atmos-workflows（atmos 工作流程）

Atmos Workflows

Atmos workflows combine multiple commands into executable units of work. They allow you to define multi-step sequences of Atmos commands and shell scripts, run them in order, and coordinate operations across multiple components and stacks.

What Workflows Are

A workflow is a named sequence of steps defined in a YAML file. Each step is either:

• An Atmos command (type atmos, the default) -- any Atmos CLI command without the atmos prefix
• A Shell command (type shell) -- any shell script or command

Workflows solve the problem of needing to run multiple Terraform operations in a specific order, such as provisioning a VPC before an EKS cluster, or deploying the same component across multiple environments sequentially.

You can use Atmos Custom Commands in workflows, and workflows in Custom Commands -- they are fully interoperable.

Workflow File Location and Discovery

Configuration

Workflow files are stored in the directory specified by workflows.base_path in atmos.yaml:

# atmos.yaml
workflows:
  # Can also be set using ATMOS_WORKFLOWS_BASE_PATH env var or --workflows-dir flag
  base_path: "stacks/workflows"

If base_path is set at the root level of atmos.yaml, the workflows.base_path is resolved relative to it.

Directory Structure

Organize workflow files by purpose, environment, or team:

stacks/workflows/
  deploy.yaml
  destroy.yaml
  validate.yaml
  networking.yaml
  eks.yaml
  maintenance/
    backup.yaml
    rotate-credentials.yaml

File names can be anything with a .yaml extension. Recommended naming conventions include grouping by environment (workflows-dev.yaml, workflows-prod.yaml), by service (workflows-eks.yaml), or by function (deploy.yaml, validate.yaml).

Auto-Discovery

Atmos can automatically discover workflow files. When you run atmos workflow <name> without the --file flag:

1. Atmos scans all YAML files in workflows.base_path
2. Finds workflows matching the given name
3. If exactly one file contains that workflow, runs it automatically
4. If multiple files contain a workflow with the same name, prompts for interactive selection (or errors in non-interactive mode with hints)

# Auto-discovery: Atmos finds the file automatically
atmos workflow eks-up --stack tenant1-ue2-dev

# Explicit file: always works, required when names conflict
atmos workflow eks-up -f networking --stack tenant1-ue2-dev

The --file value is relative to workflows.base_path. The .yaml extension is optional -- Atmos adds it automatically if not provided.

Workflow YAML Syntax

Every workflow file must have a top-level workflows: key containing a map of named workflows.

Basic Structure

workflows:
  workflow-name:
    description: "What this workflow does"
    stack: tenant1-ue2-dev        # Optional: default stack for all steps
    steps:
      - command: terraform plan vpc
        name: plan-vpc            # Optional: step identifier
        type: atmos               # Optional: default is atmos
        stack: tenant1-ue2-dev    # Optional: per-step stack override
        identity: superadmin      # Optional: per-step authentication

Step Types

Atmos steps (default type, type: atmos is implicit):

steps:
  - command: terraform plan vpc
  - command: terraform apply vpc -auto-approve
  - command: terraform deploy eks/cluster

The atmos prefix is automatically prepended. Write commands as you would after atmos on the command line.

Shell steps (require type: shell):

steps:
  - command: echo "Starting deployment..."
    type: shell
  - command: |
      echo "Running multi-line script"
      aws sts get-caller-identity
      kubectl get nodes
    type: shell

Shell commands can be any simple or complex script. Use YAML multiline strings for complex scripts.

Retry Configuration

Steps can be retried on failure:

steps:
  - command: terraform apply vpc -auto-approve
    retry:
      max_attempts: 3               # Maximum attempts (default: 1, meaning no retry)
      delay: 5s                     # Delay between retries (default: 5s)
      backoff_strategy: exponential  # constant, exponential, or linear
      initial_delay: 3s             # Initial delay for backoff
      random_jitter: 0.0            # Random jitter added to delay
      multiplier: 2                 # Multiplier for exponential backoff
      max_elapsed_time: 4m          # Total timeout for all retries (default: 30m)

Working Directory

Control where steps execute:

workflows:
  build-all:
    description: Build from repository root
    working_directory: !repo-root    # Workflow-level default
    steps:
      - command: make build
        type: shell
      - command: wget https://example.com/archive.tar.gz
        working_directory: /tmp      # Step-level override
        type: shell

Path resolution:

• Absolute paths are used as-is
• Relative paths resolve against base_path
• !repo-root resolves to the git repository root

Workflow Dependencies

Declare tool dependencies that are auto-installed before execution:

workflows:
  validate:
    description: Validate infrastructure
    dependencies:
      tools:
        tflint: "^0.54.0"
        checkov: "latest"
    steps:
      - command: terraform validate vpc
      - command: tflint --recursive
        type: shell

Dependencies support SemVer constraints: exact (1.10.3), pessimistic (~> 1.10.0), compatible (^1.10.0), or latest.

