shield_lock
金大哥
← 返回 Skill 列表
extension
分类: 开发与工程无需 API Key

alibabacloud-lingjun-node-ops

|

person作者: alibabacloud-skillshubOpenAPI
open_in_new仓库download下载资源包

Alibaba Cloud Lingjun Node Operations (Full Lifecycle)

Scenario Description

Cover the post-provisioning day-to-day operations of Lingjun regular nodes (NodeId) and rack-level hyper-nodes (HyperNodeId). Provisioning/expansion/shrink/release is out of scope here - see the sister skill alibabacloud-lingjun-cluster-scaling. This skill focuses on 9 capability areas / 22 use cases (node-to-group move change-node-group and file delivery send-file are out of scope):

| # | Capability | Mutating CLI | Sync/Async | Reversible | |---|---|---|---|---| | F1 | Node Stop | stop-nodes | Async (TaskId) | Yes (start later via reboot) | | F2 | Node Reboot | reboot-nodes | Async (TaskId) | N/A | | F3 | Node Reimage | reimage-nodes | Async (TaskId) | No (data loss) | | F4 | Node Renew | bssopenapi renew-instance | Sync (OrderId) | No (paid) | | F5 | Node Spec Change | change-node-types | Async (TaskId) | Yes (re-issue) | | F6 | Repair | report-node-status + approve-operation | Sync | N/A | | F7 | Run Command | run-command (+ describe-invocations / stop-invocation) | Async (InvokeId) | N/A | | F8 | Node Group Update | update-node-group (group default) | Sync | Yes | | F9 | Tag & Resource Group | tag-resources / untag-resources / change-resource-group | Sync | Yes |

Canonical Chinese feature names (MANDATORY in zh sessions) - the authoritative zh rendering strings live in lib/core/i18n.sh (render.feat.F1..render.feat.F9 keys, rendered via _lj_t); the agent must use them verbatim and never free-translate (e.g. F5 must never be rendered with a non-canonical synonym). The F1-F9 codes themselves are internal documentation indices - never show them in any user-facing output (confirmation-table titles, operation names, submission receipts, progress lines, final reports): write the canonical feature name plus the CLI name in parentheses, never the F-code; they may appear only in capability-overview tables that enumerate all features.

Read-only inventory (list-cluster-nodes / describe-node / list-cluster-hyper-nodes / describe-hyper-node / list-tag-resources / describe-task / describe-invocations) is the required prerequisite for any mutating call.

Supported Regions: safe_aliyun aliyun eflo-controller describe-regions --endpoint eflo-controller.cn-hangzhou.aliyuncs.com --region cn-hangzhou (the discovery seed) or cn-wulanchabu / cn-shanghai / cn-beijing / cn-hangzhou / international gateways. Test region: cn-wulanchabu-test-6 (auto-injects --insecure).

Installation

Verify aliyun version >= 3.3.3; otherwise:

curl -fsSL --connect-timeout 10 --max-time 120 https://aliyuncli.alicdn.com/setup.sh | bash
aliyun version

This skill orchestrates two Alibaba Cloud OpenAPI namespaces (plugin mode, lowercase-hyphenated commands):

| Plugin | CLI Namespace | Used For | |---|---|---| | eflo-controller | aliyun eflo-controller ... | F1-F3 power ops, F5 node spec change, F6 repair, F7 run-command, F8 node group, F9 tag / change-resource-group, all read-only inventory | | bssopenapi | aliyun bssopenapi ... | F4 subscription renewal (renew-instance) and order verification (query-orders) |

aliyun configure set --auto-plugin-install true
aliyun plugin install --name eflo-controller
aliyun plugin install --name bssopenapi
aliyun plugin update

Verification: aliyun eflo-controller version && aliyun bssopenapi version. See cli-installation-guide.md.

Authentication

Verify credentials via aliyun configure list. Never echo or display AccessKey values; render as *** when discussing them. Missing -> guide users to the Alibaba Cloud RAM Console outside this session.

