Skip to main content
This alert indicates that no barrier has been committed in your cluster for longer than the configured threshold. Data freshness can be affected while barriers remain pending.

​
Common triggers

  • Streaming graph bottlenecks. Typical causes include join amplification, insufficient resources, and suboptimal streaming queries such as over-window queries and joins.
  • Compaction write stalls that result in longer barrier synchronization times.

​
Diagnosis

  • Check CPU and memory utilization for all nodes. If the nodes are fully utilized, the cluster might have insufficient resources.
  • Run SHOW JOBS and check for jobs that are being created. Backfilling a newly created job can increase pressure on the cluster.

​
Resolution

If node resources are fully utilized or a backfill is running, scale out the cluster to alleviate the pressure. For more information, see Scale a cluster manually.