Working with Stacks in Workflows

The stack for Atmos-type steps can be specified in four ways (lowest to highest priority):

1. Inline in the Command

steps:
  - command: terraform plan vpc -s tenant1-ue2-dev

2. Workflow-Level Stack

workflows:
  deploy-vpc:
    stack: tenant1-ue2-dev
    steps:
      - command: terraform plan vpc
      - command: terraform apply vpc -auto-approve

3. Step-Level Stack

steps:
  - command: terraform plan vpc
    stack: tenant1-ue2-dev
  - command: terraform plan vpc
    stack: tenant1-ue2-staging

4. Command-Line Stack (Highest Priority)

atmos workflow deploy-vpc -f networking -s tenant1-ue2-dev

The command-line --stack flag overrides all other stack definitions.

Best practice: Create generic workflows without hardcoded stacks and provide the stack on the command line. This makes workflows reusable across environments.

Executing Workflows

Basic Execution

# With auto-discovery
atmos workflow eks-up --stack tenant1-ue2-dev

# With explicit file
atmos workflow eks-up -f networking --stack tenant1-ue2-dev

# Dry run (preview steps without executing)
atmos workflow eks-up -f networking --dry-run

Interactive Mode

Run atmos workflow without arguments to start an interactive TUI:

atmos workflow

Use arrow keys to navigate, / to search, and Enter to execute.

Starting from a Specific Step

Use --from-step to resume a workflow from a named step:

# Start from step3 (skip step1 and step2)
atmos workflow eks-up -f networking --from-step step3

This is useful for resuming after a failure. If a step does not have an explicit name, Atmos auto-generates names as step1, step2, step3, etc.

Failure Handling

When a step fails, Atmos displays:

• Which step failed
• The exact command that failed
• A ready-to-use command to resume from the failed step

Step 'step2' failed!

Command failed:
terraform plan vpc -s plat-ue2-staging

To resume the workflow from this step, run:
atmos workflow provision-vpcs -f networking --from-step step2

Using Authentication with Workflows

Per-Step Identity

Different steps can use different authentication identities:

workflows:
  deploy-with-auth:
    description: Deploy with different identities
    steps:
      - command: terraform apply vpc -s plat-ue2-prod
        identity: superadmin
        name: deploy-vpc
      - command: terraform apply app -s plat-ue2-prod
        identity: developer
        name: deploy-app
      - command: |
          aws sts get-caller-identity
        type: shell
        identity: superadmin
        name: verify-identity

Default Identity via Flag

Apply a default identity to all steps without their own:

atmos workflow deploy-all -f workflows --identity superadmin

Precedence: Step-level identity > --identity flag > no authentication.

Mixed Authentication

Steps without an identity field use ambient credentials. Steps with identity authenticate before execution. The --identity flag provides a default for steps without explicit identity.

workflows:
  mixed:
    steps:
      - command: terraform plan vpc -s dev    # Uses ambient or --identity flag
      - command: terraform apply vpc -s prod
        identity: superadmin                  # Always uses superadmin

Common Workflow Patterns

Deploy All Components in Order

workflows:
  deploy-all:
    description: Deploy all infrastructure in order
    steps:
      - command: terraform deploy vpc
      - command: terraform deploy eks/cluster
      - command: terraform deploy rds
      - command: terraform deploy app

atmos workflow deploy-all -f deploy -s plat-ue2-dev

Mixed Atmos and Shell Commands

workflows:
  deploy-and-verify:
    description: Deploy and verify infrastructure
    steps:
      - command: terraform deploy vpc
        name: deploy-vpc
      - command: kubectl get nodes
        type: shell
        name: verify

Nesting Workflows

Workflows can call other workflows via command: workflow <name>.

For additional patterns (multi-env deploy, teardown, validation pipelines), see references/workflow-syntax.md.

atmos workflow Command Reference

Syntax

atmos workflow [workflow_name] [flags]

Arguments

Flags

Listing Workflows

# List all available workflows
atmos list workflows

# Describe all workflows with details
atmos describe workflows

Best Practices

1. Create generic, stack-agnostic workflows. Do not hardcode stacks in workflow definitions. Instead, pass the stack on the command line with -s. This makes workflows reusable across all environments.
2. Name your steps. Use the name attribute so workflows can be resumed from a specific step with --from-step after a failure.
3. Organize files by purpose. Use separate files for deployment, teardown, validation, and maintenance workflows. Name files descriptively (e.g., networking.yaml, eks.yaml).
4. Use dry-run before executing. Always preview workflows with --dry-run before running them, especially in production environments.
5. Order steps by dependency. List steps in dependency order (VPC before EKS, database before application) since steps execute sequentially.
6. Use shell steps for verification. Insert shell steps between Atmos commands to verify infrastructure state before proceeding.
7. Use retry for flaky operations. Configure retry for steps that may fail transiently, such as cloud API calls that hit rate limits.
8. Leverage authentication per step. Use the identity field to switch credentials between steps that operate in different accounts or require different permission levels.
9. Use unique workflow names. Keep workflow names unique across all files to enable auto-discovery without ambiguity. If names must overlap, always use --file in scripts and CI/CD.
10. Use --from-step for recovery. When a workflow fails, fix the issue and resume from the failed step rather than re-running the entire workflow.

Additional Resources

• For the complete workflow YAML syntax reference, see references/workflow-syntax.md