PAI-DLC Job Diagnostics and Health Inspection

Read-only diagnostic analysis for PAI-DLC distributed training jobs, covering

three scenarios:

• Queuing-stuck root cause analysis — quota check, node scheduling

diagnosis, hyper-node availability

• Failed-job localization — failure classification, logs/events evidence

chain, root cause identification

• Cluster health inspection — training throughput, hang detection,

SanityCheck, restart stability

Architecture: PAI-DLC Job (read-only queries) + PAI Studio Resource

Diagnosis API (queuing scenario).

0. Dependencies

This skill performs read-only diagnostics only. All write operations

(create / update / stop jobs, resource discovery, etc.) live in the companion

skill alibabacloud-pai-dlc-job. The two skills are complementary in

responsibility and share a common field contract.

Discover and install the prerequisite skill:

# Discover available skills
npx skills add aliyun/alibabacloud-aiops-skills --skill alibabacloud-find-skills
# Install the alibabacloud-pai-dlc-job skill itself
npx skills add aliyun/alibabacloud-aiops-skills --skill alibabacloud-pai-dlc-job

Cross-skill field contract: The --job-id / --pod-id values this skill

consumes are produced verbatim by alibabacloud-pai-dlc-job via

list-jobs / get-job --cli-query "Pods[0].PodId" — no transformation needed.

--region / --workspace-id follow the same resolution rules in both skills.

Installation Requirements

> Pre-check: Aliyun CLI >= 3.3.1 required

> Run aliyun version to verify >= 3.3.1. If not installed or version too low,

> see references/cli-installation-guide.md.

> Then [MUST] run aliyun configure set --auto-plugin-install true.

> Note on --user-agent: Every API-invoking aliyun command in this skill MUST

> include --user-agent AlibabaCloud-Agent-Skills/alibabacloud-pai-dlc-job-diagnostics. Client-side helpers

> (aliyun version, aliyun configure ..., aliyun plugin ...,

> aliyun --help) do not invoke remote APIs and therefore do not require

> the flag.

aliyun version
aliyun configure set --auto-plugin-install true
aliyun plugin update
aliyun pai-dlc --help
aliyun paistudio --help >/dev/null 2>&1 || aliyun plugin install --names aliyun-cli-paistudio

aliyun configure ai-mode enable
aliyun configure ai-mode set-user-agent --user-agent "AlibabaCloud-Agent-Skills/alibabacloud-pai-dlc-job-diagnostics"
# After session: aliyun configure ai-mode disable

Authentication

> Pre-check: Alibaba Cloud Credentials Required

>

> Security Rules:

> - NEVER read, echo, or print AK/SK values

> - NEVER ask the user to input AK/SK directly

> - ONLY use aliyun configure list to check credential status

>

> ```bash

> aliyun configure list

> ```

> Check the output for a valid profile (AK, STS, or OAuth identity).

>

> If no valid profile exists, STOP here.

> 1. Obtain credentials from Alibaba Cloud Console

> 2. Configure credentials outside of this session

> 3. Return and re-run after aliyun configure list shows a valid profile

RAM Permissions

> [MUST] Permission Failure Handling: When any command fails due to permission errors:

> 1. Read references/ram-policies.md for the full permission list

> 2. Use ram-permission-diagnose skill to guide the user

> 3. Pause and wait until the user confirms permissions have been granted

Parameter Confirmation

> IMPORTANT: Parameter Confirmation — Before executing any command,

> ALL user-customizable parameters (RegionId, JobId, etc.) MUST be confirmed with the user.

Entry Routing

When a diagnostic request arrives, first call get-job to fetch job status,

then route by status:

Edge case — job was queuing but is now Stopped/Succeeded: If the user

describes the job as "stuck in queue" but get-job shows Stopped or

Succeeded, still route to Scenario 1 (queuing analysis) but expect the

resource diagnosis API to return HTTP 400. Follow the "Fallback on API Failure"

procedure in Scenario 1.

Users may also directly request a specific scenario (e.g., "run a health

inspection" even when status is not Running).

