huawei-cloud-cci-instance-management

Huawei Cloud CCI (Cloud Container Instance) full lifecycle management using hcloud CLI. Covers Namespace, Network, Deployment, StatefulSet, Pod creation/update/delete/status, EIPPool for public IP, logs and metrics. Two-step confirmation for all destructive operations (delete namespace cascades all resources under it). Use this skill when the user wants to operate CCI serverless containers via command line. Triggers: CCI, 云容器实例, serverless container, 容器实例, namespace, deployment, statefulset, pod, EIPPool, CCI负载, 无服务器容器, 创建容器实例, 删除容器实例, 容器状态, 容器日志

Huawei Cloud CCI Container Instance Lifecycle Management

Overview

Manage Huawei Cloud CCI (Cloud Container Instance) full lifecycle using hcloud CLI (KooCLI). CCI is a serverless container service — no cluster management needed, just create a Namespace, define a Network, then deploy workloads directly.

Architecture: hcloud CLI → CCI OpenAPI → Namespace / Network / Deployment / StatefulSet / Pod / EIPPool (plus a helper script for annotation keys that hcloud CLI cannot express)

Constraints and Rules

Security Rules

• Two-step confirmation: All destructive operations (delete Namespace/Network/Deployment/StatefulSet/Pod/EIPPool/PVC) and all billable creation operations require explicit user confirmation — preview command, resource details, and cost/risk warning first; execute only after user confirms. Creating an EIPPool auto-creates EIPs and bandwidth and therefore incurs recurring charges: state the cost implication before asking for confirmation, and never create it silently.
• Credential security: Never expose AK/SK values in conversation, commands, or output. Only use hcloud configure list to check credential status (presence only). Prefer profile mode or environment variables over explicit AK/SK parameters.

Supported Scope

This skill manages only these CCI resources and their lifecycle: Namespace, Network, Deployment, StatefulSet, Pod, EIPPool, plus read-only status/log/event/quota queries for them.

Explicitly out of scope — decline these requests, say why, and point the user at the matching skill or the Huawei Cloud console. Do not attempt them and do not invoke another skill on the user's behalf:

Out-of-scope request	Where it belongs
CCE clusters, nodes, node pools	huawei-cloud-cce-* skills / CCE console
ECS / BMS instances, disks	huawei-cloud-ecs-* skills / ECS console
RDS, DDS, DCS databases	corresponding database skill / console
ELB, SFS, OBS, DNS, CDN resources	networking/storage skills / console
Creating VPC, subnet or security group themselves	huawei-cloud-vpc-* skills / VPC console
Billing, invoices, cost breakdown	Billing Center in the console
kubectl workflows	huawei-cloud-kubectl-cce-installer

This skill does not manage the lifecycle of other Kubernetes objects that CCI may also expose — Service, Ingress, ConfigMap, Secret, PVC, Job, Volcano Job, RoleBinding — and kubectl is out of scope. CCI does offer some write operations for those objects, but using them is not part of this skill, with one documented exception: the confirmation-gated cleanup of PVCs left behind after deleting a StatefulSet (see references/task-statefulset-management.md). Read-only inspection of ConfigMap, Secret, Service, ResourceQuota, Event, Pod status and logs is in scope and is exactly what the diagnosis playbooks use — see references/diagnosis-playbooks.md.

Resource Constraints

• Namespace flavor annotation is mandatory: Every namespace must carry the flavor annotation (value: general-computing or gpu-accelerated). Omitting it fails creation, and the hyphenated key form only takes effect for general-computing — see hcloud CLI Constraints.
• limits must equal requests: CCI enforces resources.limits == resources.requests. Mismatch causes "limit and request doesn't equal" error. Set both to the same values (e.g., 500m/1Gi).
• Network must precede workloads: Pod/Deployment/StatefulSet creation fails or stays Pending if no Network exists in the namespace. Always create Network before deploying workloads.
• VPC CIDR restriction: VPC subnet CIDR must NOT be 10.247.0.0/16 — CCI reserves this range for Service networking. Using it causes IP conflicts and workload creation failures.
• Deletion order: Pod → Deployment/StatefulSet → EIPPool → Network → Namespace. Deleting a Namespace cascades all resources under it.

