1
0
mirror of https://github.com/oceanprotocol/docs.git synced 2024-11-26 19:49:26 +01:00
docs/developers/compute-to-data/compute-options.md
2023-06-28 15:47:46 +00:00

8.0 KiB

title section description
Compute Options developers Specification of compute options for assets in Ocean Protocol.

Compute Options

Compute Options

An asset categorized as a compute type incorporates additional attributes under the compute object.

These attributes are specifically relevant to assets that fall within the compute category and are not required for assets classified under the access type. However, if an asset is designated as compute, it is essential to include these attributes to provide comprehensive information about the compute service associated with the asset.

AttributeTypeDescription
allowRawAlgorithm*booleanIf true, any passed raw text will be allowed to run. Useful for an algorithm drag & drop use case, but increases risk of data escape through malicious user input. Should be false by default in all implementations.
allowNetworkAccess*booleanIf true, the algorithm job will have network access.
publisherTrustedAlgorithmPublishers*Array of stringIf not defined, then any published algorithm is allowed. If empty array, then no algorithm is allowed. If not empty any algo published by the defined publishers is allowed.
publisherTrustedAlgorithms*Array of publisherTrustedAlgorithmsIf not defined, then any published algorithm is allowed. If empty array, then no algorithm is allowed. Otherwise only the algorithms defined in the array are allowed. (see below).

* Required

Trusted Algorithms

The publisherTrustedAlgorithms is an array of objects that specifies algorithm permissions. It controls which algorithms can be used for computation. If not defined, any published algorithm is allowed. If the array is empty, no algorithms are allowed. However, if the array is not empty, only algorithms published by the defined publishers are permitted.

The structure of each object within the publisherTrustedAlgorithms array is as follows:

AttributeTypeDescription
didstringThe DID of the algorithm which is trusted by the publisher.
filesChecksumstringHash of algorithm's files (as string).
containerSectionChecksumstringHash of algorithm's image details (as string).

To produce filesChecksum, call the Provider FileInfoEndpoint with parameter withChecksum = True. If the algorithm has multiple files, filesChecksum is a concatenated string of all files checksums (ie: checksumFile1+checksumFile2 , etc)

To produce containerSectionChecksum:

{% code overflow="wrap" %}

sha256(algorithm_ddo.metadata.algorithm.container.entrypoint + algorithm_ddo.metadata.algorithm.container.checksum);

{% endcode %}

Compute Options Example

Example:

{
  "services": [
    {
      "id": "1",
      "type": "access",
      "files": "0x044736da6dae39889ff570c34540f24e5e084f...",
      "name": "Download service",
      "description": "Download service",
      "datatokenAddress": "0x123",
      "serviceEndpoint": "https://myprovider.com",
      "timeout": 0
    },
    {
      "id": "2",
      "type": "compute",
      "files": "0x6dd05e0edb460623c843a263291ebe757c1eb3...",
      "name": "Compute service",
      "description": "Compute service",
      "datatokenAddress": "0x124",
      "serviceEndpoint": "https://myprovider.com",
      "timeout": 0,
      "compute": {
        "allowRawAlgorithm": false,
        "allowNetworkAccess": true,
        "publisherTrustedAlgorithmPublishers": ["0x234", "0x235"],
        "publisherTrustedAlgorithms": [
          {
            "did": "did:op:123",
            "filesChecksum": "100",
            "containerSectionChecksum": "200"
          },
          {
            "did": "did:op:124",
            "filesChecksum": "110",
            "containerSectionChecksum": "210"
          }
        ]
      }
    }
  ]
}

Consumer Parameters

Sometimes, the asset needs additional input data before downloading or running a Compute-to-Data job. Examples:

  • The publisher needs to know the sampling interval before the buyer downloads it. Suppose the dataset URL is https://example.com/mydata. The publisher defines a field called sampling and asks the buyer to enter a value. This parameter is then added to the URL of the published dataset as query parameters: https://example.com/mydata?sampling=10.
  • An algorithm that needs to know the number of iterations it should perform. In this case, the algorithm publisher defines a field called iterations. The buyer needs to enter a value for the iterations parameter. Later, this value is stored in a specific location in the Compute-to-Data pod for the algorithm to read and use it.

The consumerParameters is an array of objects. Each object defines a field and has the following structure:

AttributeTypeDescription
name*stringThe parameter name (this is sent as HTTP param or key towards algo)
type*stringThe field type (text, number, boolean, select)
label*stringThe field label which is displayed
required*booleanIf customer input for this field is mandatory.
description*stringThe field description.
default*string, number, or booleanThe field default value. For select types, string key of default option.
optionsArray of optionFor select types, a list of options.

* Required

Each option is an object containing a single key: value pair where the key is the option name, and the value is the option value.

Consumer Parameters Example
[
  {
    "name": "hometown",
    "type": "text",
    "label": "Hometown",
    "required": true,
    "description": "What is your hometown?",
    "default": "Nowhere"
  },
  {
    "name": "age",
    "type": "number",
    "label": "Age",
    "required": false,
    "description": "Please fill your age",
    "default": 0
  },
  {
    "name": "developer",
    "type": "boolean",
    "label": "Developer",
    "required": false,
    "description": "Are you a developer?",
    "default": false
  },
  {
    "name": "languagePreference",
    "type": "select",
    "label": "Language",
    "required": false,
    "description": "Do you like NodeJs or Python",
    "default": "nodejs",
    "options": [
      {
        "nodejs": "I love NodeJs"
      },
      {
        "python": "I love Python"
      }
    ]
  }
]

Algorithms will have access to a JSON file located at /data/inputs/algoCustomData.json, which contains the keys/values input data required. Example:

Key Value Example
{ 
    "hometown": "São Paulo",
    "age": 10, 
    "developer": true, 
    "languagePreference": "nodejs" 
}