dominodatalab 2.0.0

Python bindings for the Domino API

This library provides bindings for the Domino APIs. It ships with the Domino Standard Environment (DSE).

See this documentation for details about the APIs:

• Latest public Domino APIs

• Legacy APIs

The latest released version of python-domino is 2.0.0.

Version compatibility matrix

The python-domino library is compatible with different versions of Domino:

Domino Versions	Python-Domino
3.6.x or lower	0.3.5
4.1.0 or higher	1.0.0 or Higher
5.3.0 or higher	1.2.0 or Higher
5.5.0 or higher	1.2.2 or Higher
5.10.0 or higher	1.3.1 or Higher
5.11.0 or higher	1.4.1 or Higher
6.0.0 or higher	1.4.8 or Higher
6.2.0 or higher	2.0.0 or Higher

Development

The current python-domino is based on Python 3.10, which is therefore recommended for development. Pipenv is also recommended to manage the dependencies.

To use the Python binding in a Domino workbook session, include dominodatalab in your project's requirements.txt file. This makes the Python binding available for each new workbook session (or batch run) started within the project.

To install dependencies from setup.py for development:

pipenv --python 3.10 install -e ".[dev]"

Use the same process for Airflow, agents, and data:

pipenv --python 3.10 install -e ".[data]" ".[airflow]" ".[agents]"

Set up the connection

You can set up the connection by creating a new instance of Domino:

_class_ Domino(project, api_key=None, host=None, domino_token_file=None, auth_token=None)

• project: A project identifier (in the form of ownerusername/projectname).

• api_proxy: (Optional) Location of the Domino API reverse proxy as host:port.

  If set, this proxy is used to intercept any Domino API requests and insert an authentication token. This is the preferred method of authentication. Alternatively, set the DOMINO_API_PROXY environment variable. In Domino 5.4.0 or later, this variable is set inside a Domino run container.

  NOTE: This mechanism does not work when connecting to an HTTPS endpoint; it is meant to be used inside Domino runs.

• api_key: (Optional) An API key to authenticate with.

  If not provided, the library expects to find one in the DOMINO_USER_API_KEY environment variable. If you are using the Python package in code that is already running in Domino, the DOMINO_API_USER_KEY variable is set automatically to be the key for the user who started the run.

• host: (Optional) A host URL.

  If not provided, the library expects to find one in the DOMINO_API_HOST environment variable.

• domino_token_file: (Optional) Path to the Domino token file containing the auth token.

  If not provided, the library expects to find one in the DOMINO_TOKEN_FILE environment variable. If you are using Python package in code that is already running in Domino, the DOMINO_TOKEN_FILE is set automatically to be the token file for the user who started the run.

• auth_token: (Optional) Authentication token.

Authentication

Domino looks for the authentication method in the following order and uses the first one it finds:

1. api_proxy
2. auth_token
3. domino_token_file
4. api_key
5. DOMINO_API_PROXY
6. DOMINO_TOKEN_FILE
7. DOMINO_USER_API_KEY

The API proxy is the preferred method of authentication. See Use the API Proxy to Authenticate Calls to the Domino API.

Additional environment variables

• DOMINO_LOG_LEVEL

  The default log level is INFO. You can change the log level by setting DOMINO_LOG_LEVEL, for example to DEBUG.

• DOMINO_VERIFY_CERTIFICATE

  For testing purposes and issues with SSL certificates, set DOMINO_VERIFY_CERTIFICATE to false. Be sure to unset this variable when not in use.

• DOMINO_MAX_RETRIES

  Default Retry is set to 4 Determines the number of attempts for the request session in case of a ConnectionError Get more info on request max timeout/error durations based on Retry and backoff factors here

• MLFLOW_TRACKING_URI

  Must be set for the domino.agents package to work properly. This will be set automatically in executions, e.g. Jobs, Workspaces, Scheduled Jobs, etc. For testing, the value must be set on the command line to "http://localhost:5000".