hcloud CLI Constraints

• Network creation must use Python helper script: hcloud CLI cannot pass annotation keys containing dots (network.alpha.kubernetes.io/default-security-group). Neither dot notation nor --cli-jsonInput works. Use scripts/cci_network_helper.py.
• Namespace flavor annotation — the hyphen workaround is NOT safe for every flavor: The hyphen form (--metadata.annotations.namespace-kubernetes-io/flavor=<value>) is normalized by CCI back to namespace.kubernetes.io/flavor only for general-computing. For gpu-accelerated the hyphen form is silently dropped and the namespace ends up as general-computing, so the GPU capability is never enabled. The dotted key namespace.kubernetes.io/flavor is the required form, but hcloud CLI cannot pass dotted annotation keys (verified: the dotted, {...}-escaped and JSON-object forms are all rejected with USE_ERROR). Create GPU namespaces with the helper script's direct API call instead: python scripts/cci_network_helper.py create-namespace --name=<ns> --flavor=gpu-accelerated --region=<region> then confirm the canonical value with hcloud CCI readCoreV1Namespace --name=<ns>. Hyphen replacement never works for Network.
• Always verify parameters with --help: CCI has hundreds of parameters. Run hcloud CCI <Operation> --help before constructing any command. The help output is the authoritative source.

Detailed breakdowns of these rules are in Security Constraints, hcloud CLI Limitations, and Precautions.

Standard Workflow

text

1. Create Namespace (with flavor annotation)
2. Create Network (requires Python helper script for annotation — see hcloud CLI Limitations)
3. Create Deployment / StatefulSet / Pod (run workloads)
4. Query status, view logs
5. (Optional) Create EIPPool for Pod public IP access
6. Cleanup: delete workload → delete Network → delete Namespace

Prerequisites

1. hcloud CLI Requirements (MANDATORY)

• hcloud CLI installed (version >= 7.2.2)
• Run hcloud version to verify installation
• First-time usage: printf "y\n" | hcloud version to accept privacy statement
• hcloud must be resolvable on PATH. If it is not (or you want to pin a specific build), set HCLOUD_PATH to the executable — scripts/cci_network_helper.py honours it. You can also bypass project-ID auto-detection entirely by passing --project-id to the helper.

2. Credential Configuration

hcloud CLI supports two credential modes. See references/credential-configuration.md for full details.

Quick setup (choose one):

bash

# Mode A — Long-term AK/SK
export HUAWEI_CLOUD_AK=<your-ak>
export HUAWEI_CLOUD_SK=<your-sk>
export HUAWEI_CLOUD_REGION=cn-north-4

# Mode B — Temporary AK/SK + SecurityToken
export HUAWEI_CLOUD_AK=<your-temp-ak>
export HUAWEI_CLOUD_SK=<your-temp-sk>
export HUAWEI_CLOUD_SECURITY_TOKEN=<your-security-token>
export HUAWEI_CLOUD_REGION=cn-north-4

• Security rules: Never expose AK/SK/SecurityToken values. Use hcloud configure list to check presence only.

⚠️ Known limitation — Python helper script credentials are independent of hcloud CLI: The Python helper script (scripts/cci_network_helper.py) uses HW_ACCESS_KEY / HW_SECRET_KEY (and optionally HW_SECURITY_TOKEN) environment variables for authentication, which are separate from hcloud CLI's credential source (profile or HUAWEI_CLOUD_AK/HUAWEI_CLOUD_SK). If the credentials in HW_ACCESS_KEY/HW_SECRET_KEY lack the necessary IAM permissions for CCI Network creation, the script will fail with a 403 error. Ensure these variables contain credentials with sufficient CCI permissions (e.g., CCI FullAccess). The hcloud CLI continues using its own credential source independently — running the helper script does not affect subsequent hcloud CLI commands.

