Huawei Cloud RDS Troubleshoot Skill

Overview

This skill provides full-scenario intelligent service for Huawei Cloud RDS (Relational Database Service) for MySQL and PostgreSQL engines. It covers:

1. Basic RDS Q&A — explain instance concepts, status meanings, HA architecture, and common terms from live instance data.
2. SQL performance optimization — locate slow queries via slow logs, correlate with CPU/connection pressure, and suggest index/parameter improvements.
3. Daily instance operations — inventory, status, storage, replication, and configuration inspection.
4. Online fault location & troubleshooting (core scenario) — guided step-by-step diagnosis of the most common production faults: instance unreachable/crash, slow SQL & high pressure, disk full, primary-standby replication failure, connection limit exceeded, and memory overrun.
5. Parameter tuning — inspect current parameter values, compare templates, and modify parameters (with user confirmation).
6. Backup/restore guidance — list backups, create manual backups, and guide restore flows (with user confirmation before write operations).

Execution mode: CLI-first (hcloud RDS <Operation> --cli-region={region}), falling back to the huaweicloudsdkrds Python SDK, then REST API.

Output form: interactive guided troubleshooting — one diagnostic step at a time, presenting findings and the next recommended step, never dumping raw JSON at the user.

Architecture:

User/Agent → hcloud CLI (KooCLI RDS, primary) → Huawei Cloud RDS v3 API
           ↘ huaweicloudsdkrds Python SDK (fallback)      ↗
           ↘ REST API (curl, last resort)                 ↗

API base (all operations below are KooCLI RDS operations against the v3 API; paths verified from KooCLI operation definitions):

能力边界（Capability Boundary）： 本 Skill 只针对华为云 RDS（MySQL/PostgreSQL），提供查询、诊断、运维、 参数调优与备份恢复引导。不 创建/删除实例、不 直接执行 DDL/DML、 不 修改数据库账号密码、不 管理 DRS 迁移任务。所有写操作 （重启、改参数、手动备份、恢复）都必须先得到用户明确确认。若用户询问 上述范围外的操作（如"帮我删掉这个实例""直接执行 UPDATE"），请明确告知 本 Skill 不提供该能力，并给出建议（如使用 RDS 控制台或对应管理类 skill）。

Prerequisites

1. hcloud CLI (KooCLI) v3.2.0+ installed and authenticated with an AK/SK profile — see references/cli-installation-guide.md.
2. Python 3.8+ with huaweicloudsdkrds (SDK fallback only; optional when CLI is available).
3. IAM permissions — read checks need rds:instance:list, rds:instance:get, rds:backup:list, rds:configuration:list, rds:configuration:get (or system policy RDS ReadOnlyAccess); write operations additionally need rds:instance:action, rds:backup:create, rds:configuration:update (or RDS FullAccess). See references/iam-policies.md.
4. The region of the target instance(s) — pass --cli-region={region} on every command (default cn-north-4).
5. The target instance ID ({instance_id}) — obtainable from hcloud RDS ListInstances --cli-region={region}.

Workflow

A. Guided Fault Troubleshooting (Core Scenario)

Follow this decision flow step by step. Run the listed command for the current step, read the result, then continue per the outcome. Never jump ahead; never run mutating commands without explicit user confirmation.

Symptom reported by user
   │
   ▼
Step 1  Locate the instance
   │     hcloud RDS ListInstances --cli-region={region} [--id={instance_id}]
   │     → is the instance in the list? what is "status"?
   │
   ▼
Step 2  Classify the symptom (see tables below)
   │
   ├── Instance unreachable / status not ACTIVE
   │     → check error logs (ListErrorLogsNew), check storage (Step 3),
   │       then offer restart (Step 8, CONFIRM)
   ├── Slow queries / high pressure
   │     → diagnosis summary (Step 5), slow logs (Step 6), parameter tuning (Step 7)
   ├── Disk full / storage alarms
   │     → storage check (Step 3), diagnosis insufficient_capacity (Step 5)
   ├── Primary-standby replication broken
   │     → replication status (Step 4)
   ├── Connection limit exceeded / memory overrun
   │     → diagnosis connections_exceed / mem_overrun (Step 5), tuning (Step 7)
   └── Data safety / recovery concerns
         → backup list (Step 9), restore guidance (Step 10)
   │
   ▼
Step 3..10  Execute the matching diagnostic/action steps below
   │
   ▼
Step 11  Summarize: root cause + evidence + resolution taken + prevention