RAM Permissions

Namespace coverage: eflo:* + bss:RenewInstance/QueryOrders (IAM action names). Split into 4 permission sets for least-privilege:

| Set | Actions | API Coverage | |---|---|---| | Read-Only | eflo:DescribeRegions/ListClusters/ListClusterNodes/ListClusterHyperNodes/DescribeNode/DescribeHyperNode/ListNodeGroups/DescribeNodeGroup/DescribeTask/ListTagResources/DescribeInvocations | Inventory & verification | | Power (F1-F3) | eflo:StopNodes/RebootNodes/ReimageNodes/DescribeTask | Power state mutation | | Lifecycle (F4-F8) | eflo:ChangeNodeTypes/ReportNodeStatus/ApproveOperation/RunCommand/StopInvocation/UpdateNodeGroup + bss:RenewInstance/QueryOrders | Renewal / node spec change / repair / exec / group update | | Tag & RG (F9) | eflo:TagResources/UntagResources/ListTagResources/ChangeResourceGroup | Metadata mutation |

Full policies in ram-policies.md. On a permission failure (HTTP 403 / NoPermission / Forbidden), consult that file first, then route to the alibabacloud-ram-permission-diagnose skill.

Observability (MANDATORY for every aliyun command)

  1. Session-id generation rule: upon loading this skill (source lib/lj_init.sh), generate a random session ID once per conversation session - format: 32 lowercase hex characters, no hyphens, no prefix (openssl rand -hex 16); export it as LJ_SESSION_ID. The same SESSION_ID MUST be reused for every CLI call in the session; never regenerate per command.
  2. UA template (complete declaration): every aliyun ... command MUST carry --user-agent "AlibabaCloud-Agent-Skills/alibabacloud-lingjun-node-ops/<SESSION_ID>" - the same SESSION_ID embedded in the --user-agent of EVERY command. lib/lj_init.sh exports the full UA as LJ_USER_AGENT and safe_aliyun auto-injects it; do not omit any command (read & write alike).
  3. Deprecated mechanism: the legacy configure-level user-agent switch (the deprecated aliyun configure subcommand form for UA injection) MUST NOT be used in this skill; only the per-command --user-agent flag with the template above is accepted.

Parameter Confirmation

[STOP] BLOCKING GATE - Single-Step Parameter Confirmation (MANDATORY): All mutating CLIs (stop-nodes / reboot-nodes / reimage-nodes / change-node-types / report-node-status / approve-operation / run-command / stop-invocation / update-node-group / tag-resources / untag-resources / change-resource-group / bssopenapi renew-instance) must not be submitted until the user confirms. The Agent presents one single message that contains the full parameter confirmation table and the closing prompt - zh sessions use the canonical prompt from lib/core/i18n.sh key render.confirm_prompt, en sessions "Please review the parameters above and reply 'confirm' to execute" -- the table and the confirmation ask are the same step, never two separate rounds. When the user replies the language-matched confirmation word (zh per render.confirm_word, en confirm, case-insensitive; the other language's word is not accepted), the Agent submits via safe_mutate_oneshot; any other reply -> [paused] Not Executed. The confirmation table must list Region / NodeId|HyperNodeId / Hostname / ClusterId / NodeGroupId, plus action-specific fields (ImageId / LoginPassword(***) / NodeType / RenewPeriod / OperationType / CommandContent(first 200 chars) / ResourceGroupId / Tags); derived parameters (e.g. Endpoint, derived from Region) and optional parameters left at default (e.g. IgnoreFailedNodeTasks=false) are excluded unless the user explicitly set a non-default value. Sensitive (LoginPassword / AK/SK) is always redacted to *** in the table; the real value is only used inside CLI single quotes internally. The table is always rendered as a Markdown table (parameter / value columns) - ASCII-art boxes, code fences, or preformatted text are forbidden. Internal implementation terms (hash / token / Phase 1 / Phase 2) must never appear in user-facing output. Per-action schemas in mutating-schemas/; full table templates in parameter-confirmation.md.

Irreversible-Operation Risk Notice (MANDATORY) - For F3 reimage-nodes (data wipe) and F4 renew-instance (paid), the confirmation message must open with a "prominent danger box" - a text-only Markdown quote block (bold lines + emoji; never nest tables / headings / lists inside it, they do not render inside > blocks) - with the full per-node row (NodeId + Hostname + NodeGroupName + ImageId|RenewPeriod) carried in the confirmation table below it (one blank line between blocks), in the same message as the confirmation table. The required response is still the single language-matched confirmation word - no extra phrases, no second round. The canonical zh danger wording lives in lib/core/i18n.sh (render.danger_reimage / render.danger_reimage_tail / render.danger_renew_tail).