Working with docs

building locally

• Install the app in dev mode and all dependencies and pandoc: brew install pandoc ghc cabal-install haskell-stack and cabal update && cabal install --lib pandoc-types --package-env .
• install dependencies in this order: pipenv --python 3.10 install -e ".[dev]" && pipenv --python 3.10 install -e ".[docs]" && pipenv --python 3.10 install -e ".[data]" ".[airflow]" ".[agents]"
• Build: ./docs/build.sh
• View: open docs/build/html/index.html
• Manually pick the changes you want and add to README.adoc and README.md. Update the styling of the docs as needed.

transform agents docs for README

• Run: ./docs/transform_agents_docs_for_readme.sh
• copy/paste the outputs for asciidoc and markdown into respective README files

Methods

Budgets and Billing Tags

See example_budget_manager.py for example code.

budget_defaults_list()

Get a list of the available default budgets with the assigned (if any) limits Requires Admin permission

budget_defaults_update(budget_label, budget_limit)

Update default budgets by BudgetLabel Requires Admin roles

• budget_label: (required) label of budget to be updated ex: BillingTag, Organization

• budget_limit: (required) new budget quota to assign to default label

budget_overrides_list()

Get a list of the available budgets overrides with the assigned limits. Requires Admin permission

budget_override_create(budget_label, budget_id, budget_limit)

Create Budget overrides based on BudgetLabels, ie BillingTags, Organization, or Projects the object id is used as budget ids Requires Admin roles

• budget_label: label of budget to be updated

• budget_id: id of project or organization to be used as new budget override id.

• budget_limit: budget quota to assign to override

budget_override_update(budget_label, budget_id, budget_limit)

Update Budget overrides based on BudgetLabel and budget id Requires Admin roles

• budget_label: label of budget to be updated

• budget_id: id of budget override to be updated.

• budget_limit: new budget quota to assign to override

budget_override_delete(budget_id)

Delete an existing budget override Requires Admin roles

• budget_id: id of budget override to be deleted.

budget_alerts_settings()

Get the current budget alerts settings Requires Admin permission

budget_alerts_settings_update(alerts_enabled, notify_org_owner)

Update the current budget alerts settings to enable/disable budget notifications and whether to notify org owners on projects notifications Requires Admin permission

• alerts_enabled: whether to enable or disable notifications.

• notify_org_owner: whether to notify organizations owners on projects reaching threshold.

budget_alerts_targets_update(targets)

Update the current budget alerts settings with additional email targets per budget label Requires Admin permission

• targets: dictionary of budget labels and list of email addresses

billing_tags_list_active()

Get a list of active billing tags Requires Admin permission

billing_tags_create(tags_list)

Create a list of active billing tags Requires Admin permission

• tags_list: list of billing tags names to be created

active_billing_tag_by_name(name)

Get detailed info on active or archived billing tag Requires Admin permission

• name: name of existing billing tag

billing_tag_archive(name)

Archive an active billing tag Requires Admin permission

• name: name of existing billing tag to archive

billing_tag_settings()

Get the current billing tag settings Requires Admin permission

billing_tag_settings_mode()

Get the current billing tag settings mode Requires Admin permission

billing_tag_settings_mode_update(mode)

Update the current billing tag settings mode Requires Admin permission

• mode: new mode to set the billing tag settings (see BillingTagSettingMode)

project_billing_tag(project_id)

Get a billing tag assigned to a particular project by project id Requires Admin permission

• project_id: id of the project to find assigned billing tag

project_billing_tag_update(billing_tag, project_id)

Update project's billing tag with new billing tag. Requires Admin permission

• billing_tag: billing tag to assign to a project

• project_id: id of the project to assign a billing tag

project_billing_tag_reset(project_id)

Remove a billing tag from a specified project Requires Admin permission

• project_id: id of the project to reset billing tag field

projects_by_billing_tag( billing_tag, offset, page_size, name_filter, sort_by, sort_order, missing_tag_only=False)

Remove a billing tag from a specified project Requires Admin permission

• billing_tag: billing tag string to filter projects by

• offset: The index of the start of the page, where checkpointProjectId is index 0. If the offset is negative the project it points to will be the end of the page.

• page_size: The number of records to return per page.

