that (a) dividing those two metrics gives a meaningful result (which is your responsibility), and (b) the two metrics have exactly the same label sets (so the corresponding timeseries can be matched 1:1).

If they don't have identical label sets then using on(), ignoring() for 1:1 matches, and group_left or group_right for 1-to-many matches are possible.  They are documented here:

https://prometheus.io/docs/prometheus/latest/querying/operators/

There are also various blogs and tutorials - those on www.robustperception.io are especially worth digesting.