Alibaba Cloud Flink Instance Manage

Operate Alibaba Cloud Flink VVP resources with a strict create/query scope through one wrapper script.

Scope and Entrypoint

• Always run operations through:

  python scripts/instance_ops.py <command> [options]
  

• Allowed commands: create, create_namespace, describe, describe_regions, describe_zones, describe_namespaces, list_tags
• Out of scope: update/delete, Flink SQL/job runtime operations, and non-Flink services

Trigger Rules

Use this skill when prompts are about Flink instance/namespace lifecycle operations.

• Positive intent examples:
  • "Create a Flink instance in cn-beijing"
  • "List Flink instances and status"
  • "Describe namespaces for instance f-cn-xxx"
  • "查询 Flink 实例标签"
  • "Flink 可用区有哪些"
• Negative intent examples:
  • ECS/Kafka/OSS/DataWorks operations
  • Generic questions (weather, translation, etc.)
  • Flink SQL / Flink job authoring or runtime tuning
• Ambiguous prompts:
  • Ask one clarification question: instance/namespace management vs SQL/job operations.

Intent to Command Mapping

| User intent | Command | |---|---| | Query all instances in a region | describe --region_id <REGION> | | Create instance | create ... --confirm | | Query namespaces under an instance | describe_namespaces --region_id <REGION> --instance_id <ID> | | Create namespace | create_namespace ... --confirm | | Query supported regions/zones | describe_regions / describe_zones --region_id <REGION> | | Query tags | list_tags --region_id <REGION> --resource_type <TYPE> [--resource_ids ...] |

Operating Rules

1. Confirmation is mandatory for create commands
   • create and create_namespace must include --confirm.
2. Verify create results with read-back
   • Do not conclude success from create response alone.
3. Retry policy is strict
   • Maximum 2 attempts for the same command (initial + one corrected retry).
4. No automatic operation switching
   • If an operation fails, do not switch to a different operation without user approval.
5. Lifecycle target lock
   • In create -> create_namespace flow, namespace must target the same newly created InstanceId unless user approves fallback.
6. Namespace pre-check is required
   • Before create_namespace, check instance status/resources and existing namespace allocation.
7. No secret exposure
   • Do not output or request plaintext AK/SK. Use default credential chain guidance.
8. Do not invent parameters
   • Never fabricate VPC/VSwitch/instance IDs.
9. Keep auditable confirmation evidence
   • Lifecycle outputs must contain SafetyCheckRequired or explicit --confirm evidence.
10. No partial-completion claims for lifecycle flows

• For flows requiring both create and create_namespace, overall status can be completed only when both create operations succeed.

11. No automatic capacity scaling

• If create_namespace fails due to insufficient resources, report it clearly and ask user to manually scale resources outside this skill scope.

Execution Protocol

Step 1: Classify request

• In-scope create/query for Flink instance/namespace/tag/region/zone -> continue.
• Out-of-scope or non-Flink -> reject or route with explanation.

Step 2: Validate parameters

• Apply references/parameter-validation.md.
• If required parameters are missing, ask user or return clear remediation.

Step 3: Execute command

• Query commands: run once unless transient query error.
• Create commands: construct final command string and verify --confirm is present before execution.

Step 4: Verify create outcomes

• For create: verify with describe --region_id <REGION>.
• For create_namespace: verify with describe_namespaces --region_id <REGION> --instance_id <ID>.
• Use up to 3 read checks with short backoff before concluding the create is not reflected yet.
• For chained create -> create_namespace:
  • poll describe --region_id <REGION> on the same InstanceId every 30 seconds
  • max wait: 10 minutes
  • if still not RUNNING, stop and provide next action (wait/retry later)
  • do not switch to another instance without explicit user approval
  • if namespace create fails, mark lifecycle chain as failed/not_ready, not completed
  • for InsufficientResources, ask user to manually scale the instance and retry later

Key References

• Start here:
  • references/README.md
  • references/quick-start.md
  • references/trigger-recognition-guide.md
  • references/core-execution-flow.md
  • references/command-templates.md

| Document | Purpose | |----------|---------| | references/parameter-validation.md | Pre-execution validation checklist | | references/e2e-playbooks.md | Complete execution sequences | | references/common-failures.md | Typical mistakes and fixes | | references/required-confirmation-model.md | Confirmation gate rules | | references/instance-state-management.md | Instance state and readiness checks | | references/output-handling.md | Output parsing and retry policy | | references/verification-method.md | Verification patterns after create/query | | references/acceptance-criteria.md | Completion checklist for normal operations | | references/python-environment-setup.md | Python dependency and auth setup | | references/cli-installation-guide.md | Aliyun CLI diagnostics setup | | references/ram-policies.md | Required RAM permissions | | references/related-apis.md | API and command mapping |

Output Format

All commands return JSON:

{
  "success": true,
  "operation": "<command>",
  "confirmation_check": {
    "required_flag": "--confirm",
    "provided": true,
    "status": "passed"
  },
  "data": {},
  "request_id": "..."
}

confirmation_check appears on create operations and is used for auditable safety evidence.

Exit codes: 0 = success, 1 = error.