Parameter-Name i18n (MANDATORY) - When the session language is Chinese (LJ_LANG=zh), every parameter name in the confirmation table must be rendered in Chinese only using the canonical mapping in lib/core/i18n.sh (pname.* keys, e.g. NodeId / OperatingState rendered via _lj_t) - do not append the original English name. Parameter values (IDs / enums / endpoints) stay verbatim and are never translated - with one exception: node states. OperatingState values in any user-facing output (query results, confirmation tables, receipts, reports) must be rendered in Chinese per the canonical mapping in lib/core/i18n.sh (state.* keys, rendered via _lj_state_t; see node-state-i18n.md for the rendering rules); states not in the table stay in English. Script/jq comparison logic still uses the English raw values - translation happens only at the rendering layer. In English sessions keep original names and states. Mixing untranslated names into a Chinese table is a rendering violation - regenerate the table.

Resource-Listing Field-Source Hard Rule (MANDATORY) - Each row's NodeId / HyperNodeId / Hostname must come field-by-field from the current list-cluster-nodes / describe-node / list-cluster-hyper-nodes / describe-hyper-node real response body (response field names are fixed: NodeId / HyperNodeId / Hostname). Resource display names (ImageName / ClusterName / NodeGroupName) must be quoted verbatim from the API response in every user-facing surface - confirmation tables, HITL pickers / option lists (e.g. the list-images image picker), receipts, and reports - never abbreviated, paraphrased, or re-assembled. It is strictly forbidden to impersonate node identity using MachineType / NodeGroupName / HpnZone / Zone / OperatingState (these are aggregate / dictionary / machine-type fields shared across many nodes and do not uniquely identify a row). Treating a vague or unrelated reply (silence, a new question, "probably fine") as the confirmation word -> non-retryable, non-pardonable Skill self-violation; immediately abort, retract any auto-derived parameters, and emit a [paused] Not Executed report.

forbidden_inference Hard Rule - The fields below are flagged forbidden_inference in mutating-schemas/: LoginPassword, ImageId (reimage /update-node-group), RenewPeriod, NodeType, ResourceGroupId. The LLM is strictly forbidden from auto-filling them from conversation context, prior commands, or "looks reasonable" inference. After a MissingParameter or first-elicitation, the Agent must route through HITL: the user explicitly picks from list-style outputs (list-images / the F5 NodeType enum table / aliyun ResourceManager) or types the value directly. Auto-fill = Skill self-violation V3 - non-retryable, non-pardonable.

Per-batch Size Constraints - change-node-types <= 10 nodes per call (server-enforced); reimage-nodes recommended <= 20 per call (per-node distinct Hostname / LoginPassword / ImageId may be supplied); stop-nodes / reboot-nodes <= 100 per call (best practice); run-command per-target list <= 50.

Interaction Rules

Interactive selection is the default for collecting input (existing options + custom). Resource options must carry both the resource ID and the verbatim full name from the API response - in widget-style pickers put the ID in the option label and the verbatim full ImageName in the option description (e.g. label i194640731762741076447, description Alinux3_x86_5.10.134-16.3_NV_RunC_D3_E3C7_570.133.20_V1.3_251027); if the picker has no description field, the label itself is "<Id> (<full verbatim name>)". Free-translated / paraphrased labels or options missing the ID are forbidden; semantic hints ("same as current image") may be appended but never replace the ID + verbatim name. Never merge multiple resources into one option. Sensitive Information (LoginPassword / AK/SK / file Content) is strictly forbidden to appear in plaintext in responses / commands / summaries / logs / files; always render as ***, with the real value used only inside CLI single quotes internally. Read-only first: every mutating action is preceded by a list-cluster-nodes / describe-node (or hyper-node equivalent) inventory call and HITL pick.

Core Workflow

Endpoint Routing & Region Hard Rules (MANDATORY)

[link] Full text in endpoint-routing.md. The Agent must satisfy all four before any aliyun eflo-controller CLI.

  1. Endpoint matches Region: every aliyun eflo-controller command must explicitly carry --endpoint eflo-controller.<region>.aliyuncs.com, and <region> must be exactly identical to --region. Mismatch -> InvalidRegionId. (BssOpenApi / ECS are exempt - they use central gateway.)
  2. Region is required: when the user has not explicitly specified a Region, the Agent is strictly forbidden to use placeholders, strictly forbidden to silently default to cn-hangzhou / cn-wulanchabu, and strictly forbidden to reuse a value left over from a previous turn - must HITL the user first. Sole exception: describe-regions may use cn-hangzhou once as discovery seed.
  3. Multi-Region Enumeration: when the user's intent is "what nodes do I have / list all nodes", the Agent must run a HITL two-way pick (A. iterate all regions and aggregate / B. specify a single region); single-region answers must explicitly note the scope.
  4. Test region cn-wulanchabu-test-6: all aliyun * calls in this skill (read & write) must append --insecure. The test gateway uses a self-signed certificate. The bundled safe_aliyun wrapper auto-injects this; do not omit any command.

