terraform-provider-iterative

terraform-provider-iterative copied to clipboard

From users feedback

tear down the created resources one relies on a brittle condition based on grepping some text in the logs

Its far from ideal, specially having status not totally working or not working as expected #388

This probably belongs to a separate epic/milestone titled “using task from CI/CD systems”; not saying it's not equally important, though.

While it's still far from ideal, have you tried using terraform console or terraform show and jq instead?

Example

resource "iterative_task" "example" {
  ...
}

terraform console

terraform console <<< 'iterative_task.example.status["succeeded"]'

terraform show --json and jq

terraform show --json | jq --exit-status '
  .values.root_module.resources[] |
  select(.address == "iterative_task.example") |
  .values.status.succeeded
'

I tried jq also but is not very friendly. However terraform console looks very interesting

Both console and jq would solve the "brittle" problem. Should be solved by some docs?

@iterative/cml definitely it's worth to use console and we could this just convert it into a doc issue. Tips:

• note the shell: bash due to the redirection
• resource "iterative_task" "train resource name is the one used in the check if terraform console <<< 'iterative_task.train.status["succeeded"]'; then

Here is a proven workflow. In my workflow the output models folder and report.md is created with the train.py using cml to do a live metrics alike

name: train-my-model

on: [push]

jobs:
  train-tpi:
    runs-on: [ubuntu-latest]

    steps:
    - uses: actions/checkout@v2
      with:
        fetch-depth: 0

    - uses: iterative/setup-cml@v1

    - name: tpi
      env:
        REPO_TOKEN: ${{ secrets.REPO_TOKEN }}
        GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        AZURE_CLIENT_ID: ${{ secrets.AZURE_CLIENT_ID }}
        AZURE_CLIENT_SECRET: ${{ secrets.AZURE_CLIENT_SECRET }}
        AZURE_SUBSCRIPTION_ID: ${{ secrets.AZURE_SUBSCRIPTION_ID }}
        AZURE_TENANT_ID: ${{ secrets.AZURE_TENANT_ID }}

      shell: bash
      run: |
        cat <<EOF > main.tf
        terraform {
          required_providers {
            iterative = {
              source = "iterative/iterative",
            }
          }
        }

        provider "iterative" {}

        resource "iterative_task" "train" {
          cloud     = "az"
          machine   = "Standard_D2S_v3"
          region    = "us-west"
          spot      = 0

          workdir {
            input = "."
            output = "."
          }
          
          environment = {
            EPOCHS = 1
          }

          script = <<-END
            #!/bin/bash
            sudo apt update
            sudo apt-get install -y software-properties-common build-essential python3-pip
            pip3 install -r requirements.txt
            python3 train.py
          END
        }
        EOF

        terraform init 
        terraform apply --auto-approve

        if terraform console <<< 'iterative_task.train.status["succeeded"]'; then
          echo 'Destroying...'
          terraform destroy --auto-approve
 
          cml send-github-check --token=$GITHUB_TOKEN --conclusion=success --title='CML report' report.md

          cml pr --md output/* >> report.md
          cml send-comment --update report.md
          cml send-comment --update --pr --commit-sha HEAD report.md
        else
          echo 'Creating report...'
          echo 'In progress...' > report.md
          cml send-github-check --token=$GITHUB_TOKEN --conclusion=neutral --title='CML report' report.md
        fi

note the shell: bash due to the redirection

GNU Bash version

if terraform console <<< 'iterative_task.example.status["succeeded"]'; then
  ...
fi

POSIX compliant version ™

if echo 'iterative_task.example.status["succeeded"]' | terraform console; then
  ...
fi

Here you're a tentative solution to the https://github.com/iterative/terraform-provider-iterative/issues/357 potential XY problem.

It would be nice to have an easier way of checking if a task has either failed or succeeded, but I'm afraid it would either involve exposing a redundant attribute (e.g. completed: boolean) or finding another way of invoking task from CI/CD systems.

Succeeded + failed check

if terraform console <<< 'try(iterative_task.example.status["succeeded"], 0) + try(iterative_task.example.status["failed"], 0)' | grep --quiet --invert-match 0; then
  echo the task has either succeeded or failed, destroying the resources...
fi

Loop–based alternative (8 bytes less)

if terraform console <<< 'sum([for status in ["succeeded", "failed"] : try(iterative_task.example.status[status], 0)])' | grep --quiet --invert-match 0; then
  echo the task has either succeeded or failed, destroying the resources...
fi

I don't quite grok if succeeded; then destroy; report pass; else report running; fi

Surely it should be report running; while !succeeded; do sleep; done; destroy; report pass?

Are you referring to https://github.com/iterative/terraform-provider-iterative/issues/389#issuecomment-1047102568? 🤔 If so, you're probably missing a detail: task was modified (https://github.com/iterative/terraform-provider-iterative/pull/339) to restart workflows when machines shut down. Intuitive as it gets.

I just realised :nauseated_face:

🦄 🤢

No longer a valid use case, use leo read --status or propose new output formats for the standalone command-line tool