Deploy DCM Project

Overview

This sub-skill provides a safe, structured workflow for deploying DCM project changes to Snowflake. Deployment is a critical operation that can create, modify, or delete database objects.

⚠️ CRITICAL SAFETY RULES

1. •NEVER deploy without running plan first
2. •NEVER deploy without explicit user confirmation
3. •ALWAYS highlight destructive changes (DROP, data-affecting ALTER)
4. •ALWAYS recommend using deployment aliases

Prerequisites

Before deploying, ensure:

• • Analyze has been run successfully (no errors)
• • Plan has been run successfully (no PLAN_FAILED status)
• • User has reviewed the plan summary
• • User has given explicit confirmation to proceed

Deployment Workflow

Step 1: Verify Analyze Passed

If analyze hasn't been run recently:

snow dcm analyze <identifier> -c <connection> \
    --configuration <config> \
    --output-path ./out/analyze

⚠️ CRITICAL: Read and Parse the Output

You MUST read and parse out/analyze/analyze_output.json.

For detailed instructions, see: Parent SKILL.md - Critical: Reading Output Files

Do NOT proceed if there are errors.

Step 2: Run Plan (or Use Existing Output)

Check if plan output already exists:

• •If out/plan/plan_metadata.json exists and is current, read it instead of rerunning
• •Only rerun plan if explicitly requested by user or if definitions have changed

If plan needs to be run:

snow dcm plan <identifier> -c <connection> \
    --configuration <config> \
    --output-path ./out/plan

⚠️ CRITICAL: Read and Parse the Output

You MUST read and parse out/plan/plan_metadata.json.

For detailed instructions, see: Parent SKILL.md - Critical: Reading Output Files

Do NOT proceed to deployment without parsing the plan output.

Step 3: Present Plan Summary

Parse the plan output and present a clear summary:

╔══════════════════════════════════════════════════════════════╗
║                    📊 DEPLOYMENT PLAN                        ║
╠══════════════════════════════════════════════════════════════╣
║ Project:       <DATABASE.SCHEMA.PROJECT_NAME>                ║
║ Configuration: <config_name>                                 ║
║ Connection:    <connection_name>                             ║
╠══════════════════════════════════════════════════════════════╣
║                                                              ║
║ ✅ CREATE: X objects                                         ║
║    • TABLE: MY_DB.RAW.NEW_TABLE                              ║
║    • DYNAMIC_TABLE: MY_DB.ANALYTICS.SUMMARY                  ║
║    • TASK: MY_DB.ETL.TSK_DAILY_LOAD                          ║
║    • ROLE: MY_DB_READER                                      ║
║                                                              ║
║ ⚠️  ALTER: Y objects                                          ║
║    • TABLE: MY_DB.RAW.CUSTOMERS                              ║
║      └─ Add column: EMAIL (VARCHAR)                          ║
║      └─ Change column: AMOUNT (NUMBER → NUMBER(15,2))        ║
║    • TASK: MY_DB.ETL.TSK_PROCESS_ORDERS                      ║
║      └─ Task will be auto-suspended, modified, and resumed   ║
║                                                              ║
║ 🚨 DROP: Z objects                                           ║
║    • VIEW: MY_DB.SERVE.OLD_REPORT                            ║
║      ⚠️  WARNING: This object and its data will be deleted   ║
║                                                              ║
╚══════════════════════════════════════════════════════════════╝

Note on Tasks: When tasks are modified during deployment, DCM automatically suspends/resumes them - no manual intervention needed.

Step 4: Highlight Risky Changes

For each of these scenarios, add explicit warnings:

Step 5: Offer Preview (Optional)

Ask the user:

"Before deployment, would you like to preview data in any of the objects that will be modified or dropped?"