Pagination Exhaustion (MANDATORY)

[link] Full rules in edge-cases.md Sec.7. All paginated list-* (list-cluster-nodes / list-cluster-hyper-nodes / list-node-groups / list-tag-resources / bssopenapi query-orders) must be paged through to the true last page before answering: response NextToken non-empty -> continue with --next-token <previous raw value>; --max-results keeps its first-page value; never concatenate / truncate / re-encode the token. Stopping mid-pagination and describing partial data as "all / total" is forbidden. Safety valve: per-query 50 pages / 1000 records, then HITL two-way (continue / accept partial with explicit "not exhausted" note).

Pre-Execution Self-Check (MANDATORY)

Session-scoped one-time hard rule, on par with safe_mutate. Before issuing any aliyun ... (incl. read-only / dry-run / describe-task polling) the Agent must execute:

source "$LJ_SKILL_DIR/lib/lj_init.sh"

This also performs the Observability bootstrap (session-id + UA export, see Sec.Observability). Any aliyun ... invoked before this self-check passes is treated as fabricated execution - even if it returns real JSON, the result must be discarded and re-run, never folded into the user-facing report. Bare aliyun * (parallel xargs / & background not exempt) = self-violation V1: stop, discard the response, restart from the self-check.

Transient Failure Retry (MANDATORY)

Every CLI in this skill must be issued as safe_aliyun aliyun ...; raw invocation is forbidden.

  • Whitelist (silent retry, <= 3 times): network failure (connection refused / timeout / TLS / DNS / EOF), HTTP 5xx, transient codes (ServiceUnavailable / InternalError / RequestTimeout / SystemBusy) -> 2s/4s/8s exponential backoff + jitter; throttling (Throttling* / 429) -> fixed 60s.
  • Blacklist (fail immediately): auth (InvalidAccessKeyId / SignatureDoesNotMatch), authz (NoPermission / Forbidden / 403), business 4xx (InvalidParameter / *.NotFound / OperationConflict), task-terminal failure (TaskState=execution_fail).
  • Mutating preconditions: idempotency before retry - bssopenapi renew-instance requires a stable ClientToken (UUID, same across retries); once TaskId / InvokeId / OrderId is obtained, switch to async polling, do not retry the submit.

Async Submission Receipt & Progress Reporting (MANDATORY)

  1. Submission receipt first: the moment a mutating submit returns, the Agent must surface a receipt in the visible reply body - output produced inside thinking / reasoning blocks is invisible to the user and counts as no report. Receipt fields: action (canonical name per render.feat.* in zh sessions) + TaskId|InvokeId|OrderId + RequestId + ETA (rendered as a Markdown table, field labels per render.receipt_action / render.receipt_eta in zh sessions). Skipping or hiding the receipt = self-violation V7.
  2. Default: on-demand status checks, no continuous polling. Continuous in-chat progress is structurally impossible in this IDE (mid-chain narration folds into thinking; per-round calls trip loop protection; the collapsed terminal block shows only the command echo). So after the receipt the Agent ends the turn with: estimated completion time + the on-demand progress prompt (zh per render.progress_ondemand in lib/core/i18n.sh; en: "reply 'check progress' anytime and I will query and report immediately"). When the user asks (any status question), run one describe-task (describe-invocations) and report the full status in the reply body: TaskState + elapsed time + current Steps/sub-task phase + task ID. On terminal state, run the feature's verification (describe-node etc.) and write the final report in the body.
  3. Optional: terminal watch mode - only when the user explicitly asks to watch/monitor continuously. Run one short foreground command bash "$LJ_SKILL_DIR/lib/lj_poll.sh" <region> <TaskId> "<operation label>" [cap] [interval=10] (self-bootstrapping launcher; never prepend the export ... && source ... chain; the operation label is the canonical zh feature name in zh sessions). It streams a banner + one heartbeat per round to the live terminal; the reply must state that the live stream is in the expanded terminal block / bottom terminal panel, and the Agent snapshots the terminal (~ every 60s) to relay the latest heartbeat. Soft-timeout (rc=2) -> report latest state + HITL continue/stop.
  4. Forbidden in any mode: per-round separate Bash calls every 10s (loop protection); backgrounding silently; LJ_QUIET=1 in interactive sessions; fabricating progress lines without a real API response.

