Job Insights gives you a single place to investigate why a job ran slowly, where its resources went, and how it compares to previous runs of the same job. For any job in the MaxCompute console, you can:
Review basic job information, runtime parameters, and intelligent diagnostic recommendations.
Trace resource usage over the full job lifecycle and pinpoint quota contention at any moment during execution.
Compare the current job against similar historical jobs to distinguish a one-off slowdown from a systemic performance issue.
Prerequisites
Before you begin, ensure that you have:
A MaxCompute workspace with at least one submitted job
Access to the MaxCompute console
Open job insights
Log on to the MaxCompute console and select a region in the upper-left corner.
In the left-side navigation pane, go to Workspace > Jobs.
In the job list, click Insight in the Actions column for the target instance ID. The Job Insights page opens with three tabs: Job Summary, Resource Consumption, and Similar Jobs.
Job summary
The Job Summary tab shows basic job information, runtime parameters, and intelligent diagnostic results.
Two additional actions are available in the upper-right corner of the tab:
Diagnose: Triggers a real-time intelligent diagnosis for SQL and SQLRT jobs. Click it at any time to refresh the diagnostic results. For non-SQL job types, this option is not shown.
LogView: Opens LogView to view detailed execution information for the job. For more information, see Use LogView V2.0 to view job information.
Basic information
| Parameter | Description |
|---|---|
| Job type | The compute engine that ran the job. Valid values: SQL, SQLRT (MaxCompute Query Acceleration (MCQA) SQL job), LOT (MapReduce), CUPID (Spark or Mars), Algo_Task (machine learning), GRAPH (graph computing). |
| Job owner | The account responsible for the job. |
| Job priority | The scheduling priority of the job. For details, see Job priority. |
| Billing method | How the job is billed. Subscription: uses subscription computing quota with no additional postpaid charges. Pay-as-you-go: uses pay-as-you-go computing quota and generates postpaid charges. |
| Submission time | When the job was submitted. |
| Start time | When the job obtained its first computing resource. For short-lived jobs or DDL statements that do not consume computing resources, the submission time is used instead. |
| Job status | The overall status of the job. Valid values: Running (running or not yet complete), Success (complete), Failed (failed to run), Canceled (canceled), Submitted (waiting for computing resources). |
| Wait time | The time between job submission and the start of execution. A long wait time typically indicates that computing quota is constrained and resources are being held by other jobs. |
| Execution duration | The time from job start to job end. |
| End time | When the job finished running. |
| Total run time | The total period for which the job runs from submission to end time. |
Job status reflects the overall state of the entire job. A single job can involve many concurrent processes, each with its own sub-status. Use LogView to inspect sub-process details. For more information, see Use LogView V2.0 to view job information.
Runtime parameters
| Parameter | Description |
|---|---|
| Project | The project that submitted the job. |
| CU quota | The computing quota used to run the job. |
Intelligent diagnostic information
When you open the Job Analysis page for an SQL or SQLRT job, a real-time diagnosis runs automatically. Results and optimization recommendations appear on the Job Summary tab.
Intelligent diagnostics is only available for SQL and SQLRT jobs. For other job types, the diagnostic section is not displayed.
Click Diagnose in the upper-right corner at any time to fetch the latest diagnostic results.
For a full list of diagnostic types and remediation guidance, see Intelligent diagnostics for jobs.
Resource consumption
The Resource Consumption tab shows how computing resources were used throughout the job lifecycle.
| Chart | Description |
|---|---|
| Resource usage during job lifecycle | A time-series chart showing CU usage and waiting CU at both the job level and the computing quota level. |
| Resource allocation of computing quota at a specific time | A snapshot of the quota's resource allocation at a selected moment, including the number of running and waiting jobs broken down by priority. |
Detect quota contention
If the job-level CU usage is low while the quota-level CU usage is high and continuously hits its upper limit, other jobs are preempting resources from the current job.
To investigate further:
Click any point on the horizontal axis of the resource usage chart to open the quota allocation snapshot for that moment.
Review the number and priority distribution of running and waiting jobs.
Click a priority color block to see the specific jobs assigned to it — these are the jobs consuming resources that the current job is waiting for.
To resolve contention, adjust job priorities or reallocate quota resources:
For best practices on interpreting resource usage data, see Best practices for job-level resource analysis.
Similar jobs
The Similar Jobs tab compares the current job against other jobs that share the same structure or origin, across a selected time range. Use it to determine whether a performance anomaly is isolated or recurring.
Similarity standards
MaxCompute uses two standards to identify similar jobs:
Same signature (default): Matches SQL jobs that share the same query signature as the current job.
Same ExtNodeId: Matches jobs with the same external node ID as the current job (for example, a DataWorks node ID). For more information about DataWorks node IDs, see Configure basic properties.
Select a time range of 1d, 7d, or 14d. The default is the range selected on the Jobs page.
Comparison metrics
The chart displays the following metrics for each similar job:
| Metric | Description |
|---|---|
| Waiting duration | Time from job submission to the start of execution. |
| Execution duration | Time from job start to job end. |
| CU-hour | Total resource consumption, calculated as MAX(CPU hours, ceil(Memory hours / 4)). 1 CPU-hour = 1 CPU core running for 1 hour (CPU hours = CPU cores x duration). 1 Memory-hour = 1 GB of memory used for 1 hour (Memory hours = memory size x duration). |
| Scan size | The amount of data scanned by the job. |
Job fields
Each job in the comparison list includes the following fields:
| Field | Description |
|---|---|
| Instance ID | The instance ID of the similar job. Click it to open that job's Job Insights page. |
| Submitted at | When the job was submitted. |
| Job type | The type of the job. |
| Job priority | The scheduling priority of the job. |
| Initiating project | The project that submitted the job. |
| Computing quota | The computing quota used to run the job. |
| Job owner | The account responsible for the job. |