• name_filter: matches projects by name substring

• sort_by: (Optional) field to sort the projects on

• sort_order: (Optional) Whether to sort in asc or desc order

• missing_tag_only: (Optional) determine whether to only return projects with missing tag

project_billing_tag_bulk_update(projects_tag)

Update project's billing tags in bulk Requires Admin permission

• projects_tag: dictionary of project_id and billing_tags

Projects

See example_projects_usage.py for example code.

project_create_v4(project_name, owner_id, owner_username, description, collaborators, tags, billing_tag, visibility=PUBLIC)

Newer version of projects creation using the v4 endpoints which allows more optional fields.

• project_name: (required) The name of the project.

• owner_id: (Optional) user id of the owner of the new project to be created (must be admin to create projects for other users) owner_id or owner_username can be used, both are not needed (Defaults to current owner_username)

• owner_username: (Optional) username of the owner of the new project to be created (must be admin to create projects for other users) owner_id or owner_username can be used, both are not needed (Defaults to current owner_username)

• description: (Optional) description of the project

• collaborators: (Optional) list of collaborators to be added to the project

• tags: (Optional) list of tags to add to project

• billing_tag: (Optional unless billingTag settings mode is Required) active billing tag to be added to projects for governance

• visibility: (Optional) (Defaults to Public) project visibility

project_create(project_name, owner_username=None)

Create a new project with given project name.

• project_name: The name of the project.

• owner_username: (Optional) The owner username for the project. This parameter is useful when you need to create a project under an organization.

collaborators_get()

Get the list of collaborators on a project.

collaborators_add(username_or_email, message="")

Add collaborators to a project.

• username_or_email: Name or email of the Domino user to add as collaborator to the current project.

• message: Optional - Message related to the user’s role or purpose to the project.

Project tags

Project tags are an easy way to add freeform metadata to a project. Tags help colleagues and consumers organize and find the Domino projects that interest them. Tags can be used to describe the subject explored by a project, the packages and libraries it uses, or the source of the data within.

See example_projects_usage.py for example code.

tags_list(*project_id)

List a project’s tags.

• project_id: The project identifier.

tag_details(tag_id)

Get details about a tag.

• tag_id: The tag identifier.

tags_add(tags, *project_id)

Create a tag, if it does not exist, and add it to a project.

• tags (list): One or more tag names.

• project_id: (Defaults to current project ID) The project identifier.

tag_get_id(tag_name, *project_id)

Get the tag ID using the tag string name.

• tag_name (string): The tag name.

• project_id: (Defaults to current project id) The project ID.

tags_remove(tag_name, project_id=None)

Remove a tag from a project.

• tag_name (string): The tag name.

• project_id: (Defaults to current project id) The project ID.

Executions

See these code example files:

• start_run_and_check_status.py

• export_runs.py

runs_list()

List the executions on the selected project.

runs_start(command, isDirect, commitId, title, tier, publishApiEndpoint)

Start a new execution on the selected project.

• command: The command to execution as an array of strings where members of the array represent arguments of the command. For example: ["main.py", "hi mom"]

• isDirect: (Optional) Whether this command should be passed directly to a shell.

• commitId: (Optional) The commitId to launch from. If not provided, the project launches from the latest commit.

• title: (Optional) A title for the execution.

• tier: (Optional) The hardware tier to use for the execution. This is the human-readable name of the hardware tier, such as "Free", "Small", or "Medium". If not provided, the project’s default tier is used.

• publishApiEndpoint: (Optional) Whether to publish an API endpoint from the resulting output.

runs_start_blocking(command, isDirect, commitId, title, tier, publishApiEndpoint, poll_freq=5, max_poll_time=6000)

Start a new execution on the selected project and make a blocking request that waits until job is finished.

• command: The command to execution as an array of strings where members of the array represent arguments of the command. For example: ["main.py", "hi mom"]

• isDirect: (Optional) Whether this command should be passed directly to a shell.

• commitId: (Optional) The commitId to launch from. If not provided, the project launches from the latest commit.

• title: (Optional) A title for the execution.

• tier: (Optional) The hardware tier to use for the execution. Will use project’s default tier if not provided. If not provided, the project’s default tier is used.

• publishApiEndpoint: (Optional) Whether to publish an API endpoint from the resulting output.

• poll_freq: (Optional) Number of seconds between polling of the Domino server for status of the task that is running.

• max_poll_time: (Optional) Maximum number of seconds to wait for a task to complete. If this threshold is exceeded, an exception is raised.