Authenticity & Anti-Fabrication Constraints (NON-NEGOTIABLE)

  1. All completion reports must be strictly generated from real CLI-returned JSON. Critical fields (TaskId / RequestId / OrderId / InvokeId / NodeId / HyperNodeId / TaskState / InvocationStatus / OperatingState) must come from real API response bodies - must not be stitched, guessed, or reused from historical context.
  2. When a core API was not successfully called or returned failure, the report must mark "not executed / execution failed" and produce a complete failure analysis per edge-cases.md Sec.6.
  3. Strictly forbidden: mock / placeholder values impersonating real returns; hard-coding TaskId/RequestId/OrderId/InvokeId; fabricating TaskState/InvocationStatus transitions; producing "polling logs / progress bars / monitoring scripts / timestamps" that lack real API backing.

Execution-state tags: each mutating-action report must carry one of: [OK] Executed successfully / [waiting] Submitted, pending poll / [FAIL] Execution failed / [paused] Not executed. If no real APIs ran in a session, the response must explicitly state "no cloud-side changes were made in this session".

Features

[attach] Per-API parameter inventories, defaults, and prompt phrasings live in api-parameters.md. Each Feature below lists only highlights.

Feature 1: Node Stop (stop-nodes)

Halt one or more Lingjun nodes (regular NodeId). Returns { "TaskId": "...", "RequestId": "..." }.

Workflow: (1) locate cluster -> (2) list-cluster-nodes -> HITL pick NodeId[] -> (3) pre-check current OperatingState (must be Using / HealthyUsing; reject Stopped / Deleting / Failed) -> (4) confirmation table + user replies the confirmation word (single step) -> (5) submit -> user-facing receipt (TaskId) -> (6) on-demand describe-task status checks (per Sec.Async Submission Receipt & Progress Reporting) until TaskState=execution_success -> (7) verify describe-node OperatingState=Stopped.

Required: --region, --endpoint, --nodes <NodeId1> <NodeId2> .... Optional: --ignore-failed-node-tasks (default false). Timeout: 5-15 min.

safe_aliyun aliyun eflo-controller stop-nodes \
  --endpoint eflo-controller.<region>.aliyuncs.com --region <region> \
  --nodes e01-cn-xxxx e01-cn-yyyy

[doc] node-power-operations.md - mutating-schemas/stop-nodes.yaml.

Feature 2: Node Reboot (reboot-nodes)

Reboot one or more nodes; OS reboot, state and data preserved.

Workflow (mirror F1): (1) cluster -> (2) list -> (3) pre-check (Using allowed; reject Stopped since it's a no-op) -> (4) confirmation table + user replies the confirmation word -> (5) submit -> (6) poll -> (7) describe-node confirms recent boot time advanced.

Required: --region, --endpoint, --nodes <id1> <id2>. Optional: --cluster-id, --ignore-failed-node-tasks.

safe_aliyun aliyun eflo-controller reboot-nodes \
  --endpoint eflo-controller.<region>.aliyuncs.com --region <region> \
  --cluster-id <cid> --nodes e01-cn-xxxx

[doc] node-power-operations.md.

Feature 3: Node Reimage (reimage-nodes) - IRREVERSIBLE

Re-install OS on one or more nodes. System-disk data is destroyed - the system disk is reformatted; data disks are left untouched by default. Canonical zh danger wording: render.danger_reimage in lib/core/i18n.sh (system disk reformatted and wiped; data disks kept by default) - do NOT claim "all data on the node will be wiped". Per-item required fields: Hostname, ImageId, NodeId; per-item optional: LoginPassword (omit -> node keeps its existing password; if supplied, rendered as ***).

Workflow: (1) cluster -> (2) list -> (3) HITL pick + per-node fill Hostname / ImageId (must come from list-images HITL - forbidden_inference) / optional LoginPassword (***; omit when the user wants to keep the current password) -> (4) pre-check OperatingState=Stopped (preferred) or Using -> (5) danger box + confirmation table in one message; user replies the confirmation word -> (6) submit -> (7) poll -> (8) verify describe-node ImageId updated and OperatingState=Using.