3. Validation Check

bash

hcloud version
hcloud configure list

Security Constraints

Dangerous Operation Confirmation Mechanism

This skill strictly enforces a two-step confirmation mechanism for all destructive and billable creation operations.

All destructive operations and all billable creation operations require explicit user confirmation before execution. The process:

Step 1: Preview — Show the command, resource details, risk warning, and any cost implication

Step 2: Confirm & Execute — Only after user explicitly confirms

Operations Requiring Confirmation

Operation	Risk Level	Description
Create EIPPool	💰 Billable	Auto-creates EIPs and bandwidth — recurring charges. Give the cost warning and get explicit confirmation before creating
Delete Namespace	🔴 Critical	Cascades — deletes ALL resources under this namespace (Network, Pods, Deployments, etc.)
Delete Network	🟠 High	Disconnects namespace from VPC; running pods lose network
Delete Deployment	🟠 High	Terminates all replicas of the workload
Delete StatefulSet	🟠 High	Terminates all replicas; PVC data may be lost
Delete Pod	🟠 High	Terminates the container instance
Delete EIPPool	🟡 Medium	Releases public IPs allocated to pods
Delete PVC (StatefulSet cleanup)	🟠 High	Destroys the claim and its volume; the data cannot be recovered. Preview with listCoreV1NamespacedPersistentVolumeClaim / readCoreV1NamespacedPersistentVolumeClaim, then delete one PVC at a time

Credential Security

• Never expose AK/SK/SecurityToken values in conversation, commands, or output
• Never ask user to input AK/SK/SecurityToken directly in conversation
• Only use hcloud configure list to check credential status (presence only, not values)
• Prefer profile mode or environment variables over explicit AK/SK parameters

Command Format Standard

CCI follows the standard hcloud format with Kubernetes-style nested parameters:

bash

hcloud CCI <Operation> --param=value --cli-region=<region> --cli-output=json

CCI-Specific Parameter Rules

CCI parameters follow Kubernetes API conventions — deeply nested objects with dot notation:

1. Annotations use {*} format: --metadata.annotations.namespace-kubernetes-io/flavor=general-computing
2. Labels use {*} format: --metadata.labels.app=my-app
3. Containers array (1-based): --spec.template.spec.containers.1.name=main --spec.template.spec.containers.1.image=nginx
4. Resources use {*} format: --spec.template.spec.containers.1.resources.limits.cpu=500m
5. Selector matchLabels use {*} format: --spec.selector.matchLabels.app=my-app
6. Namespaced operations require --namespace: all workload operations must specify namespace

⚠️ Critical: Before constructing any CCI command, always run hcloud CCI <Operation> --help to verify exact parameter names. CCI has hundreds of parameters; the help output is the authoritative source.

Parameter Format Details

See references/parameter-format.md for complete CCI parameter format rules and examples.

Scenario Routing

User Intent	Reference Document
Create/query/delete Namespace	references/task-namespace-management.md
Create/query/delete Network	references/task-network-management.md
Create/query/update/delete/scale Deployment	references/task-deployment-management.md
Create/query/update/delete StatefulSet	references/task-statefulset-management.md
Create/query/delete Pod	references/task-pod-management.md
Create/query/delete EIPPool	references/task-eippool-management.md
Query status, view logs, events	references/task-logs-and-status.md
Full workflow (create→run→cleanup)	references/common-workflows.md
All CCI operations quick reference	references/cci-operation-catalog.md
Troubleshooting	references/troubleshooting.md
Diagnosis playbooks (10 ticket categories)	references/diagnosis-playbooks.md
Flavors, quotas, output, debugging	references/operational-notes.md
IAM permissions	references/iam-policies.md
Verification steps	references/verification-method.md
Correct/error pattern comparison	references/acceptance-criteria.md

