gitlab-org--gitlab-foss/lib/api/ml/mlflow.rb

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

265 lines
11 KiB
Ruby
Raw Normal View History

# frozen_string_literal: true
require 'mime/types'
module API
# MLFlow integration API, replicating the Rest API https://www.mlflow.org/docs/latest/rest-api.html#rest-api
module Ml
class Mlflow < ::API::Base
include APIGuard
# The first part of the url is the namespace, the second part of the URL is what the MLFlow client calls
MLFLOW_API_PREFIX = ':id/ml/mlflow/api/2.0/mlflow/'
allow_access_with_scope :api
allow_access_with_scope :read_api, if: -> (request) { request.get? || request.head? }
feature_category :mlops
content_type :json, 'application/json'
default_format :json
before do
# MLFlow Client considers any status code different than 200 an error, even 201
status 200
authenticate!
not_found! unless Feature.enabled?(:ml_experiment_tracking, user_project)
end
rescue_from ActiveRecord::ActiveRecordError do |e|
invalid_parameter!(e.message)
end
helpers do
def resource_not_found!
render_structured_api_error!({ error_code: 'RESOURCE_DOES_NOT_EXIST' }, 404)
end
def resource_already_exists!
render_structured_api_error!({ error_code: 'RESOURCE_ALREADY_EXISTS' }, 400)
end
def invalid_parameter!(message = nil)
render_structured_api_error!({ error_code: 'INVALID_PARAMETER_VALUE', message: message }, 400)
end
def experiment_repository
::Ml::ExperimentTracking::ExperimentRepository.new(user_project, current_user)
end
def candidate_repository
::Ml::ExperimentTracking::CandidateRepository.new(user_project, current_user)
end
def experiment
@experiment ||= find_experiment!(params[:experiment_id], params[:experiment_name])
end
def candidate
@candidate ||= find_candidate!(params[:run_id])
end
def find_experiment!(iid, name)
experiment_repository.by_iid_or_name(iid: iid, name: name) || resource_not_found!
end
def find_candidate!(iid)
candidate_repository.by_iid(iid) || resource_not_found!
end
def packages_url
path = api_v4_projects_packages_generic_package_version_path(
id: user_project.id, package_name: '', file_name: ''
)
path = path.delete_suffix('/package_version')
"#{request.base_url}#{path}"
end
end
params do
requires :id, type: String, desc: 'The ID of a project'
end
resource :projects, requirements: API::NAMESPACE_OR_PROJECT_REQUIREMENTS do
desc 'API to interface with MLFlow Client, REST API version 1.28.0' do
detail 'This feature is gated by :ml_experiment_tracking.'
end
namespace MLFLOW_API_PREFIX do
resource :experiments do
desc 'Fetch experiment by experiment_id' do
success Entities::Ml::Mlflow::GetExperiment
detail 'https://www.mlflow.org/docs/1.28.0/rest-api.html#get-experiment'
end
params do
optional :experiment_id, type: String, default: '', desc: 'Experiment ID, in reference to the project'
end
get 'get', urgency: :low do
present experiment, with: Entities::Ml::Mlflow::GetExperiment
end
desc 'Fetch experiment by experiment_name' do
success Entities::Ml::Mlflow::GetExperiment
detail 'https://www.mlflow.org/docs/1.28.0/rest-api.html#get-experiment-by-name'
end
params do
optional :experiment_name, type: String, default: '', desc: 'Experiment name'
end
get 'get-by-name', urgency: :low do
present experiment, with: Entities::Ml::Mlflow::GetExperiment
end
desc 'List experiments' do
success Entities::Ml::Mlflow::ListExperiment
detail 'https://www.mlflow.org/docs/latest/rest-api.html#list-experiments'
end
get 'list', urgency: :low do
response = { experiments: experiment_repository.all }
present response, with: Entities::Ml::Mlflow::ListExperiment
end
desc 'Create experiment' do
success Entities::Ml::Mlflow::NewExperiment
detail 'https://www.mlflow.org/docs/1.28.0/rest-api.html#create-experiment'
end
params do
requires :name, type: String, desc: 'Experiment name'
optional :artifact_location, type: String, desc: 'This will be ignored'
optional :tags, type: Array, desc: 'This will be ignored'
end
post 'create', urgency: :low do
present experiment_repository.create!(params[:name]), with: Entities::Ml::Mlflow::NewExperiment
rescue ActiveRecord::RecordInvalid
resource_already_exists!
end
end
resource :runs do
desc 'Creates a Run.' do
success Entities::Ml::Mlflow::Run
detail ['https://www.mlflow.org/docs/1.28.0/rest-api.html#create-run',
'MLFlow Runs map to GitLab Candidates']