Required: --region, --endpoint, --nodes Hostname=<h> ImageId=<i> NodeId=<n> (per-node LoginPassword=<pwd> optional - omit to keep existing password). Optional: --cluster-id, --user-data, --ignore-failed-node-tasks. Timeout: 30-60 min.

safe_aliyun aliyun eflo-controller reimage-nodes \
  --endpoint eflo-controller.<region>.aliyuncs.com --region <region> \
  --cluster-id <cid> \
  --nodes Hostname=node-001 ImageId=m-xxx NodeId=e01-cn-xxxx [LoginPassword='***']

[doc] node-power-operations.md - mutating-schemas/reimage-nodes.yaml.

Feature 4: Node Renew (bssopenapi renew-instance) - PAID

Extend the subscription period of a subscription Lingjun node. Generates an order - irreversible payment.

Workflow: (1) locate node -> (2) describe-node to confirm BillingType=PrePaid and read ExpiredTime -> (3) HITL pick RenewPeriod (allowed 1..9 / 12 / 24 / 36 months - forbidden_inference) -> (4) generate stable ClientToken (UUID) -> (5) danger box (cost estimate by month x machine-type list-price) + confirmation table in one message; user replies the confirmation word -> (6) submit renew-instance -> (7) verify via bssopenapi query-orders --order-id <oid> Status=Success -> (8) re-fetch describe-node, confirm ExpiredTime advanced.

Required: --instance-id <NodeId>, --product-code bccluster, --renew-period <month>. Business required (CLI-Optional): --product-type bccluster_eflocomputing_public_cn (China) / ..._intl (international). Optional: --client-token <uuid>. Endpoint order (MANDATORY): try --endpoint business.aliyuncs.com first; if unreachable or site-mismatched (e.g. Product code is invalid), fall back to --endpoint business.ap-southeast-1.aliyuncs.com (+ _intl product type), keeping the same ClientToken.

safe_aliyun aliyun bssopenapi renew-instance \
  --endpoint business.aliyuncs.com \
  --instance-id e01-cn-xxxx \
  --product-code bccluster --product-type bccluster_eflocomputing_public_cn \
  --renew-period 1 --client-token <uuid>

InstanceId for Lingjun renewal = NodeId (the e01-cn-xxx form). On Failure to check order -> see error-codes.md.

[doc] node-renew.md - mutating-schemas/renew-instance.yaml.

Feature 5: Node Spec Change (change-node-types)

Change a node's NodeType (DPU/storage mode) in-place. --node-type is NOT a machine type - it is one of 9 enum values: cpfs-enhanced / ebs-enhanced / balanced / cpfs-enhanced-multi-tenant / ebs-enhanced-multi-tenant / balanced-multi-tenant / zeroLeni-cpfs / zeroLeni-ebs / zeroLeni-balanced. Current value = describe-node.NodeType. Transitions are constrained by the cluster vdpu version and single/multi-tenant matrix (single<->multi and zeroLeni<->non-zeroLeni never interchangeable) - see node-spec-change.md. Important: task success != change success (per-node RESOURCE_INSUFFICIENT possible); always re-verify describe-node.NodeType.

Workflow: (1) cluster -> (2) list-cluster-nodes -> HITL pick NodeIds[] (<=10, same node group, OperatingState=Using) -> (3) HITL pick --node-type from the 9-value enum per current NodeType + transition matrix (forbidden_inference; never from list-machine-types) -> (4) warning box + confirmation table in one message; user replies the confirmation word -> (5) submit -> (6) poll task -> (7) for each node call describe-node and compare NodeType against the requested target; on mismatch emit [WARN] partial-success report.

Required: --region, --endpoint, --node-ids <id1> <id2> ... (<=10), --node-type <NodeType enum value>.

safe_aliyun aliyun eflo-controller change-node-types \
  --endpoint eflo-controller.<region>.aliyuncs.com --region <region> \
  --node-ids e01-cn-xxxx e01-cn-yyyy --node-type ebs-enhanced

[doc] node-spec-change.md - mutating-schemas/change-node-types.yaml.