Core Commands

Namespace

bash

# Create namespace (general-computing flavor)
hcloud CCI createCoreV1Namespace \
  --metadata.name=<ns-name> \
  --metadata.annotations.namespace-kubernetes-io/flavor=general-computing \
  --cli-region=<region> \
  --cli-output=json

# Create namespace (gpu-accelerated flavor)
# The hyphen form above does NOT work for gpu-accelerated — hcloud CLI cannot pass the
# dotted annotation key, so use the helper script's direct API call:
python scripts/cci_network_helper.py create-namespace \
  --name=<ns-name> \
  --flavor=gpu-accelerated \
  --region=<region>

# List namespaces
hcloud CCI listCoreV1Namespace --cli-region=<region> --cli-output=json

# Read namespace details
hcloud CCI readCoreV1Namespace --name=<ns-name> --cli-region=<region> --cli-output=json

# Delete namespace (TWO-STEP CONFIRMATION REQUIRED)
hcloud CCI deleteCoreV1Namespace --name=<ns-name> --cli-region=<region>

Network

⚠️ hcloud CLI limitation: Network creation requires a Python helper script because hcloud CLI cannot pass the annotation key network.alpha.kubernetes.io/default-security-group (contains dots that hcloud treats as nested levels). The --cli-jsonInput approach also doesn't work due to an hcloud bug where annotations show in --dryrun but aren't transmitted in actual requests. See hcloud CLI Limitations below.

⚠️ Credential requirement: The Python helper script uses HW_ACCESS_KEY/HW_SECRET_KEY (and optionally HW_SECURITY_TOKEN) environment variables for authentication. This is independent from hcloud CLI's credential source (which reads from HUAWEI_CLOUD_AK/HUAWEI_CLOUD_SK or its profile). If the credentials in HW_ACCESS_KEY/HW_SECRET_KEY lack CCI Network creation permissions, the script will fail with 403. Ensure they have sufficient IAM permissions (e.g., CCI FullAccess). After the script runs, hcloud CLI commands continue using their own credential source unaffected.

bash

# Step 1: Get VPC subnet details (including neutron_network_id)
hcloud VPC ShowSubnet --subnet_id=<subnet-id> --cli-region=<region> --cli-output=json

# Step 2: Create network via Python helper script
python scripts/cci_network_helper.py create \
  --namespace=<ns-name> \
  --name=<network-name> \
  --vpc-id=<vpc-id> \
  --subnet-id=<subnet-id> \
  --network-id=<neutron-network-id> \
  --security-group-id=<sg-id> \
  --region=<region>

# Step 3: Check network status until Active
hcloud CCI readNetworkingCciIoV1beta1NamespacedNetworkStatus \
  --name=<network-name> \
  --namespace=<ns-name> \
  --cli-region=<region> \
  --cli-output=json

# List networks
hcloud CCI listNetworkingCciIoV1beta1NamespacedNetwork \
  --namespace=<ns-name> \
  --cli-region=<region> \
  --cli-output=json

Required Network spec fields: Network creation requires attachedVPC, subnetID, networkType, AND networkID (neutron network ID). The networkID field is REQUIRED — it is the neutron network ID obtained from hcloud VPC ShowSubnet.

Required Network annotation: network.alpha.kubernetes.io/default-security-group (the correct annotation key for CCI Network security group, NOT security-group-id). This annotation must be set to the security group ID.

Deployment

bash

# Create deployment
hcloud CCI createAppsV1NamespacedDeployment \
  --namespace=<ns-name> \
  --metadata.name=<deploy-name> \
  --spec.replicas=1 \
  --spec.selector.matchLabels.app=<deploy-name> \
  --spec.template.metadata.labels.app=<deploy-name> \
  --spec.template.spec.containers.1.name=<container-name> \
  --spec.template.spec.containers.1.image=<image> \
  --spec.template.spec.containers.1.resources.limits.cpu=500m \
  --spec.template.spec.containers.1.resources.limits.memory=1Gi \
  --spec.template.spec.containers.1.resources.requests.cpu=500m \
  --spec.template.spec.containers.1.resources.requests.memory=1Gi \
  --cli-region=<region> \
  --cli-output=json