• retry_count: (Optional) Maximum number of polling retries (in case of transient HTTP errors). If this threshold is exceeded, an exception is raised.

run_stop(runId, saveChanges=True):

Stop an existing execution in the selected project.

• runId: String that identifies the execution.

• saveChanges: (Defaults to True) If false, execution results are discarded.

runs_stdout(runId)

Get stdout emitted by a particular execution.

• runId: string that identifies the execution

Files and blobs

See these code example files:

• upload_file.py

• upload_and_run_file_and_download_results.py

files_list(commitId, path)

List the files in a folder in the Domino project.

• commitId: The commitId to list files from.

• path: (Defaults to "/") The path to list from.

files_upload(path, file)

Upload a Python file object into the specified path inside the project. See examples/upload_file.py for an example. All parameters are required.

• path: The path to save the file to. For example, /README.md writes to the root directory of the project while /data/numbers.csv saves the file to a sub folder named data. If the specified folder does not yet exist, it is created.

• file: A Python file object. For example: f = open("authors.txt","rb")

blobs_get(key)

Deprecated Retrieve a file from the Domino server by blob key. Use blobs_get_v2(path, commit_id, project_id) instead.

• key: The key of the file to fetch from the blob server.

blobs_get_v2(path, commit_id, project_id)

Retrieve a file from the Domino server in a project from its path and commit id.

• path: The path to the file in the Domino project.
• commit_id: ID of the commit to retrieve the file from.
• project_id: ID of the project to retrieve the file from.

Apps

app_publish(unpublishRunningApps=True, hardwareTierId=None)

Publish an app within a project, or republish an existing app.

• unpublishRunningApps: (Defaults to True) Check for an active app instance in the current project and unpublish it before re/publishing.

• hardwareTierId: (Optional) Launch the app on the specified hardware tier.

app_unpublish()

Stop the running app in the project.

Jobs

job_start(command, commit_id=None, hardware_tier_name=None, environment_id=None, on_demand_spark_cluster_properties=None, compute_cluster_properties=None, external_volume_mounts=None, title=None):

Start a new job (execution) in the project.

• command (string): Command to execute in Job. For example: domino.job_start(command="main.py arg1 arg2")

• commit_id (string): (Optional) The commitId to launch from. If not provided, the job launches from the latest commit.

• hardware_tier_name (string): (Optional) The hardware tier NAME to launch job in. If not provided, the project’s default tier is used.

• environment_id (string): (Optional) The environment ID with which to launch the job. If not provided, the project’s default environment is used.

• on_demand_spark_cluster_properties (dict): (Optional) On demand spark cluster properties. The following properties can be provided in the Spark cluster:

  {
      "computeEnvironmentId": "<Environment ID configured with spark>"
      "executorCount": "<Number of Executors in cluster>"
       (optional defaults to 1)
      "executorHardwareTierId": "<Hardware tier ID for Spark Executors>"
       (optional defaults to last used historically if available)
      "masterHardwareTierId":  "<Hardware tier ID for Spark master"
       (optional defaults to last used historically if available)
      "executorStorageMB": "<Executor's storage in MB>"
       (optional defaults to 0; 1GB is 1000MB Here)
  }
  

• param compute_cluster_properties (dict): (Optional) The compute-cluster properties definition contains parameters for launching any Domino supported compute cluster for a job. Use this to launch a job that uses a compute-cluster instead of the deprecated on_demand_spark_cluster_properties field. If on_demand_spark_cluster_properties and compute_cluster_properties are both present, on_demand_spark_cluster_properties is ignored. compute_cluster_properties contains the following fields:

  {
      "clusterType": <string, one of "Ray", "Spark", "Dask", "MPI">,
      "computeEnvironmentId": <string, The environment ID for the cluster's nodes>,
      "computeEnvironmentRevisionSpec": <one of "ActiveRevision", "LatestRevision",
      {"revisionId":"<environment_revision_id>"} (optional)>,
      "masterHardwareTierId": <string, the Hardware tier ID for the cluster's master node (required unless clusterType is MPI)>,
      "workerCount": <number, the total workers to spawn for the cluster>,
      "workerHardwareTierId": <string, The Hardware tier ID for the cluster workers>,
      "workerStorage": <{ "value": <number>, "unit": <one of "GiB", "MB"> },
      The disk storage size for the cluster's worker nodes (optional)>
      "maxWorkerCount": <number, The max number of workers allowed. When
      this configuration exists, autoscaling is enabled for the cluster and
      "workerCount" is interpreted as the min number of workers allowed in the cluster
      (optional)>
  }
  

• external_volume_mounts (List[string]): (Optional) External volume mount IDs to mount to execution. If not provided, the job launches with no external volumes mounted.