Diagnostic Toolbox

PAI-DLC Read-Only Commands

aliyun pai-dlc get-job --region <r> --job-id <id> \
  --user-agent AlibabaCloud-Agent-Skills/alibabacloud-pai-dlc-job-diagnostics
aliyun pai-dlc get-job-events --region <r> --job-id <id> --max-events-num 50 \
  --user-agent AlibabaCloud-Agent-Skills/alibabacloud-pai-dlc-job-diagnostics
aliyun pai-dlc get-pod-events --region <r> --job-id <id> --pod-id <pod> --max-events-num 20 \
  --user-agent AlibabaCloud-Agent-Skills/alibabacloud-pai-dlc-job-diagnostics
aliyun pai-dlc get-pod-logs --region <r> --job-id <id> --pod-id <pod> --max-lines 100 \
  --user-agent AlibabaCloud-Agent-Skills/alibabacloud-pai-dlc-job-diagnostics
aliyun pai-dlc list-job-sanity-check-results --region <r> --job-id <id> \
  --user-agent AlibabaCloud-Agent-Skills/alibabacloud-pai-dlc-job-diagnostics
aliyun pai-dlc get-job-sanity-check-result --region <r> --job-id <id> --sanity-check-number 1 \
  --user-agent AlibabaCloud-Agent-Skills/alibabacloud-pai-dlc-job-diagnostics

PAI Studio Resource Diagnosis (queuing scenario only)

aliyun paistudio GET /api/v1/quotas/{quota_id}/workloads/{job_id}/diagnosis \
  --region <r> --header "Content-Type=application/json" --force \
  --user-agent AlibabaCloud-Agent-Skills/alibabacloud-pai-dlc-job-diagnostics

Hard constraint: quota_id MUST come from get-job's ResourceId field.

If ResourceId is empty (public pay-as-you-go), this API is unavailable.

Full API structure: see references/resource-diagnosis-api.md.

Scenario 1: Queuing-Stuck Root Cause Analysis

Trigger: job status = Queuing / Creating and user reports it cannot be scheduled.

Tools: get-job → paistudio resource diagnosis → (optional) get-job-events.

Hard constraints:

• ResourceId empty → resource diagnosis unavailable; mine events for clues
• ResourceId non-empty → resource diagnosis API is the primary instrument

> CRITICAL: pai-dlc vs paistudio — two different products

>

> | Product | Scope | Commands |

> |---------|-------|----------|

> | pai-dlc | Job/Pod lifecycle (GetJob, GetJobEvents, ListJobs, GetPodLogs) | aliyun pai-dlc get-job ... |

> | paistudio | Platform-level services including resource diagnosis | aliyun paistudio GET /api/v1/quotas/... |

>

> The resource diagnosis API belongs to paistudio, NOT pai-dlc.

> Do NOT call pai-dlc GetResourceQuota, pai-dlc ListResourceQuotas,

> or any pai-dlc GET /api/v1/resourcequotas/... — these are wrong APIs

> and will fail. The correct command is:

> ```bash

> aliyun paistudio GET /api/v1/quotas/{quota_id}/workloads/{job_id}/diagnosis ...

> ```

Pattern knowledge: resource diagnosis returns 4 checks

(self_quota / ancestor_quota / user_limit / queue_strategy), plus node

scheduling and hyper-node analysis. Common patterns:

references/diagnostic-patterns.md §1.

Agent latitude: decide whether to compute the quota gap, whether to pull

events for corroboration, and how verbose the report should be.

Fallback on API Failure

The PAI Studio resource diagnosis API may fail with HTTP 400/404 when the

job is no longer in an active queuing state (e.g., already Stopped by user).

In this case:

1. Explicitly declare the API failure in the report:

"Resource diagnosis API unavailable: HTTP {code} — {error message}"

1. Perform qualitative analysis using only these data sources:
• get-job → ResourceRequest (GPU/CPU/Memory demand per pod)
• get-job → PodCount × per-pod resources = total demand
• get-job → EcsSpec (instance type and its per-node capacity)
• get-job-events → scheduling event timeline and queuing duration
2. Prohibited language: NEVER use hedging words such as "possibly",

"might", "perhaps", "may not be sufficient". State only confirmed facts:

• Total GPU demand = PodCount × RequestGPU = {computed value}
• Instance type = {EcsSpec}
• Queuing duration = {computed hours/minutes}
• Job final status = {Stopped/Succeeded/etc.}

1. Mandatory output structure when API is unavailable:

## Resource Diagnosis (API Unavailable — Configuration-Based Analysis)

- Diagnosis API: unavailable (HTTP {code}: {message})
- Resource demand: {PodCount} pods × {GPU/pod} GPU = {total} GPU cards
- Instance type: {EcsSpec}
- Queuing duration: {hours}h {minutes}m
- Quota ID: {ResourceId}
- Job final status: {status} ({ReasonCode})
- Conclusion: The job requested {total} GPU cards which could not be
  fulfilled within the queuing window before the job was {stopped/completed}.

Scenario 2: Failed-Job Localization

Trigger: job status = Failed.

Tools: get-job → (as needed) get-job-events / get-pod-events / get-pod-logs.