# Scale deployment
# --spec.replicas is NOT a valid parameter; the PATCH body must be a JSON Patch array.
hcloud CCI patchAppsV1NamespacedDeploymentScale \
  --name=<deploy-name> \
  --namespace=<ns-name> \
  --io.k8s.apimachinery.pkg.apis.meta.v1.Patch='[{"op":"replace","path":"/spec/replicas","value":<new-replicas>}]' \
  --cli-region=<region> \
  --cli-output=json

# Update container image (same JSON Patch form; note the 0-based array index in "path")
hcloud CCI patchAppsV1NamespacedDeployment \
  --name=<deploy-name> \
  --namespace=<ns-name> \
  --io.k8s.apimachinery.pkg.apis.meta.v1.Patch='[{"op":"replace","path":"/spec/template/spec/containers/0/image","value":"<new-image>"}]' \
  --cli-region=<region> \
  --cli-output=json

# Read deployment status
hcloud CCI readAppsV1NamespacedDeploymentStatus \
  --name=<deploy-name> \
  --namespace=<ns-name> \
  --cli-region=<region> \
  --cli-output=json

StatefulSet

bash

# Create statefulset
hcloud CCI createAppsV1NamespacedStatefulSet \
  --namespace=<ns-name> \
  --metadata.name=<sts-name> \
  --spec.replicas=1 \
  --spec.selector.matchLabels.app=<sts-name> \
  --spec.template.metadata.labels.app=<sts-name> \
  --spec.template.spec.containers.1.name=<container-name> \
  --spec.template.spec.containers.1.image=<image> \
  --spec.template.spec.containers.1.resources.limits.cpu=500m \
  --spec.template.spec.containers.1.resources.limits.memory=1Gi \
  --spec.template.spec.containers.1.resources.requests.cpu=500m \
  --spec.template.spec.containers.1.resources.requests.memory=1Gi \
  --cli-region=<region> \
  --cli-output=json

# Read statefulset status
hcloud CCI readAppsV1NamespacedStatefulSetStatus \
  --name=<sts-name> \
  --namespace=<ns-name> \
  --cli-region=<region> \
  --cli-output=json

Pod

bash

# Create pod (single container instance)
hcloud CCI createCoreV1NamespacedPod \
  --namespace=<ns-name> \
  --metadata.name=<pod-name> \
  --spec.containers.1.name=<container-name> \
  --spec.containers.1.image=<image> \
  --spec.containers.1.resources.limits.cpu=500m \
  --spec.containers.1.resources.limits.memory=1Gi \
  --spec.containers.1.resources.requests.cpu=500m \
  --spec.containers.1.resources.requests.memory=1Gi \
  --cli-region=<region> \
  --cli-output=json

# Read pod status
hcloud CCI readCoreV1NamespacedPodStatus \
  --name=<pod-name> \
  --namespace=<ns-name> \
  --cli-region=<region> \
  --cli-output=json

# Read pod logs
hcloud CCI readCoreV1NamespacedPodLog \
  --name=<pod-name> \
  --namespace=<ns-name> \
  --container=<container-name> \
  --cli-region=<region>

EIPPool

bash

# Create EIPPool (for pod public IP access — auto-create EIPs)
hcloud CCI createCrdYangtseCniV1NamespacedEIPPool \
  --namespace=<ns-name> \
  --apiVersion=crd.yangtse.cni/v1 \
  --kind=EIPPool \
  --metadata.name=<eippool-name> \
  --spec.amount=1 \
  --spec.eipAttributes.networkType=5_bgp \
  --spec.eipAttributes.ipVersion=4 \
  --spec.eipAttributes.bandwidth.shareType=PER \
  --spec.eipAttributes.bandwidth.size=5 \
  --spec.eipAttributes.bandwidth.chargeMode=bandwidth \
  --spec.eipAttributes.bandwidth.name=<bw-name> \
  --cli-region=<region> \
  --cli-output=json