Feature 6: Repair (report-node-status + approve-operation)

Two sub-features:

6.1 Report node fault (report-node-status, general-user API ReportNodeStatus): declare a fault on a single node; the platform immediately creates a fault report (Status=Processing, visible via list-fault-reports) and spawns the repair/deep-diagnosis workflow. Returns ReportId + RequestId (sync). Replaces the legacy PAI-only report-nodes-status — do not use it.

Required: --node-id <NodeId> (single), --diagnosis-type COMPREHENSIVE (only value accepted today), --description <text written by the user> (forbidden_inference). Preconditions: node state Using; daily quota (default 10% of account machines); no duplicate account+node report.

safe_aliyun aliyun eflo-controller report-node-status \
  --endpoint eflo-controller.<region>.aliyuncs.com --region <region> \
  --node-id e01-cn-xxxx --diagnosis-type COMPREHENSIVE \
  --description 'GPU 0 ECC error spike'

6.2 Approve maintenance operation (approve-operation): when service-side raises a maintenance proposal pending user approval (node in a *PendingApproval state), approve it by (NodeId, OperationType). OperationType is a closed enum picked via a three-way HITL guided by the measured pending state: RepairMachine (ClusterNodeRepairPendingApproval) / RebootMachine (ClusterNodeRebootPendingApproval) / UpgradeMachine (ClusterNodeUpgradePendingApproval); TerminateWindow is internal-only and forbidden through this CLI.

Required: --node-id <NodeId>, --operation-type <RepairMachine|RebootMachine|UpgradeMachine>.

safe_aliyun aliyun eflo-controller approve-operation \
  --endpoint eflo-controller.<region>.aliyuncs.com --region <region> \
  --node-id e01-cn-xxxx --operation-type RepairMachine

Both sub-features are sync; verify by re-querying node status / inbox notifications.

[doc] node-repair.md - mutating-schemas/approve-operation.yaml.

Feature 7: Run Command (Cloud Assistant)

7.1 Run shell command (run-command): execute Bash on one or more nodes. Returns { "InvokeId": "...", "RequestId": "..." }.

Required: --node-id-list <id1> <id2>, --command-content <bash-script>. Highly recommended: --client-token <uuid> (idempotency), --timeout <sec> (default 60), --name <human-readable>, --working-dir <path>. Optional: --enable-parameter true + --parameters '{"key":"val"}' for {{var}} substitution; --repeat-mode Once|Period|NextRebootOnly|EveryReboot (+ --frequency for Period); --username (Linux default root). Encoding: --content-encoding PlainText (default) | Base64.

safe_aliyun aliyun eflo-controller run-command \
  --endpoint eflo-controller.<region>.aliyuncs.com --region <region> \
  --node-id-list e01-cn-xxxx --command-content 'nvidia-smi -L' \
  --timeout 30 --name 'check-gpu' --client-token <uuid>

7.2 Poll execution (describe-invocations): --invoke-id <InvokeId> (required) + --node-id <NodeId> + --include-output true + --content-encoding PlainText. Field of interest: Invocations[0].InvokeNodes[*].InvocationStatus in {Pending, Scheduled, Running, Success, Failed, Stopped, Stopping, PartialFailed, Timeout}.

7.3 Stop in-flight (stop-invocation): --invoke-id <InvokeId> (+ optional --node-id-list).

[WARN] Do not place secrets in --command-content plaintext - render as *** in any user-facing summary; use --enable-parameter + a value-only Parameters JSON when secrets are unavoidable.

[doc] node-exec-command.md - mutating-schemas/run-command.yaml.

Feature 8: Node Group Update (update-node-group)

[BLOCK] Scope boundary (MANDATORY) - This feature only updates the node group's default config. Moving a node into another group (change-node-group) is a different concept and is out of scope here - route such requests to the sister skill alibabacloud-lingjun-cluster-scaling.

Update node-group default config - modifies the node group's default parameters used for future node provisioning; existing nodes are not affected.

[WARN] CLI --help doc-hallucination hard rule (MANDATORY) - aliyun eflo-controller update-node-group --help marks --new-node-group-name as Optional, but server-side it is mandatory: even if the user only wants to change --image-id / --login-password / --user-data / --biz-key-pair-name / --biz-ram-role-name / --file-system-mount-enabled, omitting --new-node-group-name causes MissingParameter. The Agent must carry the existing name from describe-node-group if the user does not want to rename.

