Openclaw Skills for Huawei Cloud DWS memory diagnosis automatically collect metrics, isolate root causes, and generate standardized HTML reports.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install bigdata-dws-huawei-cloud-dws-dymem-diag
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install bigdata-dws-huawei-cloud-dws-dymem-diag using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Huawei Cloud DWS Memory High Diagnosis?
Openclaw Skills for Huawei Cloud DWS memory diagnosis turn a memory alarm, OOM event, or high-usage trigger into an evidence-based root-cause workflow. The skill prefers KooCLI v3.2.0+ for cluster, host, and metric queries, and transparently falls back to the DWS Autopilot MCP Server when KooCLI is unavailable or the network probe fails.
It analyzes MemStat, InstanceMemory, and memory_diagnose_detail data to determine whether the incident is cluster-wide, node-specific, instance-specific, SQL-driven, or session-related. The result is a standardized HTML diagnosis report saved in the workspace root, with conclusions limited strictly to observed tool output.
Huawei Cloud DWS Memory High Diagnosis Use Cases
- DWS cluster memory usage too high, memory alarm, insufficient memory, or OOM incidents
- Need to determine whether pressure is global, partial, or isolated to a single node
- Need to trace high-memory CN/DN instances, active users, sessions, or SQL statements
- Suspected data skew, memory imbalance, work_mem pressure, shared_pool pressure, or memory leak behavior
- KooCLI is preferred for diagnosis, but MCP fallback is required for resilience
- A standardized HTML report is needed for incident handoff, review, or postmortem work
How Huawei Cloud DWS Memory High Diagnosis Works
- Detect the execution environment by running
hcloud version; if the version is at least 3.2.0, start in KooCLI mode, otherwise switch to MCP mode. Probe network availability with a lightweight KooCLI API call, and if the first KooCLI call returnsNETWORK_ERROR, fall back to MCP and keep that mode for the rest of the run. - Query
MemStatfor the window fromfirst_alarm_time - 1200000tofirst_alarm_time, using paginated reads withlimit=200. Merge all pages locally, then compute node memory usage, identify the highest-memory hosts, and classify the scope as global, partial, or single-node pressure. - Query host information for the affected host IDs so the diagnosis can map
host_idto host name and IP address. This step is used to label the suspicious node and connect host-level evidence to the alarm target. - Query
InstanceMemoryfor the same time window, again using pagination. Filter the cluster-wide response byinst_name, infer CN or DN from the instance name, and determine whether the pressure is mainly dynamic memory or process memory. - Query
memory_diagnose_detail, the largest dataset in the workflow, with the same time window and mandatory pagination. Extract active sessions, database users, SQL text, query IDs, plan types, and memory pool data, then useuserName,state,duration_ms, andmem_usedto identify high-memory users and statements. - Analyze the combined evidence to decide whether the root cause is customer-side or system-side. Check for single SQL dominance, multi-user concurrency, session leaks, long transactions, work_mem pressure, shared_pool pressure, and possible memory leak trends. Convert all displayed timestamps from UTC millisecond values to Beijing time before placing them in the report.
- Resolve the display name from
cluster_nameor the cluster lookup, generate the final HTML diagnosis report, and save it in the current working directory asdws_mem_diagnosis_report_{timestamp}.html.
Huawei Cloud DWS Memory High Diagnosis Setup
- Verify that KooCLI is installed and meets the minimum version requirement.
hcloud version
If the command is missing or the version is below 3.2.0, install or upgrade KooCLI using the linked CLI installation guide.
- Confirm that credentials are configured without exposing secrets.
hcloud configure list
Use existing Huawei Cloud authentication only; never paste AK/SK values into chat or shell history.
Install and configure the DWS Autopilot MCP Server as the fallback path when KooCLI is unavailable or returns
NETWORK_ERROR.Ensure the account has the required permissions:
dws:clusters:get,dws:clusters:list,dws:metricData:get, anddws:hostOverview:get.Prepare the required inputs for Openclaw Skills:
alarm_serial_number,project_id,cluster_id,first_alarm_time, andalarm_name. Provideregion_idwhen using hcloud commands.Validate access with a lightweight cluster query before running a full diagnosis.
hcloud DWS ListClusters --cli-region=<region> --project_id=<project_id> --offset=0 --limit=1
- Make sure the workspace is writable so the generated report file
dws_mem_diagnosis_report_{timestamp}.htmlcan be saved in the current working directory.
