With a little help from my friends…

Consensus building in Computes.

Photo by Sidharth Bhatia on Unsplash

Sometimes you want to make sure your tasks create a consistent result before trusting the outcome. We’ve added a simple consensus mechanism in Computes. When used, you can set a threshold of how many matches are required. Let’s look at an example task with a consensus threshold of 2. This means 2 tasks would have to agree on the results before they become visible in your dataset.

Consensus Task 1

{
  "input": {
    "dataset": 2
  },
  "taskDefinition": {
    "runner": {
      "type": "docker-json-runner",
      "manifest": {
        "*": {
          "image": "computes/fibonacci-sum-split:latest"
        }
      }
    },
    "result": {
      "action": "set",
      "destination": {
        "dataset": { "/": "zdpuAyMuiJxSwrpuDuwdXckFpjDNrzURXxtsayfFhxuiJ9ZEt" },
        "path": "split/results"
      },
      "consensus": {
        "threshold": 2
      }
    }
  },
  "status": {
    "/": "zdpuAwmC7rqGDaLm8rPhFgNbBoSqUXMGgi478TLuDTGohhP3q"
  },
  "id": 1
}

If I run this task and check for the results, they will not be visible, even though they are stored internally in the Computes network.

$ computes-cli dataset dumplatest \
    zdpuAo2koQJUuh4ixPiJ9jfUicCWPTEL2Rhg7r3WUYuGsbEt2
{
  "__meta__": {
    "computes": {
      "seed": "d5f0f8c7-46c0-434e-a676-5699d7287da9",
      "tasks": {
        "related": [
          "zdpuAtRc9P6piSvmJbxfxrSA24P6ATbsXzCFTFzPF3DmkwjVT"
        ]
      }
    }
  },
  "split": {}
}

Consensus Task 2

{
  "input": {
    "dataset": 2
  },
  "taskDefinition": {
    "runner": {
      "type": "docker-json-runner",
      "manifest": {
        "*": {
          "image": "computes/fibonacci-sum-split:latest"
        }
      }
    },
    "result": {
      "action": "set",
      "destination": {
        "dataset": { "/": "zdpuAyMuiJxSwrpuDuwdXckFpjDNrzURXxtsayfFhxuiJ9ZEt" },
        "path": "split/results"
      },
      "consensus": {
        "threshold": 2
      }
    }
  },
  "status": {
    "/": "zdpuAwmC7rqGDaLm8rPhFgNbBoSqUXMGgi478TLuDTGohhP3q"
  },
  "id": 2
}

If we look at task #2, there is only one small difference, which is the id property. This isn’t an official property of Computes, but allows the task to be different enough to generate a new task hash. Computes requires each attempt at consensus to be a different task, running the same task over and over again will not create consensus.

After running task #2, let’s see what the results show now.

$ computes-cli dataset dumplatest \
    zdpuAo2koQJUuh4ixPiJ9jfUicCWPTEL2Rhg7r3WUYuGsbEt2
{
  "__meta__": {
    "computes": {
      "seed": "d5f0f8c7-46c0-434e-a676-5699d7287da9",
      "tasks": {
        "related": [
          "zdpuAtRc9P6piSvmJbxfxrSA24P6ATbsXzCFTFzPF3DmkwjVT",
          "zdpuB15gD5xC8gf6NVEWCnYXosWSagzhYqWYCMT4NUQh1mQDM"
        ]
      }
    }
  },
  "split": {
    "results": [
      1,
      2
    ]
  }
}

What’s next?

We’d like to give nodes the ability to sign and verify results, offering some level of trust across the Computes network in a potentially trustless environment.