Issues with group_left to exclude specific label value

[Al]

All hosts from which I collect node_exporter metrics each have an additional node_role metric (added via textfile collector) which identifies all the Chef roles a given host has.  As an example, say we have 3 hosts with the following textfile collector metrics:

server1:

node_role{role="web_server_app1"}

server2:

node_role{role="redis_server"}

server3:

node_role{role="db_server"}

server4:

node_role{role="web_server_app2"}

I'm attempting to write a PromQL query which will return the current disk usage % for hosts that do not have a specific role "web_server" assigned to them.  I've attempted the following PromQL query although it's invalid as we end up with many results on the right hand side, which doesn't match the many-to-one nature of the group left:

100 - (

   (node_filesystem_free_bytes{mountpoint=“”/data} * on (hostname) group_left(role) node_role{role!~“web_server.*”} )

   /

   (node_filesystem_size_bytes{mountpoint=“”/data} * on (hostname) group_left(role) node_role{role!~“web_server”} )

   *

   100

)

How could I modify this query so that it correctly return the disk usage percentage of server2 and server3?

I appreciate your help.

[Christian Hoffmann]

Hi,

On 6/25/20 4:05 PM, Al wrote:
> All hosts from which I collect node_exporter metrics each have an
> additional node_role metric (added via textfile collector) which
> identifies all the Chef roles a given host has.  As an example, say we
> have 3 hosts with the following textfile collector metrics:
>

> *server1:*
> |
> node_role{role="web_server_app1"}
> |
>
> *server2:*
> |
> node_role{role="redis_server"}
> |
>
> *server3:*
> |
> node_role{role="db_server"}
> |
>
> *server4:*

> |
> node_role{role="web_server_app2"}
> |
>
>
>
> I'm attempting to write a PromQL query which will return the current
> disk usage % for hosts that do not have a specific role "web_server"
> assigned to them.  I've attempted the following PromQL query although
> it's invalid as we end up with many results on the right hand side,
> which doesn't match the many-to-one nature of the group left:
>
> |
>
> 100-(
>
>    (node_filesystem_free_bytes{mountpoint=“”/data}*on

> (hostname)group_left(role)node_role{role!~“web_server.*”})

>
>    /
>
>    (node_filesystem_size_bytes{mountpoint=“”/data}*on

> (hostname)group_left(role)node_role{role!~“web_server”})

>
>    *
>
>    100
>
> )
>
> |
>
> How could I modify this query so that it correctly return the disk usage
> percentage of server2 and server3?

The pattern basically looks fine.
Some remarks:

- Your quotation marks do not look like ASCII quotes -- I guess this
comes from pasting?
- The quotation marks around mountpoint= seem off (one should come after
/data).
- Your role regexp is not identical. The regexp in the second part lacks
the .* and will therefore match all your example servers.
- The role "join" can probably be omitted from the second part when
using on(instance, mountpoint)

In general I suggest trying the parts of your query individually and
only putting them into the larger query once both parts return what you
need.

Kind regards,
Christian