# Read EIPPool status
hcloud CCI readCrdYangtseCniV1NamespacedEIPPoolStatus \
  --name=<eippool-name> \
  --namespace=<ns-name> \
  --cli-region=<region> \
  --cli-output=json

EIPPool required fields: --apiVersion=crd.yangtse.cni/v1 and --kind=EIPPool are mandatory. spec.eipAttributes.networkType is required. The only real values are 5_telcom (China Telecom), 5_union (China Unicom), 5_bgp (dynamic BGP), 5_sbgp (static BGP) and 5_ipv6 — any other value, including the dedicated-load-balancing value shown in older revisions of this document, does not exist in any region and is rejected by the admission webhook. Which values a region accepts differs (for example cn-north-4 supports only 5_bgp and 5_sbgp); always confirm the list for your region with hcloud EIP CreatePublicip --help, whose output enumerates the supported types per region. spec.eipAttributes.bandwidth.chargeMode and name are required when auto-creating EIPs.

💰 Billable operation: creating an EIPPool auto-creates EIPs and bandwidth, which are charged. Warn the user about the cost and obtain explicit confirmation before creating an EIPPool.

Pod EIP binding: To assign an EIPPool to a Pod, add the annotation yangtse.io/eippool=<eippool-name>. hcloud CLI cannot pass dotted annotation keys, so the hyphenated form --metadata.annotations.yangtse-io/eippool=<eippool-name> is the only CLI-expressible option; CCI stores hyphenated keys verbatim, so verify the binding after creation with hcloud CCI readCoreV1NamespacedPod --name=<pod> --namespace=<ns>, and fall back to a direct API call or the console if the annotation was not recognised.

VPC/Subnet Prerequisites

A Network maps to an existing VPC subnet, so before creating one you must supply the VPC ID, the subnet ID and the neutron network ID (networkID is mandatory):

bash

hcloud VPC ListVpcs --cli-region=<region> --cli-output=json
hcloud VPC ListSubnets --vpc_id=<vpc-id> --cli-region=<region> --cli-output=json
hcloud VPC ShowSubnet --subnet_id=<subnet-id> --cli-region=<region> --cli-output=json   # neutron_network_id comes from here

⚠️ The VPC/subnet CIDR must NOT be 10.247.0.0/16 — CCI reserves that range for Service networking. Full details: references/task-network-management.md.

Namespace Flavors, Quotas, Output and Debugging

Flavor values (general-computing / gpu-accelerated), per-namespace resource quotas, JSON/table output, JMESPath filtering, the global hcloud --debug switch and best practices live in references/operational-notes.md.

Parameter Confirmation

Before executing any CCI operation, confirm these parameters:

Parameter	Required	Description	Source
--namespace	Yes	CCI namespace name	Existing or newly created
--cli-region	Yes	Huawei Cloud region ID	HUAWEI_CLOUD_REGION or config
--metadata.name	Yes	Resource name	User specified
Flavor annotation	Yes (Namespace)	general-computing or gpu-accelerated	User choice
VPC/Subnet ID	Yes (Network)	From VPC ListVpcs / VPC ShowSubnet	Query existing resources
neutron_network_id	Yes (Network)	From VPC ShowSubnet response	Query result

Run hcloud CCI <Operation> --help before any CCI command to verify parameter names, then cross-reference the table above.

Precautions

See references/troubleshooting.md for detailed troubleshooting.