• *title (string): (Optional) Title for Job.

job_stop(job_id, commit_results=True):

Stop the Job (execution) in the project.

• job_id (string): Job identifier.

• commit_results (boolean): (Defaults to true) If false, the job results are not committed.

jobs_list(self, project_id: str, order_by: str = "number", sort_by: str = "desc", page_size: Optional[int] = None, page_no: int = 1, show_archived: str = "false", status: str = "all", tag: Optional[str] = None):

Lists job history for a given project_id

• project_id (string): The project to query

• order_by (string): Field on which sort has to be applied– e.g. "title" (default "number")

• sort_by (string): Sort "desc" (default) or "asc"

• page_size (integer): The number of jobs to return (default: 3)

• page_no (integer): Page number to fetch (default: 1).

• show_archived (string): Show archived jobs in results (default: false)

• status (string): Status of jobs to fetch– e.g. "completed" (default: "all")

• tag (string): Optional tag filter

job_status(job_id):

Get the status of a job.

• job_id (string): Job identifier.

job_restart(job_id, should_use_original_input_commit=True):

Restart a previous job

• job_id (string): ID of the original job. This can be obtained with jobs_list().

• should_use_original_input_commit (bool): Should the new job run use the original code, or the current version?

job_start_blocking(poll_freq=5, max_poll_time=6000, **kwargs):

Start a job and poll until the job is finished. Additionally, this method supports all the parameters in the job_start method.

• poll_freq: Poll frequency interval in seconds.

• max_poll_time: Max poll time in seconds.

Datasets

A Domino dataset is a collection of files that are available in user executions as a filesystem directory. A dataset always reflects the most recent version of the data. You can modify the contents of a dataset through the Domino UI or through workload executions.

See Domino Datasets for more details, and example_dataset.py for example code.

datasets_list(project_id=None)

Provide a JSON list of all the available datasets.

• project_id (string): (Defaults to None) The project identifier. Each project can hold up to 5 datasets.

datasets_ids(project_id)

List the IDs the datasets for a particular project.

• project_id: The project identifier.

datasets_names(project_id)

List the names the datasets for a particular project.

• project_id: The project identifier.

datasets_details(dataset_id)

Provide details about a dataset.

• dataset_id: The dataset identifier.

datasets_create(dataset_name, dataset_description)

Create a new dataset.

• dataset_name: Name of the new dataset. NOTE: The name must be unique.

• dataset_description: Description of the dataset.

datasets_update_details(dataset_id, dataset_name=None, dataset_description=None)

Update a dataset’s name or description.

• dataset_id: The dataset identifier.

• dataset_name: (Optional) New name of the dataset.

• dataset_description: (Optional) New description of the dataset.

datasets_remove(dataset_ids)

Delete a set of datasets.

• dataset_ids (list[string]): List of IDs of the datasets to delete. NOTE: Datasets are first marked for deletion, then deleted after a grace period (15 minutes, configurable). A Domino admin may also need to complete this process before the name can be reused.

datasets_upload_files(dataset_id, local_path_to_file_or_directory, file_upload_setting, max_workers, target_chunk_size, target_relative_path)

Uploads a file or entire directory to a dataset.

• dataset_id: The dataset identifier.
• local_path_to_file_or_directory: The path to the file or directory in local machine.
• file_upload_setting: The setting to resolve naming conflict, must be one of Overwrite, Rename, Ignore (default).
• max_workers: The max amount of threads (default: 10).
• target_chunk_size: The max chunk size for multipart upload (default: 8MB).
• target_relative_path: The path on the dataset to upload the file or directory to. Note that the path must exist or the upload will fail.