Huawei Cloud DWS Memory High Diagnosis Data Schema & Taxonomy
Input contract
| Field | Required | Type | Purpose |
|---|---|---|---|
alarm_serial_number |
Yes | string | Alarm serial number |
project_id |
Yes | string | Passed to all tool calls |
cluster_id |
Yes | string | DWS cluster instance ID and alarm resourceId |
first_alarm_time |
Yes | integer | UTC Unix millisecond timestamp used as the end of the diagnosis window |
alarm_name |
Yes | string | Alarm title |
region_id |
No | string | hcloud --cli-region value |
node_name |
No | string | Host name extracted from alarm payload |
instance_name |
No | string | Instance name extracted from the alarm MOI field |
cluster_name |
No | string | Preferred display name if already known |
alarm_severity |
No | integer | 1 Critical, 2 Major, 3 Minor, 4 Info |
Collected metric layers
| Layer | Tool / Metric | Primary fields | Derived meaning |
|---|---|---|---|
| Cluster and host | MemStat |
ctime, host_id, mem_total, mem_free, cached, buffers, watermark_high, watermark_min |
Node memory usage and cluster imbalance |
| Instance | InstanceMemory |
host_id, inst_name, dynamic_used_memory, max_dynamic_memory, process_used_memory, max_process_memory, plus additional memory buckets |
CN/DN pressure and dynamic vs process memory split |
| Session and SQL | memory_diagnose_detail |
active_sessions, query, query_id, userName, mem_used, state, duration_ms, plan_type, memory_pool |
Active users, high-memory SQL, session leaks, and pool pressure |
Derived taxonomy
mem_scope: global high, partial high, single-node high, or normal/imbalancedinst_type: CN or DN, inferred frominst_namemem_type: dynamic memory high or process memory highroot_cause_category: customer-side or system-sideconcurrent_mode: multi_user, single_user, or nonehigh_memory_sql_info,session_info, andmemory_pool_summary: evidence bundles for the report
Output contract
| Field | Type | Meaning |
|---|---|---|
diagnosis_summary |
string | Short result summary |
diagnosis_report |
string | HTML diagnosis report body |
Generated artifact
dws_mem_diagnosis_report_{timestamp}.htmlis saved in the current working directorytimestampuses local machine time inyyyyMMdd_HHmmss- Report timestamps are displayed in Beijing time (
UTC+8)
Transport rules
- Tool responses follow
{"code":0,"data":...}on success and{"code":-1,"message":...}on failure - All metric queries must use paginated reads with
limit=200and be merged locally before analysis
Huawei Cloud DWS Memory High Diagnosis Advanced Features
- Automatic KooCLI-first execution with MCP fallback when KooCLI is unavailable or returns
NETWORK_ERROR - Mandatory pagination with
limit=200and local dataset merging for large metric responses - Cross-layer correlation across cluster, host, instance, session, and SQL evidence
- CN/DN inference from
inst_namewhen no explicit instance type field exists - Dynamic memory and process memory split, plus
work_memandshared_poolpressure checks - Session leak, long transaction, high concurrency, and memory leak heuristics
- Top 3 user and SQL contribution tables with percentage-bar visualization
- UTC-to-Beijing time normalization for all report timestamps
- Evidence-only diagnosis policy with graceful degradation to
unavailablewhen data is missing
SKILL.md
Loading
Related Openclaw Skills
HTTP Status Decoder
A fast HTTP status code decoder that explains web errors in plain language and shows where to start debugging.
ephemeral-eng Input-to-Deliverable Engineering Tools
A single-file HTML framework for Openclaw Skills that turns engineering inputs into print-ready calculations, diagrams, and client deliverables.
Skill Auditor
Skill Auditor is a developer-focused utility designed to analyze, score, and safely remediate Openclaw Skills across eight structural and security dimensions.
Consumer Rights Guide
An Openclaw Skills consumer-rights playbook that turns purchase disputes into evidence packs, complaint drafts, and escalation-ready filings.
MFZ Catalogue Report
Converts one DRG or DIP catalogue into a validated, provenance-rich report through a secure MCP workflow.
8D Report Generation Skill
An AI-assisted 8D reporting workflow that guides teams through D1-D8 problem solving and generates a structured report with optional Word export.
Featured*
