rdev/cookbooks/scripts/lib/checkpoint.sh

#!/bin/bash
# Checkpoint utilities for tree-based cookbook execution
#
# Usage:
#   source "$(dirname "${BASH_SOURCE[0]}")/checkpoint.sh"
#
# Provides:
#   - checkpoint_init() - Initialize a new checkpoint
#   - checkpoint_load() - Load existing checkpoint
#   - checkpoint_save() - Persist checkpoint to disk
#   - checkpoint_step_start() - Mark step as started
#   - checkpoint_step_complete() - Mark step as completed with output
#   - checkpoint_step_fail() - Mark step as failed with error
#   - checkpoint_get_output() - Get output from a completed step
#   - checkpoint_status() - Get overall checkpoint status
#   - checkpoint_list() - List all checkpoints
#   - checkpoint_delete() - Delete a checkpoint

# Checkpoint directory
CHECKPOINT_DIR="${CHECKPOINT_DIR:-$(dirname "${BASH_SOURCE[0]}")/../../.checkpoints}"

# Ensure checkpoint directory exists
_checkpoint_ensure_dir() {
    mkdir -p "$CHECKPOINT_DIR"
}

# Get checkpoint file path for a tree
_checkpoint_path() {
    local tree_name="$1"
    echo "$CHECKPOINT_DIR/${tree_name}.json"
}

# Initialize a new checkpoint
# Arguments: tree_name vars_json
# Returns: run_id
# Example: run_id=$(checkpoint_init "landing-page" '{"project_name": "test"}')
checkpoint_init() {
    local tree_name="$1"
    local vars_json="${2:-{}}"

    _checkpoint_ensure_dir

    local run_id="${tree_name}-$(date +%s)"
    local now
    now=$(date -u +"%Y-%m-%dT%H:%M:%SZ")

    local checkpoint
    checkpoint=$(jq -n \
        --arg tree "$tree_name" \
        --arg run_id "$run_id" \
        --arg started "$now" \
        --argjson vars "$vars_json" \
        '{
            tree: $tree,
            run_id: $run_id,
            status: "pending",
            vars: $vars,
            steps: {},
            started_at: $started,
            last_completed_step: null
        }')

    echo "$checkpoint" > "$(_checkpoint_path "$tree_name")"
    echo "$run_id"
}

# Load existing checkpoint
# Arguments: tree_name
# Returns: checkpoint JSON on stdout, exit 1 if not found
# Example: checkpoint=$(checkpoint_load "landing-page")
checkpoint_load() {
    local tree_name="$1"
    local path
    path="$(_checkpoint_path "$tree_name")"

    if [[ ! -f "$path" ]]; then
        return 1
    fi

    cat "$path"
}

# Save checkpoint to disk
# Arguments: tree_name checkpoint_json
# Example: checkpoint_save "landing-page" "$checkpoint"
checkpoint_save() {
    local tree_name="$1"
    local checkpoint="$2"

    _checkpoint_ensure_dir
    echo "$checkpoint" > "$(_checkpoint_path "$tree_name")"
}

# Mark step as started
# Arguments: tree_name step_name
# Example: checkpoint_step_start "landing-page" "create-project"
checkpoint_step_start() {
    local tree_name="$1"
    local step_name="$2"

    local checkpoint
    checkpoint=$(checkpoint_load "$tree_name") || return 1

    local now
    now=$(date -u +"%Y-%m-%dT%H:%M:%SZ")

    checkpoint=$(echo "$checkpoint" | jq \
        --arg step "$step_name" \
        --arg started "$now" \
        '.steps[$step] = {status: "running", started_at: $started} | .status = "partial"')

    checkpoint_save "$tree_name" "$checkpoint"
}

# Mark step as completed with output
# Arguments: tree_name step_name output_json
# Example: checkpoint_step_complete "landing-page" "create-project" '{"project_id": "test"}'
checkpoint_step_complete() {
    local tree_name="$1"
    local step_name="$2"
    local output_json="${3:-{}}"

    local checkpoint
    checkpoint=$(checkpoint_load "$tree_name") || return 1

    local now
    now=$(date -u +"%Y-%m-%dT%H:%M:%SZ")

    checkpoint=$(echo "$checkpoint" | jq \
        --arg step "$step_name" \
        --arg completed "$now" \
        --argjson output "$output_json" \
        '.steps[$step].status = "completed" |
         .steps[$step].completed_at = $completed |
         .steps[$step].output = $output |
         .last_completed_step = $step')

    checkpoint_save "$tree_name" "$checkpoint"
}

# Mark step as failed with error
# Arguments: tree_name step_name error_message
# Example: checkpoint_step_fail "landing-page" "wait-pipeline" "Pipeline failed"
checkpoint_step_fail() {
    local tree_name="$1"
    local step_name="$2"
    local error_msg="$3"

    local checkpoint
    checkpoint=$(checkpoint_load "$tree_name") || return 1

    local now
    now=$(date -u +"%Y-%m-%dT%H:%M:%SZ")

    checkpoint=$(echo "$checkpoint" | jq \
        --arg step "$step_name" \
        --arg completed "$now" \
        --arg error "$error_msg" \
        '.steps[$step].status = "failed" |
         .steps[$step].completed_at = $completed |
         .steps[$step].error = $error |
         .status = "failed"')

