I was writing an ansible playbook to deploy an Apache Flink cluster, and I ran into this problem while editing the configuration files:

How can I get the total number of the vCPUs of all the “slave” machines running on the cluster?

I thought of the hostvars ansible variable, which stores the gathered facts about the hosts, so using the debug module, I printed it out. The result was a JSON like that:

{
    "node1": {"flink_type": "master", "ansible_processor_vcpus": 2, ...},
    "node2": {"flink_type": "slave", "ansible_processor_vcpus": 4, ...},
    "node3": {"flink_type": "slave", "ansible_processor_vcpus": 8, ...}
}

So all I needed to do was to filter the nodes on the value of the attribute flink_type and get a list of their ansible_processor_vcpus values, so that I can then sum them up with jinja, right?

Well it wasn’t very easy. As it turns out, if my case was:

{
    "nodes": [
        {"flink_type": "master", "ansible_processor_vcpus": 2, ...},
        {"flink_type": "slave", "ansible_processor_vcpus": 4, ...},
        {"flink_type": "slave", "ansible_processor_vcpus": 8, ...}
    ]
}

the query-solution would be as easy as nodes[?flink_type=='slave'].ansible_processor_vcpus, which would return the list [4, 8].

But in my case, I first needed to convert the JSON into a list of objects, then get the first element of this list, and then perform the filtering and attribute selection.

The solution I have found to be working correctly is this one:

[@.*][0][?flink_type==`slave`].ansible_processor_vcpus

, which is very ugly. I was hoping I could do something like *[?flink_type=='slave'].ansible_processor_vcpus but this doesn’t work.

If there is a more elegant solution to this, I would really like to hear it, contact links are below.