Agents

domino.agents.environment_variables

DOMINO_AGENT_CONFIG_PATH:
For configuring the location of the agent_config.yaml file. If not set, defaults to ‘./agent_config.yaml’.

type:
str

DOMINO_AGENT_IS_PROD:
Indicates if the Agent is running in production mode. Set to ‘true’ to optimize for production.

type:
str

DOMINO_APP_ID:
Indicates the ID of the Agent application. Must be set in production mode.

type:
str

MLFLOW_TRACKING_URI:
Used to configure Mlflow functionality. It is required in order for library to work and will be set automatically when running in Domino.

type:
str

domino.agents.logging

Functions

log_evaluation(trace_id, name, value)	This logs evaluation data and metadata to a parent trace.

Classes

DominoRun([experiment_name, run_id, ...])	DominoRun is a context manager that starts an Mlflow run and attaches the user's Agent configuration to it, create a Logged Model with the Agent configuration, and computes summary metrics for evaluation traces made during the run.

class domino.agents.logging.DominoRun(experiment_name: str | None = None, run_id: str | None = None, agent_config_path: str | None = None, custom_summary_metrics: list[str, Literal['mean', 'median', 'stdev', 'max', 'min']] | None = None)

Bases: object

DominoRun is a context manager that starts an Mlflow run and attaches the user’s Agent configuration to it, create a Logged Model with the Agent configuration, and computes summary metrics for evaluation traces made during the run. Average metrics are computed by default, but the user can provide a custom list of evaluation metric aggregators. This is intended to be used in development mode for Agent evaluation. Context manager docs: https://docs.python.org/3/library/contextlib.html

Parallelism: DominoRun is not thread-safe. Runs in different threads will work correctly. This is due to Mlflow’s architecture. Parallelizing operations within a single DominoRun context however, is supported.

Example

import mlflow

mlflow.set_experiment(“my_experiment”)

with DominoRun():
train_model()

Parameters:

• experiment_name – the name of the mlflow experiment to log the run to.

• run_id – optional, the ID of the mlflow run to continue logging to. If not provided a new run will start.

• agent_config_path – the optional path to the Agent configuration file. If not provided, defaults to the DOMINO_AGENT_CONFIG_PATH environment variable.

• custom_summary_metrics – an optional list of tuples that define what summary statistic to use with what evaluation metric. Valid summary statistics are: “mean”, “median”, “stdev”, “max”, “min” e.g. [(“hallucination_rate”, “max”)]

Returns: DominoRun context manager

 

domino.agents.logging.log_evaluation(trace_id: str, name: str, value: float | str)

This logs evaluation data and metadata to a parent trace. This is used to log the evaluation of a span after it was created. This is useful for analyzing past performance of an Agent component.

Parameters:

• trace_id – the ID of the trace to evaluate

• name – a label for the evaluation result. This is used to identify the evaluation result

• value – the evaluation result to log. This must be a float or string

Modules

dominorun
logging

domino.agents.tracing

Functions

add_tracing(name[, autolog_frameworks, ...])	This is a decorator that starts an mlflow span for the function it decorates.
init_tracing([autolog_frameworks])	Initialize Mlflow autologging for various frameworks and sets the active experiment to enable tracing in production.
search_agent_traces(agent_id[, ...])	This allows searching for traces that have a certain name and returns a paginated response of trace summaries that include the spans that were requested.
search_traces(run_id[, trace_name, ...])	This allows searching for traces that have a certain name and returns a paginated response of trace summaries that inclued the spans that were requested.

Classes

EvaluationResult(name, value)	An evaluation result for a trace.
SearchTracesResponse(data, page_token)	The response from searching for traces.
SpanSummary(id, name, trace_id, inputs, outputs)	A span in a trace.
TraceSummary(name, id, spans, evaluation_results)	A summary of a trace.

class domino.agents.tracing.EvaluationResult(name: str, value: float | str)

Bases: object

An evaluation result for a trace.

name*: str*
The name of the evaluation

value*: float | str*
The value of the evaluation

 

class domino.agents.tracing.SearchTracesResponse(data: list[TraceSummary], page_token: str | None)

Bases: object

The response from searching for traces.

