foundry-compute-modules 0.1.0

The official Python library for creating Compute Modules

Compute Module Lib

[!WARNING] This SDK is incubating and subject to change.

An open-source python library for compute modules for performing tasks like service discovery, getting a token, external source credentials, etc

Functions Mode

Sources can be used to store secrets for use within a Compute Module, they prevent you from having to put secrets in your container or in plaintext in the job specification. Retrieving a source credential using this library is simple, if you are in Functions Mode they are passed to the context

Basic usage

Option 1 - @function annotation on each endpoint

# app.py
from compute_modules.annotations import function

@function
def add(context, event) -> int:
    return event["x"] + event["y"]

@function
def get_sources(context, event) -> List[str]:
    return context["sources"].keys()

Option 2 - Explicitly register function(s) & start the Compute Module

You can either add a single function via add_function or several at once with add_functions.

# functions/add.py
def add(context, event) -> int:
    return event["x"] + event["y"]

# functions/hello.py
def hello(context, event) -> str:
    return "Hello " + event["x"] + "!"

# app.py
from compute_modules import add_functions, start_compute_module

from functions.add import add
from functions.hello import hello

if __name__ == "__main__":
    add_functions(
        hello,
        add,
    )
    start_compute_module()

Advanced Usage - automatic function discovery

This library includes functionality that will inspect the functions registered for the Compute Module, inspect the input/output types of those functions, and then convert those to FunctionSpecs that can be imported as a Foundry Function without any modifications needed. Below are some considerations to ensure this feature works as expected.

1. The Input class must be a complex type

Foundry function specs require the input type of a Function to be a complex type. If your function takes only a single primitive type as input, make sure to wrap that param in a complex type to have your function schema inferred properly.

2. Input type definition

✅ TypedDict as input type

from typing import TypedDict


class HelloInput(TypedDict):
    planet: str

def hello(context, event: HelloInput) -> str:
    return "Hello " + event["planet"] + "!"

✅ dataclass as input type

from dataclasses import dataclass


@dataclass
class TypedInput:
    bytes_value: bytes
    bool_value: bool
    date_value: datetime.date
    decimal_value: decimal.Decimal
    float_value: float
    int_value: int
    str_value: str
    datetime_value: datetime.datetime
    other_date_value: datetime.datetime

def typed_function(context, event: TypedInput) -> str:
    diff = event.other_date_value - event.datetime_value
    return f"The diff between dates provided is {diff}"

✅ regular class with both class AND constructor type hints

class GoodExample:
    some_flag: bool
    some_value: int

    def __init__(self, some_flag: bool, some_value: int) -> None:
        self.some_flag = some_flag
        self.some_value = some_value

def typed_function(context, event: GoodExample) -> int:
    return return event.some_value

❌ AVOID python class with no class type hints

# This will raise an exception
class BadClassNoTypeHints:
    def __init__(self, arg1: str, arg2: int):
        ...

❌ AVOID python class with no constructor type hints

# This will raise an exception
class BadClassNoInitHints:
    arg1: str
    arg2: int

    def __init__(self, arg1, arg2):
        ...

❌ AVOID python class with args in constructor

# This will raise an exception
class BadClassArgsInit:
    arg1: str
    arg2: int

    def __init__(self, arg1: str, arg2: int, *args):
        ...

❌ AVOID python class with kwargs in constructor

# This will raise an exception
class BadClassKwargsInit:
    arg1: str
    arg2: int

    def __init__(self, arg1: str, arg2: int, **kwargs):
        ...

3. Serialization/De-serialization of various types

Python Type	Foundry Type	Serialized over HTTP as
int	Integer	int
str	Byte	string
bool	Boolean	boolean
bytes	Binary	string
datetime.date	Date	string
datetime.datetime	Timestamp	int (Unix timestamp)
decimal.Decimal	Decimal	string
float	Float	float
list	Array	array
set	Array	array
dict	Map	JSON
class/TypedDict	Struct	JSON

Pipelines Mode

Retrieving source credentials

Sources allow you to store secrets securely for use within a Compute Module, eliminating the need to include secrets in your container or in plaintext within the job specification. Retrieving a source credential using this library is straightforward:

from compute_modules.sources import get_sources, get_source_secret

# retrive a dict with all sources
sources = get_sources()

# retrive the credentials of a specific source 
my_creds = get_source_secret("mySourceApiName", "MyCredential")

Retrieving pipeline resources

The SDK offers a convenient method for retrieving information on the resources configured for your pipeline module. This allows you to obtain the rid (& branch, if present) of a Foundry resource via the alias provided for that resource in the Configure tab of your compute module.

from compute_modules.resources import PipelineResource, get_pipeline_resources

resources: dict[str, PipelineResource] = get_pipeline_resources()
print(f"My reource's rid is: {resources['your-alias-name'].rid}")

Retriving pipeline token

To obtain an auth token for interacting with Foundry resources in Pipeline mode use the following function:

from compute_modules.auth import retrieve_pipeline_token
import requests

pipeline_token = retrieve_pipeline_token()
requests.post(..., headers={"Authorization": f"Bearer {pipeline_token}")

Application's permissions/ Third Party App

If you have configured your Compute Module (CM) to use Application's permissions, your application will use a service user for permissions instead of relying on the user's permissions. This configuration requires you to obtain the client ID and credentials to grant permission to the service token. This library facilitates this process:

from compute_modules.auth import retrieve_third_party_id_and_creds, oauth

CLIENT_ID, CLIENT_CREDS = retrieve_third_party_id_and_creds()

# get a scoped token for your 3pa
HOSTNAME = "myenvironment.palantirfoundry.com"
access_token = oauth(HOSTNAME, ["api:datasets-read"])

Retrieving Arguments

This SDK provides utilities for retrieving arguments passed into the compute module. There are two different functions available: get_raw_arguments and get_parsed_arguments. Below is an example showing the difference between the two.

For a Compute Module with the following arguments configured:

If we log the result of both as such:

# app.py

import logging as log
from compute_modules.annotations import function
from compute_modules.arguments import get_raw_arguments, get_parsed_arguments

log.basicConfig(level=log.INFO)

@function
def hello(context, event) -> str:
    raw_args = get_raw_arguments()
    parsed_args = get_parsed_arguments()
    log.info(f"raw_args: {raw_args}")
    log.info(f"parsed_args: {parsed_args}")
    ...

We would then receive the following log output:

INFO: raw_args: ['--test', 'hello', '--another-param', 'world']
INFO: parsed_args: Namespace(test='hello' , another_param= 'world' )

Logging

To ensure your logs are emitted to properly we recommend you use the get_logger utility function provided by the SDK. This returns a normal logging.Logger instance so once you have the logger, you can use it as a drop-in replacement for logging.getLogger.

from compute_modules.logging import get_logger

logger = get_logger(__name__)
logger.setLevel(logging.INFO)

logger.debug("Can't see me")
logger.info("Peekaboo!")
logger.warning("Peekaboo!")
logger.error("Peekaboo!")
logger.critical("Peekaboo!")

Surfacing logs from the compute_modules library

By default, the logs emitted from within the compute_modules library have a level of ERROR, meaning only error- or critical-level logs will be emitted. If for any reason you want to see other logs being emitted from within compute_modules you can use the set_compute_modules_log_level function.

from compute_modules.logging import set_compute_modules_log_level

set_compute_modules_log_level(logging.DEBUG)