Quick reference:

Issue	Cause	Quick Fix
Namespace creation fails	Missing flavor annotation	Add --metadata.annotations.namespace-kubernetes-io/flavor=general-computing
Network creation fails (400/403)	Missing VPC/subnet/annotation/networkID, or credential scope insufficient	Verify subnet/neutron IDs, security group; use Python helper; use long-term AK/SK (Mode A)
Pod stays Pending	No Network in namespace	Create Network first
403 permission error	Insufficient IAM	Check references/iam-policies.md
Deep nested param errors	Wrong dot notation	Use --help to verify exact parameter path
Annotation with dots not passed	hcloud CLI limitation	Use Python helper script for Network creation
EIPPool creation fails (400/422)	Missing apiVersion/kind/networkType	Add all required fields (see EIPPool section)
limit/request mismatch	CCI requires limits == requests	Set requests same as limits (e.g., both 500m/1Gi)

Verification Method

See references/verification-method.md for complete verification steps.

Quick checklist:

Step	Command	Expected Result
Namespace	hcloud CCI readCoreV1Namespace --name=<ns> --cli-region=<region>	status.phase=Active
Network	hcloud CCI readNetworkingCciIoV1beta1NamespacedNetworkStatus --name=<net> --namespace=<ns>	status.phase=Active
Deployment	hcloud CCI readAppsV1NamespacedDeploymentStatus --name=<deploy> --namespace=<ns>	readyReplicas >= 1
Pod	hcloud CCI readCoreV1NamespacedPodStatus --name=<pod> --namespace=<ns>	status.phase=Running

Implementation Notes

• Namespace isolation: use different namespaces for different teams/projects.
• EIPPool on-demand: EIPs and bandwidth are billable — create an EIPPool only when Pod public IP access is needed, and confirm the cost with the user first.
• Best practices, flavors, quotas, output and debugging: references/operational-notes.md

hcloud CLI Limitations

⚠️ Critical: hcloud CLI has known limitations that affect CCI operations. Understanding these is essential for successful Network creation.

Limitation	Impact	Workaround
Cannot pass annotation keys containing dots (.) via CLI parameters	hcloud treats dots in parameter names as nested object levels, so --metadata.annotations.network.alpha.kubernetes.io/default-security-group creates a deeply nested structure instead of a single annotation key	Use Python helper script (scripts/cci_network_helper.py) for Network creation
--cli-jsonInput is unreliable for annotations	Annotations appear in --dryrun output but are not transmitted in real API requests; a UTF-8 BOM also breaks parsing	Use the Python helper script; if JSON input is unavoidable, save it as plain ASCII (no BOM)
Namespace flavor: hyphen replacement is only partial	The hyphen form is normalized back by CCI only for general-computing; for gpu-accelerated it is silently dropped and the namespace falls back to general-computing	Use scripts/cci_network_helper.py create-namespace (dotted annotation). The workaround never applies to Network

Why the Python helper is needed: annotation keys such as network.alpha.kubernetes.io/default-security-group and namespace.kubernetes.io/flavor contain dots, which hcloud CLI interprets as nested object levels — so they cannot be passed as CLI parameters. scripts/cci_network_helper.py builds the correct API request body directly.

References

Document	Description
task-namespace-management.md	Namespace lifecycle operations
task-network-management.md	Network lifecycle operations
task-deployment-management.md	Deployment lifecycle operations
task-statefulset-management.md	StatefulSet lifecycle operations
task-pod-management.md	Pod lifecycle operations
task-eippool-management.md	EIPPool operations
task-logs-and-status.md	Status queries and log viewing
cci-operation-catalog.md	Full CCI operation quick reference
parameter-format.md	CCI parameter format rules and examples
common-workflows.md	Complete workflow sequences
credential-configuration.md	Credential setup (long-term AK/SK & temporary AK/SK+SecurityToken)
iam-policies.md	IAM permission policies
troubleshooting.md	Error troubleshooting
diagnosis-playbooks.md	Read-only diagnosis playbooks for the 10 high-frequency ticket categories
operational-notes.md	Namespace flavors, resource quotas, output format, debugging, best practices
verification-method.md	Verification steps
acceptance-criteria.md	Correct/error pattern comparison
scripts/cci_network_helper.py	Python helper script for Network creation (bypasses hcloud CLI annotation limitations)