data*: list[TraceSummary]*
The list of trace summaries

page_token*: str | None*
The token for the next page of results

 

class domino.agents.tracing.SpanSummary(id: str, name: str, trace_id: str, inputs: Any, outputs: Any)

Bases: object

A span in a trace.

id*: str*
the mlflow ID of the span

inputs*: Any*
The inputs to the function that created the span

name*: str*
The name of the span

outputs*: Any*
The outputs of the function that created the span

trace_id*: str*
The parent trace ID

 

class domino.agents.tracing.TraceSummary(name: str, id: str, spans: list[SpanSummary], evaluation_results: list[EvaluationResult])

Bases: object

A summary of a trace.

evaluation_results*: list[EvaluationResult]*
The evaluation results for this trace

id*: str*
The mlflow ID of the trace

name*: str*
The name of the trace

spans*: list[SpanSummary]*
The child spans of this trace

 

domino.agents.tracing.add_tracing(name: str, autolog_frameworks: list[str] | None = [], evaluator: Callable[[mlflow.entities.Span], dict[str, int | float | str]] | None = None, trace_evaluator: Callable[[mlflow.entities.Trace], dict[str, int | float | str]] | None = None, eagerly_evaluate_streamed_results: bool = True, allow_tracing_evaluator: bool = False)

This is a decorator that starts an mlflow span for the function it decorates. If there is an existing trace a span will be appended to it. If there is no existing trace, a new trace will be created.

It also enables the user to run evaluators when the code is run in development mode. Evaluators can be run on the span and/or trace generated for the wrapped function call. The trace evaluator will run if the parent trace was started and finished by the related decorator call. The trace will contain all child span information. The span evaluator will always run. The evaluation results from both evaluators will be combined and saved to the trace.

This decorator must be used directly on the function to be traced without any intervening decorators, because it must have access to the arguments.

@add_tracing(
name=”assistant_chat_bot”, evaluator=evaluate_helpfulness,

) def ask_chat_bot(user_input: str) -> dict:

…

Parameters:

• name – the name of the span to add to existing trace or create if no trace exists yet.

• autolog_frameworks – an optional list of mlflow supported frameworks to autolog

• evaluator – an optional function that takes the span created for the wrapped function and returns a dictionary of evaluation results. The evaluation results will be saved to the trace

• trace_evaluator – an optional function that takes the trace for this call stack and returns a dictionary of evaluation results. This evaluator will be triggered if the trace was started and finished by the add tracing decorator. The evaluation results will be saved to the trace

• eagerly_evaluate_streamed_results – optional boolean, defaults to true, this determines if all yielded values should be aggregated and set as outputs to a single span. This makes evaluation easier, but will impact performance if you expect a large number of streamed values. If set to false, each yielded value will generate a new span on the trace, which can be evaluated post-hoc. Inline evaluators won’t be executed. Each span will have a group_id set in their attributes to indicate that they are part of the same function call. Each span will have an index to indicate what order they arrived in.

• allow_tracing_evaluator – optional boolean, defaults to false. This determines if inline evaluators will be traced by mlflow autolog.

Returns:
A decorator that wraps the function to be traced.

 

domino.agents.tracing.init_tracing(autolog_frameworks: list[str] | None = None)

Initialize Mlflow autologging for various frameworks and sets the active experiment to enable tracing in production. This may be used to initialize logging and tracing for the Agent in dev and prod modes.

In prod mode, environment variables DOMINO_AGENT_IS_PROD, DOMINO_APP_ID must be set. Call init_tracing before your app starts up to start logging traces to Domino.

Parameters:
autolog_frameworks – list of frameworks to autolog

 

domino.agents.tracing.search_agent_traces(agent_id: str, agent_version: str | None = None, trace_name: str | None = None, start_time: datetime | None = None, end_time: datetime | None = None, page_token: str | None = None, max_results: int | None = None) → SearchTracesResponse

This allows searching for traces that have a certain name and returns a paginated response of trace summaries that include the spans that were requested.

Parameters:

• agent_id – string, the ID of the Agent to filter by

• agent_version – string, the version of the Agent to filter by, if not provided will search throuh all versions

• trace_name – the name of the traces to search for

• start_time – python datetime

• end_time – python datetime, defaults to now

• page_token – page token for pagination. You can use this to request the next page of results and may find a page_token in the response of the previous search_traces call.