B. Step Reference

1. Locate the instance — ListInstances (with --id for exact match).
2. Confirm region and instance ID with the user before any instance-scoped query; if the user gives an instance name, resolve it to an ID via ListInstances --name={name} first.
3. Check storage usage — ShowStorageUsedSpace; compare used against the volume size from ListInstances.
4. Check replication/HA — ShowReplicationStatus; for PostgreSQL also ShowReplayDelayStatus when applicable.
5. Run intelligent diagnosis — ListInstanceDiagnosis for the summary, then ListInstancesInfoDiagnosis --diagnosis={item} for detail.
6. Analyze slow/error logs — ListSlowLogs / ListErrorLogsNew with the incident time window; download files with ListSlowLogFile / DownloadSlowlog / DownloadErrorlog for deep analysis.
7. Parameter tuning — ShowInstanceConfiguration to inspect, ListConfigurations to compare templates, then UpdateInstanceConfiguration / UpdatePostgresqlParameterValue (CONFIRM with the user; note restart_required).
8. Restart instance — StartInstanceRestartAction (CONFIRM; explain downtime impact first).
9. Backup list — ListBackups; verify latest successful backup time and status before any recovery discussion.
10. Backup/restore — CreateManualBackup for a pre-change safety backup (CONFIRM), RestoreToExistingInstance / RestoreExistInstance for recovery (CONFIRM; explain that restore overwrites the target).
11. Summarize — root cause, evidence (command outputs), resolution, and prevention tips.

C. Common Fault → Step Map

Core Commands

KooCLI format: hcloud <Service> <Operation> --cli-region=<region> [--key=value]

1. Instance inventory & status (fault triage start)

hcloud RDS ListInstances --cli-region={region} --cli-output=json
# Filter by engine:
hcloud RDS ListInstances --cli-region={region} --datastore_type=MySQL --cli-output=json
hcloud RDS ListInstances --cli-region={region} --datastore_type=PostgreSQL --cli-output=json

Compact status table (name, id, status, engine, volume):

hcloud RDS ListInstances --cli-region={region} --cli-output=json \
  | jq -r '.instances[] | [.name, .id, .status, .datastore.type, .volume.size] | @tsv'

Instance status values include: ACTIVE, BUILD, REBOOTING, RESTORING, BACKING_UP, STORAGE_FULL, ERROR, FAILED, DELETING.

2. Instance detail (spec, HA, nodes)

hcloud RDS ListInstances --cli-region={region} --id={instance_id} --cli-output=json

There is no standalone ShowInstance KooCLI operation; ListInstances --id={instance_id} returns the exact instance with type (Ha/Single), nodes[].role (master/slave), nodes[].availability_zone, ha.replication_mode, volume, and datastore.

3. Storage usage / disk-full check

hcloud RDS ShowStorageUsedSpace --cli-region={region} --instance_id={instance_id} --cli-output=json

used is returned in GB; compare with the volume size from ListInstances. If used is close to size (or instance status is STORAGE_FULL), guide the user to expand the volume (StartInstanceEnlargeVolumeAction, write op — confirm) or clean up data.

4. Replication / HA status

hcloud RDS ShowReplicationStatus --cli-region={region} --instance_id={instance_id} --cli-output=json

replication_status values: normal (healthy), abnormal (broken — check error logs, network/security-group, then consider failover StartFailover, write op — confirm), unavailable.

5. Intelligent diagnosis summary & detail

hcloud RDS ListInstanceDiagnosis --cli-region={region} --engine=mysql --cli-output=json
hcloud RDS ListInstanceDiagnosis --cli-region={region} --engine=postgresql --cli-output=json

Diagnosis items (ListInstancesInfoDiagnosis --diagnosis={item}): high_pressure, lock_wait, insufficient_capacity, slow_sql_frequency, disk_performance_cap, mem_overrun, age_exceed, connections_exceed.

hcloud RDS ListInstancesInfoDiagnosis --cli-region={region} --engine=mysql \
  --diagnosis=high_pressure --cli-output=json

6. Slow SQL & error log analysis

Time format: yyyy-mm-ddThh:mm:ss+0000 (UTC — the RDS API rejects the Z suffix, e.g. 2026-08-05T00:00:00+0000). Only logs from the last month are queryable.