    checkpoint_save "$tree_name" "$checkpoint"
}

# Get output from a completed step
# Arguments: tree_name step_name [output_key]
# Returns: output value or full output JSON
# Example: project_id=$(checkpoint_get_output "landing-page" "create-project" "project_id")
checkpoint_get_output() {
    local tree_name="$1"
    local step_name="$2"
    local output_key="${3:-}"

    local checkpoint
    checkpoint=$(checkpoint_load "$tree_name") || return 1

    if [[ -n "$output_key" ]]; then
        echo "$checkpoint" | jq -r ".steps[\"$step_name\"].output[\"$output_key\"] // empty"
    else
        echo "$checkpoint" | jq ".steps[\"$step_name\"].output // {}"
    fi
}

# Get variable from checkpoint
# Arguments: tree_name var_name
# Returns: variable value
# Example: project_name=$(checkpoint_get_var "landing-page" "project_name")
checkpoint_get_var() {
    local tree_name="$1"
    local var_name="$2"

    local checkpoint
    checkpoint=$(checkpoint_load "$tree_name") || return 1

    echo "$checkpoint" | jq -r ".vars[\"$var_name\"] // empty"
}

# Get overall checkpoint status
# Arguments: tree_name
# Returns: status string (pending, partial, completed, failed)
# Example: status=$(checkpoint_status "landing-page")
checkpoint_status() {
    local tree_name="$1"

    local checkpoint
    checkpoint=$(checkpoint_load "$tree_name") || {
        echo "none"
        return 0
    }

    echo "$checkpoint" | jq -r '.status // "unknown"'
}

# Get detailed checkpoint status (for display)
# Arguments: tree_name
# Returns: formatted status output
checkpoint_status_detail() {
    local tree_name="$1"

    local checkpoint
    checkpoint=$(checkpoint_load "$tree_name") || {
        echo "No checkpoint found for tree: $tree_name"
        return 1
    }

    echo "$checkpoint" | jq -r '
        "Tree: \(.tree)",
        "Run ID: \(.run_id)",
        "Status: \(.status)",
        "Started: \(.started_at)",
        "",
        "Variables:",
        (.vars | to_entries | .[] | "  \(.key): \(.value)"),
        "",
        "Steps:",
        (.steps | to_entries | sort_by(.value.started_at // "") | .[] |
            if .value.status == "completed" then
                "  \u001b[32m\u2713\u001b[0m \(.key): completed"
            elif .value.status == "failed" then
                "  \u001b[31m\u2717\u001b[0m \(.key): failed - \(.value.error)"
            elif .value.status == "running" then
                "  \u001b[33m\u25d0\u001b[0m \(.key): running..."
            else
                "  \u25cb \(.key): \(.value.status)"
            end
        ),
        "",
        "Last completed: \(.last_completed_step // "none")"
    '
}

# Get list of completed steps
# Arguments: tree_name
# Returns: newline-separated list of completed step names
checkpoint_completed_steps() {
    local tree_name="$1"

    local checkpoint
    checkpoint=$(checkpoint_load "$tree_name") || return 1

    echo "$checkpoint" | jq -r '.steps | to_entries | .[] | select(.value.status == "completed") | .key'
}

# Get list of failed steps
# Arguments: tree_name
# Returns: newline-separated list of failed step names
checkpoint_failed_steps() {
    local tree_name="$1"

    local checkpoint
    checkpoint=$(checkpoint_load "$tree_name") || return 1

    echo "$checkpoint" | jq -r '.steps | to_entries | .[] | select(.value.status == "failed") | .key'
}

# List all checkpoints
# Returns: newline-separated list of tree names with checkpoints
# Example: trees=$(checkpoint_list)
checkpoint_list() {
    _checkpoint_ensure_dir

    for f in "$CHECKPOINT_DIR"/*.json; do
        [[ -e "$f" ]] || continue
        basename "$f" .json
    done
}

# Delete a checkpoint
# Arguments: tree_name
# Example: checkpoint_delete "landing-page"
checkpoint_delete() {
    local tree_name="$1"
    local path
    path="$(_checkpoint_path "$tree_name")"

    if [[ -f "$path" ]]; then
        rm "$path"
        return 0
    fi
    return 1
}

# Mark entire tree as completed
# Arguments: tree_name
checkpoint_mark_completed() {
    local tree_name="$1"

    local checkpoint
    checkpoint=$(checkpoint_load "$tree_name") || return 1

    local now
    now=$(date -u +"%Y-%m-%dT%H:%M:%SZ")

    checkpoint=$(echo "$checkpoint" | jq \
        --arg completed "$now" \
        '.status = "completed" | .completed_at = $completed')

    checkpoint_save "$tree_name" "$checkpoint"
}

# Update vars in checkpoint (for resume with different vars)
# Arguments: tree_name vars_json
checkpoint_update_vars() {
    local tree_name="$1"
    local vars_json="$2"

    local checkpoint
    checkpoint=$(checkpoint_load "$tree_name") || return 1

    checkpoint=$(echo "$checkpoint" | jq \
        --argjson new_vars "$vars_json" \
        '.vars = (.vars + $new_vars)')

    checkpoint_save "$tree_name" "$checkpoint"
}