Nodes types filters order not kept for _nodes API

[fpompermaier]

In the cluster API the following NOTE is wrong (I think):

node filters run in the order in which they are given, which is important if using filters that remove nodes from the set. For example _all,master:false means all the nodes except the master-eligible ones, but master:false,_all means the same as _all because the _all filter runs after the master:false filter.

It should be:

node filters run in the order in which they are given, which is important if using filters that remove nodes from the set. For example _all,master:false means all the nodes except the master-eligible ones, but master:false,_all means the same as master:false because the _all filter runs after the master:false filter.

[elasticmachine]

Pinging @elastic/es-distributed

[astefan]

@fpompermaier for this one, I don't think you are right.

master: false removes from the set of nodes those that are master eligible, and _all adds back all the nodes, because _all is evaluated after master:false. It's a combination of removing master eligible nodes and adding all nodes (in this order), which means _all in the end.

[astefan]

Also, when putting master:false as the first filter, it's more or less a useless one because the filtering starts from an empty set of nodes and it's like removing the master eligible nodes from an empty set of nodes.

[fpompermaier]

Actually I'm trying the following requests on http://demo.elastic.co/app/kibana (7.0.0) but the documented statements are not true..

GET _nodes/_all # OK, returns all 9 nodes
GET _nodes/master:true # OK, returns all 3 master nodes
GET _nodes/master:false # OK, returns all 0 nodes
GET _nodes/_all,master:false == GET _nodes/_all
From documentation, GET _nodes/_all,master:false should be != GET _nodes/_all

Am I wrong @astefan ?

[astefan]

@fpompermaier you are right and we might have a bug in our code. Looking into it. And I'm re-opening this issue. Thank you for your diligence on this issue.

[astefan]

This is fixed in 8.0.0 with 74a7438 and in 7.2.0 with 912c6bd.