hcloud RDS ListSlowLogs --cli-region={region} --instance_id={instance_id} \
  --start_date={start_ts} --end_date={end_ts} --cli-output=json
hcloud RDS ListErrorLogsNew --cli-region={region} --instance_id={instance_id} \
  --start_date={start_ts} --end_date={end_ts} --cli-output=json
# Slow-log file listing + download link:
hcloud RDS ListSlowLogFile --cli-region={region} --instance_id={instance_id} --cli-output=json
hcloud RDS DownloadSlowlog --cli-region={region} --instance_id={instance_id} --cli-output=json
hcloud RDS DownloadErrorlog --cli-region={region} --instance_id={instance_id} --cli-output=json

Slow-log entries include query_sample, time (execution seconds), lock_time, rows_sent, rows_examined, database, type. Guide the user to optimize queries with high rows_examined / time (missing indexes, full table scans) and to correlate with --diagnosis=slow_sql_frequency.

7. Parameter tuning

# List parameter templates (default + custom):
hcloud RDS ListConfigurations --cli-region={region} --cli-output=json
# Inspect the parameters of a specific instance:
hcloud RDS ShowInstanceConfiguration --cli-region={region} --instance_id={instance_id} --cli-output=json

Each parameter shows value, value_range, restart_required, readonly. Common tuning targets for troubleshooting: max_connections (connection limit exceeded), innodb_buffer_pool_size / shared_buffers (memory), slow_query_log & long_query_time (slow SQL capture).

Modify parameters (WRITE — confirm with the user first; note restart_required):

# MySQL / SQLServer (general):
hcloud RDS UpdateInstanceConfiguration --cli-region={region} --instance_id={instance_id} \
  --values.max_connections=500 --cli-output=json
# PostgreSQL (single parameter):
hcloud RDS UpdatePostgresqlParameterValue --cli-region={region} \
  --instance_id={instance_id} --name=max_connections --value=500 --cli-output=json

8. Restart instance (fault recovery)

hcloud RDS StartInstanceRestartAction --cli-region={region} \
  --instance_id={instance_id} --restart={} --cli-output=json

⚠️ WRITE operation. Always explain the downtime impact and get explicit user confirmation before running. Prefer checking logs/status first.

9. Backup list

hcloud RDS ListBackups --cli-region={region} --instance_id={instance_id} --cli-output=json
# Type filter: auto | manual | fragment | incremental
hcloud RDS ListBackups --cli-region={region} --instance_id={instance_id} \
  --backup_type=manual --cli-output=json

Verify the latest COMPLETED backup and its begin_time/end_time before any recovery discussion.

10. Backup / restore

# Create a manual backup before any risky change (WRITE — confirm):
hcloud RDS CreateManualBackup --cli-region={region} \
  --instance_id={instance_id} --name={backup_name} --cli-output=json
# Restore a backup to an existing instance (WRITE — confirm; overwrites target):
hcloud RDS RestoreToExistingInstance --cli-region={region} \
  --source.instance_id={instance_id} --source.backup_id={backup_id} \
  --target.instance_id={target_instance_id} --cli-output=json

Backup name: 4–64 chars, start with a letter, only letters/digits/-/_.

Parameter Confirmation

KooCLI Command Format Standard

The KooCLI command format is:

hcloud <Service> <Operation> --cli-region=<region> [--key=value ...]

Note: this is a generic format template, not an executable command — replace <Service>, <Operation>, <region> and the parameters with real values (e.g. hcloud RDS ListInstances --cli-region=cn-north-4).

• Service name: RDS (starts with uppercase).
• Operation name: PascalCase — ListInstances, ShowReplicationStatus, ListSlowLogs, UpdateInstanceConfiguration.
• Region: always pass --cli-region=<region>.
• Simple parameters use --key=value, e.g. --instance_id=....
• Nested/object parameters use indexed keys, e.g. --values.max_connections=500 or --source.instance_id=....

Reference Documents

• IAM Policies — Least-privilege IAM permissions
• CLI Installation Guide — hcloud CLI setup and authentication
• Troubleshooting Guide — Detailed symptom→diagnosis→resolution scenarios
• Verification Method — How to verify the skill works
• Dataflow Diagram — Request flow diagram
• Acceptance Criteria — Acceptance test checklist