When we changed how Materialize manages memory via swap (details in our blog post), we also introduced a new Memory Utilization metric. This better reflects a cluster’s capacity, since we can now spill most memory to disk before a cluster OOM’s.
info
Under the hood, Memory Utilization is defined as (Memory Usage + Swap Usage) / (Memory Size + Swap Size). Within our mz_cluster_replica_utilization and mz_cluster_replica_utilization_history internal system tables, memory utilization is calculated using the heap_percent field.
If you’d like to run the query yourself against your cluster replicas, the query is:
SELECT replica_id, process_id, heap_percent FROM mz_internal.mz_cluster_replica_utilization;We recommend updating your alerting thresholds in your preferred monitoring tool with this updated query.
You can find more details in our Troubleshooting guide.
Our Console will be updated to display Memory Utilization using the heap_percent field to simplify the calculation in an upcoming release.
Any additional questions? Please ask Matty (via the chatbot in the right hand corner of our website) or contact support to be connected with our team.