Why is my job running slowly?
Slow jobs in MaxCompute typically have one of three causes: resource contention, a job-level issue such as data skew or data bloat, or a mode rollback from MaxCompute Query Acceleration (MCQA) to normal mode. Start with the most likely cause for your billing model.
| Cause | When it happens | What to do |
|---|---|---|
| Insufficient resources | High resource usage, many concurrent jobs, or low job priority | Check quota consumption; optimize the job or adjust your resource configuration |
| Job exceptions | Data skew, inefficient user-defined functions (UDFs), or data bloat | Use Logview to diagnose the root cause; optimize your SQL |
| Mode rollback | An MCQA job falls back to normal mode | Review MCQA eligibility; check whether your job processes large data volumes without returning query results |
Insufficient resources
Subscription computing resources: If your quota is consistently at or near its upper limit, jobs wait in a queue for resource allocation — especially when runtime data volume is large, resource requests are high, or job priority is low. Check your current quota usage in the MaxCompute console using the instructions in the "View quota consumption" section of Manage quotas in the new MaxCompute console. If usage stays at the ceiling for extended periods, either optimize job execution or change your resource configuration. For optimization options, see Manage jobs in the new MaxCompute console.
Pay-as-you-go computing resources: The pay-as-you-go resource pool is shared across all users. Jobs compete for resources on demand, and you cannot reserve a fixed amount. When many jobs run simultaneously, resource contention causes slower execution.
Job exceptions
Data skew, low UDF execution efficiency, and data bloat are common causes of slow SQL jobs. Use Logview to identify which factor is affecting your job — see Diagnose a job that runs slowly for step-by-step instructions. Once you identify the root cause, see Optimize SQL statements to resolve it.
Mode rollback
MaxCompute Query Acceleration (MCQA) speeds up interactive queries on small data volumes compared to normal mode. However, MaxCompute cannot guarantee that a job always runs in MCQA mode. In some cases, an MCQA job rolls back to normal mode, which is slower than expected.
Jobs that process large data volumes and do not need to return query results can only run in normal mode — MCQA is not applicable. In most cases, if resources and jobs are normal, execution times in normal mode are generally consistent.