hopsworks.core.costs_api #

Get the cost of each department over a recent time window.

This is the namespace view of get_namespace_costs grouped by the department of each namespace's projects, with the same totals and buckets. Each department gets one DEPARTMENT row, whose namespace_count says how many namespaces it sums. Projects are counted in their current department, so a project moved to another department moves its whole history with it, and a project created without a department is in the default department. A namespace that is no project's, such as Hopsworks' own or kube-system, or that is shared by projects of different departments, is summed into one UNASSIGNED row. The IDLE and UNMOUNTED rows are the same as in the namespace view.

import hopsworks

hopsworks.login()
costs = hopsworks.get_costs_api().get_department_costs("month")

for row in costs.items:
    if row.kind == "DEPARTMENT":
        print(row.department_name, row.namespace_count, row.total_cost)

Get the cost of each Kubernetes namespace over a recent time window.

Each Hopsworks project runs in its own namespace, so this is the cost per project, plus the cluster's idle capacity and unmounted volumes. Hour and day are computed live by OpenCost. Week, month and year combine the daily history Hopsworks records with live OpenCost data for the days not recorded yet, so a stored day is not recomputed and the window covers only the days since recording started or still within OpenCost's retention. The returned coverage says how many of the window's buckets have data, and a bucket without data is unknown cost rather than zero.

import hopsworks

hopsworks.login()
costs = hopsworks.get_costs_api().get_namespace_costs("month")

print(costs.total_cost, costs.currency)
for row in costs.items:
    print(row.namespace, row.total_cost)

Get the cost of one project over a recent time window.

Only the project's data owners can call it. The result has the shape of get_namespace_costs with the project's own NAMESPACE row alone: no other project, no idle capacity, and no share. Its totals are the project's, and coverage is the cluster's, so a bucket with data and no row is one in which the project cost nothing.

import hopsworks

project = hopsworks.login()
costs = hopsworks.get_costs_api().get_project_costs("month")

print(costs.total_cost, costs.currency)
for bucket in costs.buckets:
    if not bucket.observed:
        print(bucket.start, "no data")
        continue
    print(bucket.start, sum(row.total_cost for row in bucket.items))