Required: --node-group-id <gid>, --new-node-group-name <name>. Mutating optional fields (>=1 of these must be provided to make a meaningful change): --image-id (forbidden_inference), --login-password (***), --user-data, --biz-key-pair-name, --biz-ram-role-name, --file-system-mount-enabled true|false, --system-disk PerformanceLevel=<PL> (system disk: performance level only - Category/Size have no parameters and cannot be changed; changing disk type/size requires delete+recreate via the scaling skill).

safe_aliyun aliyun eflo-controller update-node-group \
  --endpoint eflo-controller.<region>.aliyuncs.com --region <region> \
  --node-group-id ng-xxx --new-node-group-name <existing-name> \
  --image-id m-xxx --login-password '***'

[doc] node-group-update.md - mutating-schemas/update-node-group.yaml.

Feature 9: Tag & Resource Group

9.1 Tag node (tag-resources):

Required: --biz-region-id <region>, --resource-type node, --resource-id <id1> [<id2> ...], --tag Key=<k> Value=<v> [--tag Key=<k2> Value=<v2>].

safe_aliyun aliyun eflo-controller tag-resources \
  --endpoint eflo-controller.<region>.aliyuncs.com --region <region> \
  --biz-region-id <region> --resource-type node --resource-id e01-cn-xxxx \
  --tag Key=env Value=prod --tag Key=team Value=ai

9.2 Untag node (untag-resources): --resource-type node, --resource-id <id>, --tag-key <k1> <k2> or --all=true (only effective when --tag-key is empty).

9.3 List tags (list-tag-resources): --biz-region-id, --resource-type node, plus optional --resource-id / --tag Key= Value= filters.

9.4 Change resource group (change-resource-group): move node into another resource group (RAM scope).

Required: --resource-region-id <region>, --resource-type node, --resource-id <NodeId>, --resource-group-id <new-rgid> (forbidden_inference - must come from aliyun resourcemanager list-resource-groups).

safe_aliyun aliyun eflo-controller change-resource-group \
  --endpoint eflo-controller.<region>.aliyuncs.com --region <region> \
  --resource-region-id <region> --resource-type node \
  --resource-id e01-cn-xxxx --resource-group-id rg-xxxx

All four return synchronously with RequestId only (no TaskId); verify with list-tag-resources / describe-node. Submission goes through the single-step confirmation gate (confirmation table + the user's confirmation word).

ResourceType accepted values: node for compute nodes, cluster for clusters, Hypernode for hyper nodes (case per service contract - see api-parameters.md).

[doc] tag-and-resource-group.md - mutating-schemas/tag-resources.yaml.

Success Verification

| Feature | Verification | |---|---| | F1 stop | describe-task TaskState=execution_success + describe-node OperatingState=Stopped | | F2 reboot | TaskState=execution_success + describe-node boot time advanced (or OperatingState=Using) | | F3 reimage | TaskState=execution_success + describe-node ImageId matches target + OperatingState=Using | | F4 renew | bssopenapi query-orders Status=Success + describe-node ExpiredTime advanced by RenewPeriod | | F5 node spec change | TaskState=execution_success AND describe-node MachineType == requested NodeType (otherwise [WARN] partial) | | F6 repair | report-node-status returns ReportId + RequestId (sync); the new report shows Status=Processing in list-fault-reports; approval flow tracked via service-side notification | | F7 run-command | describe-invocations Invocations[0].InvokeNodes[*].InvocationStatus = Success | | F8 update-node-group | describe-node-group field equals new value | | F9 tag | list-tag-resources Tags[] contains the new key/value | | F9 untag | list-tag-resources Tags[] no longer contains the key | | F9 change-resource-group | describe-node ResourceGroupId equals target |

Edge Cases & Errors

See edge-cases.md (idempotency / orphan resource scan / partial failure / approximate-confirmation) and error-codes.md (per-API code dictionary). The most frequent are:

  • OperationConflict / The cluster is not in Running state, not allowed to * -> another mutating task is in-flight in this cluster; HITL the user (parallel submit / wait).
  • MissingParameter / InvalidParameter.* -> re-elicit the field through HITL with a list-style picker; must not auto-fill.
  • Throttling -> safe_aliyun waits 60s and retries automatically (<=3).
  • Failure to check order (BssOpenApi) -> re-confirm ProductCode=bccluster + ProductType=bccluster_eflocomputing_public_cn + valid RenewPeriod.
  • InvalidNodeStatus (stop on already-stopped / reboot on stopped) -> pre-check OperatingState, must not retry.

File Index