end
params do
requires :experiment_id, type: Integer,
desc: 'Id for the experiment, relative to the project'
optional :start_time, type: Integer,
desc: 'Unix timestamp in milliseconds of when the run started.',
default: 0
optional :user_id, type: String, desc: 'This will be ignored'
optional :tags, type: Array, desc: 'This will be ignored'
end
post 'create', urgency: :low do
present candidate_repository.create!(experiment, params[:start_time]),
with: Entities::Ml::Mlflow::Run, packages_url: packages_url
end
desc 'Gets an MLFlow Run, which maps to GitLab Candidates' do
success Entities::Ml::Mlflow::Run
detail 'https://www.mlflow.org/docs/1.28.0/rest-api.html#get-run'
end
params do
requires :run_id, type: String, desc: 'UUID of the candidate.'
optional :run_uuid, type: String, desc: 'This parameter is ignored'
end
get 'get', urgency: :low do
present candidate, with: Entities::Ml::Mlflow::Run, packages_url: packages_url
end
desc 'Updates a Run.' do
success Entities::Ml::Mlflow::UpdateRun
detail ['https://www.mlflow.org/docs/1.28.0/rest-api.html#update-run',
'MLFlow Runs map to GitLab Candidates']
end
params do
requires :run_id, type: String, desc: 'UUID of the candidate.'
optional :status, type: String,
values: ::Ml::Candidate.statuses.keys.map(&:upcase),
desc: "Status of the run. Accepts: " \
"#{::Ml::Candidate.statuses.keys.map(&:upcase)}."
optional :end_time, type: Integer, desc: 'Ending time of the run'
end
post 'update', urgency: :low do
candidate_repository.update(candidate, params[:status], params[:end_time])
present candidate, with: Entities::Ml::Mlflow::UpdateRun, packages_url: packages_url
end
desc 'Logs a metric to a run.' do
summary 'Log a metric for a run. A metric is a key-value pair (string key, float value) with an '\
'associated timestamp. Examples include the various metrics that represent ML model accuracy. '\
'A metric can be logged multiple times.'
detail 'https://www.mlflow.org/docs/1.28.0/rest-api.html#log-metric'
end
params do
requires :run_id, type: String, desc: 'UUID of the run.'
requires :key, type: String, desc: 'Name for the metric.'
requires :value, type: Float, desc: 'Value of the metric.'
requires :timestamp, type: Integer, desc: 'Unix timestamp in milliseconds when metric was recorded'
optional :step, type: Integer, desc: 'Step at which the metric was recorded'
end
post 'log-metric', urgency: :low do
candidate_repository.add_metric!(
candidate,
params[:key],
params[:value],
params[:timestamp],
params[:step]
)
{}
end
desc 'Logs a parameter to a run.' do
summary 'Log a param used for a run. A param is a key-value pair (string key, string value). '\
'Examples include hyperparameters used for ML model training and constant dates and values '\
'used in an ETL pipeline. A param can be logged only once for a run, duplicate will be .'\
'ignored'
detail 'https://www.mlflow.org/docs/1.28.0/rest-api.html#log-param'
end
params do
requires :run_id, type: String, desc: 'UUID of the run.'
requires :key, type: String, desc: 'Name for the parameter.'
requires :value, type: String, desc: 'Value for the parameter.'
end
post 'log-parameter', urgency: :low do
bad_request! unless candidate_repository.add_param!(candidate, params[:key], params[:value])
{}
end
desc 'Logs multiple parameters and metrics.' do
summary 'Log a batch of metrics and params for a run. Validation errors will block the entire batch, '\
'duplicate errors will be ignored.'
detail 'https://www.mlflow.org/docs/1.28.0/rest-api.html#log-param'
end
params do
requires :run_id, type: String, desc: 'UUID of the run.'
optional :metrics, type: Array, default: [] do
requires :key, type: String, desc: 'Name for the metric.'
requires :value, type: Float, desc: 'Value of the metric.'
requires :timestamp, type: Integer, desc: 'Unix timestamp in milliseconds when metric was recorded'
optional :step, type: Integer, desc: 'Step at which the metric was recorded'
end
optional :params, type: Array, default: [] do
requires :key, type: String, desc: 'Name for the metric.'
requires :value, type: String, desc: 'Value of the metric.'
end
end
post 'log-batch', urgency: :low do
candidate_repository.add_metrics(candidate, params[:metrics])
candidate_repository.add_params(candidate, params[:params])
{}
end
end
end
end
end
end
end