Script in development used to calculate profile_list request for memory and cpu for `n2-highmem-[4|16|64]` nodes and `r5.[|4|16]xlarge` nodes

calculating_profile_list_options.py

"""
This is a basic script that prints shares of CPU/Memory for machines of
different size. The script reads a list of nodes specifications, then prints
node share choices based on how much memory is required.

The mem_guarantee specification is calculated for each node by first subtracting
4 GB of memory for each node. This is to ensure that all nodes retain ~4 GB of
memory.
"""


import sys

from ruamel.yaml import YAML

# I declared three node types all having 1:16 ratio between CPU:GB_RAM, starting
# at 4, stopping at 64, excluding the often available 96 CPU machine.
#
# I opted for mem_choices based on the following ideas:
#
# - Choose multiples of 2.
#
# - Exclude options with less than 1 GB of memory.
#
# - Exclude options that could fit more than 128 users per node, as most k8s
#   clusters and nodes only allows us to schedule a bit more than 100 users per
#   node.
#

# EKS
eks_nodes = [
    {
        "cpu": 4,
        "mem": 29.937,
        "mem_display": 32,
        "mem_choices": [1, 2, 4, 8, 16, 32],
    },
    {
        "cpu": 16,
        "mem": 121.513,
        "mem_display": 128,
        "mem_choices": [1, 2, 4, 8, 16, 32, 64, 128],
    },
    {
        "cpu": 64,
        "mem": 490.130,
        "mem_display": 512,
        "mem_choices": [4, 8, 16, 32, 64, 128, 256, 512],
    }
]

# GKE
gke_nodes = [
    {
        "cpu": 4,
        "mem": 27.738,
        "mem_display": 32,
        "mem_choices": [1, 2, 4, 8, 16, 32],
    },
    {
        "cpu": 16,
        "mem": 116.549,
        "mem_display": 128,
        "mem_choices": [1, 2, 4, 8, 16, 32, 64, 128],
    },
    {
        "cpu": 64,
        "mem": 486.949,
        "mem_display": 512,
        "mem_choices": [4, 8, 16, 32, 64, 128, 256, 512],
    }
]
nodes = gke_nodes

for node in nodes:
    print(f"Node type: {node['cpu']} CPU, {node['mem_display']} GB")
    choices = {}
    for mem_choice in node["mem_choices"]:
        mem_minus_one_factor = (node["mem"] - 1) / node["mem"]
        mem_gibi_label_factor = node["mem"] / node["mem_display"]
        mem_guarantee = round(mem_minus_one_factor * mem_gibi_label_factor * mem_choice, 3)
        cpu_guarantee = round(mem_choice / 8 * 0.1, 3)

        choices[f"mem_{mem_choice}"] = {
            "display_name": f"~{mem_choice} GB, ~{mem_choice / 8} CPU",
            "kubespawner_override": {
                "mem_guarantee": f"{mem_guarantee}G",
                "cpu_guarantee": cpu_guarantee,
            }
        }

    yaml = YAML()
    yaml.indent(mapping=2, offset=2, sequence=4)
    choices_string = yaml.dump(choices, sys.stdout)
    print("")