If yes:

snow dcm preview <identifier> -c <connection> \
    --object <fully.qualified.object.name> \
    --configuration <config> \
    --limit 10

This is especially useful for:

• •Verifying data before DROP operations
• •Checking current state before ALTER operations

Step 6: Request Explicit Confirmation

Present a clear confirmation prompt:

╔══════════════════════════════════════════════════════════════╗
║              ⚠️  DEPLOYMENT CONFIRMATION                     ║
╠══════════════════════════════════════════════════════════════╣
║                                                              ║
║ You are about to deploy changes to LIVE Snowflake            ║
║ infrastructure.                                              ║
║                                                              ║
║ • Creates: X objects                                         ║
║ • Alters:  Y objects                                         ║
║ • Drops:   Z objects                                         ║
║                                                              ║
║ Target Database: <database>                                  ║
║ Using Connection: <connection>                               ║
║                                                              ║
║ This action CANNOT be automatically undone.                  ║
║                                                              ║
╠══════════════════════════════════════════════════════════════╣
║                                                              ║
║ Type "yes" to confirm deployment, or "no" to cancel.         ║
║                                                              ║
╚══════════════════════════════════════════════════════════════╝

DO NOT PROCEED unless user explicitly confirms with "yes".

Step 7: Execute Deployment

Suggest an alias for tracking:

"Would you like to add a deployment alias for tracking? (e.g., 'v1.0', 'add-email-column', 'initial-setup')"

snow dcm deploy <identifier> -c <connection> \
    --configuration <config> \
    --alias "<user-provided-alias-or-timestamp>"

Step 8: Verify Deployment

After successful deployment:

1. •

   Confirm success - Report the deployment completed

2. •

   Check for tests - From the analyze output, determine if the project has data quality tests:

   code

   Do any definitions have DATA_METRIC_SCHEDULE or ATTACH DATA METRIC FUNCTION?
   

3. •

   Offer to run tests:

   "This project has data quality tests defined. Would you like to run them to verify the deployment?"

Step 8.5: Apply Unsupported Grants (If Any)

Check if the project has a post_deployment_grants.sql file (grants that DCM cannot apply):

If file exists, present to user:

"The following grants could not be applied by DCM and require manual execution:

• •[list grants from post_deployment_grants.sql]

These typically include: GRANT ... ON ACCOUNT, GRANT IMPORTED PRIVILEGES, warehouse grants to database roles, etc.

Would you like me to execute these grants now? (Requires appropriate privileges)"

⚠️ CHECKPOINT: Get explicit approval before running post-deployment grants.

Step 9: Run Tests (If Requested)

If the project has tests and user wants to run them:

1. •

   Refresh dynamic tables first (ensures fresh data):

   bash

   snow dcm refresh <identifier> -c <connection>
   

2. •

   Run tests:

   bash

   snow dcm test <identifier> -c <connection> \
       --output-path ./out/test
   

3. •

   Report test results - Summarize pass/fail status

Post-Deployment

Viewing Deployment History

snow dcm list-deployments <identifier> -c <connection>

Rolling Back

DCM doesn't have automatic rollback. To "undo" a deployment:

1. •Modify definition files to previous state
2. •Run plan to see what will change
3. •Deploy the reverted definitions

Recommendation: Keep definition files in version control (git) for easy rollback.

Error Handling

Plan Fails

If plan returns PLAN_FAILED:

• •Check the error field in the output
• •Common causes:
  • •Invalid SQL syntax
  • •Missing dependencies
  • •Permission issues
  • •Object state conflicts

Deployment Fails

If deploy fails:

• •Note the error message
• •Check Snowflake query history for failed DDL
• •Some objects may have been created/modified before failure
• •Run plan again to see remaining changes

Partial Deployment

If deployment partially succeeds:

• •Some objects were created/modified
• •Run plan to see what's left
• •Fix any errors and re-deploy

Quick Reference

Commands Used

Always Include

• •-c <connection>: Connection name
• •--configuration <config>: Configuration name (if using configurations)
• •--output-path <path>: For analyze/plan/test output

Recommended

• •--alias "<name>": For deploy, to track deployment history