Hard constraints:

• Stopped is not a failure — do not diagnose
• [OUTPUT GUARD — RECOMMENDATIONS STRICTLY FORBIDDEN] When the failure

falls into any of the categories below, the diagnostic report **MUST strictly

omit** any "Recommendations" / "Suggested fixes" / "Next steps" / "Solution"

section, and output only the root cause and objective facts:

• ResourceAllocateFailed (insufficient resources)
• Job preempted / evicted (ReasonMessage contains preempted or evicted)
• Spot instance reclamation
• Report MUST end on the diagnostic conclusion. No sentence anywhere in

the output may instruct the user to change configuration, request more

quota, switch instance types, add retry logic, or modify any job parameter.

• Negative examples (ALL forbidden):
• "Recommend using pay-as-you-go instances"
• "Consider requesting more quota"
• "Try switching to another zone"
• "Disable preemptible jobs by setting EnablePreemptibleJob: false"
• "Implement checkpointing and retry logic"
• "Solution: Use guaranteed quota instead of oversold quota"
• Positive examples (allowed):
• "Root cause: spot instance reclaimed by the cloud platform, triggering pod eviction and job failure."
• "The job failed due to preemption/eviction of pods. ReasonCode: JobFailed."
• Only permitted user-facing suggestion: "For quota policy adjustments, please contact your platform administrator."

Mandatory output template for preemption/eviction/resource-shortage failures:

## Diagnosis Conclusion

- Failure reason: {classification} ({ReasonCode}: {ReasonMessage})
- Affected pods: {pod list with SubStatus}
- Timeline: {key timestamps from events}
- Evidence: {quoted ReasonMessage or event details}

For quota policy or resource allocation adjustments, please contact your
platform administrator.

Pattern knowledge: failure-classification priority

(network > image > runtime > resource > config > system), exit-code meanings,

keyword-matching rules. See

references/diagnostic-patterns.md §2.

Agent latitude: when ReasonCode is clear, logs may be unnecessary; when

logs already explain the issue, events may be unnecessary. Decide investigation

depth based on information sufficiency.

Scenario 3: Health Inspection

Trigger: job status = Running and user requests inspection / health check.

Tools: get-job + get-job-events + get-pod-logs

• list-job-sanity-check-results.

Execution steps:

1. get-job → obtain status, WorkspaceId, Pod list, whether EnableSanityCheck is set
2. get-job-events → event-chain analysis (scheduling, restarts, etc.)
3. get-pod-logs → target the master pod (rank=0) first; extract structured training metrics if present
4. list-job-sanity-check-results → execute only if EnableSanityCheck=true in job settings
5. Generate console links (MANDATORY — must always output both links):
• Job overview: https://pai.console.aliyun.com/?regionId={region}&workspaceId={workspace_id}#/dlc/jobs/{job_id}/overview
• Monitoring dashboard: https://pai.console.aliyun.com/?regionId={region}&workspaceId={workspace_id}#/dlc/jobs/{job_id}/monitor
• Mandatory closing notice (append verbatim to every inspection report):

```

> GPU/memory real-time resource utilization metrics require the monitoring

> dashboard above. This skill's CLI commands do not support fetching

> runtime utilization data directly.

```

This step is mandatory regardless of job status (Running, Succeeded, or

any other state). Even if the job has already completed, the user needs

the monitoring link to review historical resource utilization.

Dimension matrix (mandatory vs optional):

Note on resource utilization: GPU/memory metrics are NOT available via this

skill's CLI commands. The monitoring dashboard link (step 5) and its closing

notice are mandatory in every inspection report — do not omit them.

Interpretation rules per dimension: see

references/healthcheck-dimensions.md.

Agent latitude: kilo-card jobs focus on Hang + SanityCheck; small jobs

look at training throughput from logs. Decide depth and report verbosity

from scale and user intent.

Best Practices

1. get-job first, route second — never assume status; query and then route
2. Cap response size — --max-lines 100 / --max-events-num 50 to avoid context blow-up
3. Find the problem pod — among get-job Pods, focus on Status=Failed/Unknown or those with ReasonMessage
4. Read log tail — errors usually live in the last few dozen lines; no need to pull the whole log
5. Exit codes are clues, not conclusions — exit 137 may be OOM or external kill; combine with context
6. Don't over-diagnose — when ReasonCode already states the cause, skip the full log dump
7. Console links — generate both overview and monitoring links so the user can jump into the console for details and resource utilization metrics
8. Read-only — this skill MUST NEVER execute stop / update / create
9. Report summary — present a per-dimension rating table at the end of the report:

Reference Links