in the operator, as well as with other changes from last year which
assume that ansible runner is returning memory in bytes.

So you're talking about the behavior of the operator, or of something from the past?

If I check my API locally I find

"memory": 33391198208,

which is ~31 Gb, not ~31 Mb (in which case I would not be able to type this in Chrome). So hopefully we're on the same page that we're consistent in units from runner to AWX, because it is documented along all that way. See runner's behavior:

bash-5.1$ ansible-runner worker --worker-info
{cpu_count: 8, errors: [], mem_in_bytes: 33391198208, runner_version: 2.1.1.dev24,
  uuid: 126cf42a-9bd1-408d-87d4-80fd0250d289}

I'm unclear exactly how the bug here is documented. What values are getting put in? We have the expectation set correctly, it should be an integer number of bytes. You're not getting that, somewhere, for some reason?

@AlanCoding So couple things going on here:

• SYSTEM_TASK_ABS_MEM used be used by a now deprecated openshift installer
• SYSTEM_TASK_ABS_MEM used to be assumed to be Gi (gigabytes)
• now I want to assume SYSTEM_TASK_ABS_MEM is a string that conforms to k8s accepted ways of defining memory resource request/limit https://kubernetes.io/docs/tasks/configure-pod-container/assign-memory-resource/#memory-units
• with this PR, I can do set memory setting based on resource settings awx-operator#791
• I agree that runner returns bytes, I just updated the local variable "memory_mb" to be "memory_bytes" to match that
• I updated the awx/main/dispatch/pool.py to consider the memory in bytes instead of Gi, so we can use same conversion of SYSTEM_TASK_ABS_MEM to bytes as in the other place

So, there is nothing wrong with how runner is returning memory -- I'm just updating it so if we set the env vars, we get the right behavior and everything is in bytes so it works the same as the value when we get it from runner instead of a setting.

@shanemcd pushed new changes to migrate CPU to a decimal field and also to not deduct the 2GB memory penalty on k8s that we do when all the services are colocated

Here is a control pod on a minikube deployment with operator built from ansible/awx-operator#791

{
    "id": 3,
    "type": "instance",
    "url": "[/api/v2/instances/3/](http://192.168.49.2:31406/api/v2/instances/3/)",
    "related": {
        "named_url": "[/api/v2/instances/awx-7d6b447ccc-tqw9h/](http://192.168.49.2:31406/api/v2/instances/awx-7d6b447ccc-tqw9h/)",
        "jobs": "[/api/v2/instances/3/jobs/](http://192.168.49.2:31406/api/v2/instances/3/jobs/)",
        "instance_groups": "[/api/v2/instances/3/instance_groups/](http://192.168.49.2:31406/api/v2/instances/3/instance_groups/)",
        "health_check": "[/api/v2/instances/3/health_check/](http://192.168.49.2:31406/api/v2/instances/3/health_check/)"
    },
    "uuid": "cbfbfcca-55f6-4ca3-83fb-cd05cfba9089",
    "hostname": "awx-7d6b447ccc-tqw9h",
    "created": "2022-02-14T17:51:10.153995Z",
    "modified": "2022-02-14T17:51:10.154129Z",
    "last_seen": "2022-02-14T18:13:26.629703Z",
    "last_health_check": "2022-02-14T18:13:26.629703Z",
    "errors": "",
    "capacity_adjustment": "1.00",
    "version": "19.5.2.dev111+g9db07d9d7b.d20220214",
    "capacity": 10,
    "consumed_capacity": 0,
    "percent_capacity_remaining": 100.0,
    "jobs_running": 0,
    "jobs_total": 0,
    "cpu": "0.5",
    "memory": 1073741824,
    "cpu_capacity": 2,
    "mem_capacity": 10,
    "enabled": true,
    "managed_by_policy": true,
    "node_type": "control"
}

the request on the pod (this is default for awx-operator right now)

  awx-task:
....omitted...
    Requests:
      cpu:     500m
      memory:  1Gi