• max_results – defaults to 100

Returns:
a token based pagination response that contains a list of trace summaries
data: list of TraceSummary page_token: the next page’s token

Return type:
SearchTracesResponse

 

domino.agents.tracing.search_traces(run_id: str, trace_name: str | None = None, start_time: datetime | None = None, end_time: datetime | None = None, page_token: str | None = None, max_results: int | None = None) → SearchTracesResponse

This allows searching for traces that have a certain name and returns a paginated response of trace summaries that inclued the spans that were requested.

Parameters:

• run_id – string, the ID of the development mode evaluation run to search for traces.

• trace_name – the name of the traces to search for

• start_time – python datetime

• end_time – python datetime, defaults to now

• page_token – page token for pagination. You can use this to request the next page of results and may find a page_token in the response of the previous search_traces call.

• max_results – defaults to 100

Returns:
a token based pagination response that contains a list of trace summaries
data: list of TraceSummary page_token: the next page’s token

Return type:
SearchTracesReponse

Modules

inittracing
tracing

Airflow

The python-domino client comes bundled with an Operator for use with Apache Airflow as an extra.

When installing the client from PyPI, add the airflow flag to extras:

pip install "dominodatalab[airflow]"

Similarly, when installing the client from GitHub, use the following command:

pip install -e git+https://github.com/dominodatalab/python-domino.git@1.0.6#egg="dominodatalab[airflow]"

See also example_airflow_dag.py for example code.

DominoOperator

from domino.airflow import DominoOperator

Allows a user to schedule Domino executions via Airflow. Follows the same function signature as domino.runs_start with two extra arguments:

• startup_delay: Optional[int] = 10 | Add a startup delay to your job, useful if you want to delay execution until after other work finishes.
• include_setup_log: Optional[bool] = True | Determine whether or not to publish the setup log of the job as the log prefix before stdout.

DominoSparkOperator

from domino.airflow import DominoSparkOperator

Allows a user to schedule Domino executions via the v4 API, which supports onDemandSparkClusters. Follows the same function signature as domino.job_start, with the addition of startup_delay from above.

Example

from domino import Domino

# By and large your commands will run against a single project,
# so you must specify the full project name
domino = Domino("chris/canon")

# List all runs in the project, most-recently queued first
all_runs = domino.runs_list()['data']

latest_100_runs = all_runs[0:100]

print(latest_100_runs)

# all runs have a commitId (the snapshot of the project when the
# run starts) and, if the run completed, an "outputCommitId"
# (the snapshot of the project after the run completed)
most_recent_run = all_runs[0]

commitId = most_recent_run['outputCommitId']

# list all the files in the output commit ID -- only showing the
# entries under the results directory.  If not provided, will
# list all files in the project.  Or you can say path=“/“ to
# list all files
files = domino.files_list(commitId, path='results/')['data']

for file in files:
print file['path'], '->', file['url']

print(files)

# Get the content (i.e. blob) for the file you're interested in.
# blobs_get returns a connection rather than the content, because
# the content can get quite large and it's up to you how you want
# to handle it
print(domino.blobs_get(files[0]['key']).read())

# Start a run of file main.py using the latest copy of that file
domino.runs_start(["main.py", "arg1", "arg2"])

# Start a "direct" command
domino.runs_start(["echo 'Hello, World!'"], isDirect=True)

# Start a run of a specific commit
domino.runs_start(["main.py"], commitId="aabbccddee")

Manual installation

Because python-domino ships with the DSE, normally you do not need to install it.
This section provides instructions for installing it in another environment or updating it to a newer version.

Starting from version 1.0.6, python-domino is available on PyPI as dominodatalab:

pip install dominodatalab

If you are adding install instructions for python-domino to your Domino Environment Dockerfile Instructions field, you must add RUN to the beginning:

RUN pip install dominodatalab

To install a specific version of the library from PyPI, such as 1.0.6:

pip install dominodatalab==1.0.6

To install a specific version of the library from GitHub, such as 1.0.6:

pip install https://github.com/dominodatalab/python-domino/archive/1.0.6.zip

License

This library is made available under the Apache 2.0 License. This is an open-source project of Domino Data Lab.