Metrics
VeloDB Cloud provides built-in metrics in the console and exposes raw metrics through the Metrics API. Use the console to inspect warehouse and cluster health interactively, or connect the Metrics API to your own Prometheus and Grafana stack for centralized observability.
View metrics in the console
On the Metrics page you can:
- View metrics by warehouse or by cluster.
- Use Starred to pin the metrics you care about across warehouses and clusters so they display together.
- Adjust the time selector to look at historical data up to the past 15 days.
- Enable auto-refresh for near-real-time updates (5-second interval).
Metrics are split into two categories: Resource Metrics (physical resource utilization) and Service Metrics (query and workload performance).
Resource Metrics
Resource metrics track physical utilization across warehouse and cluster nodes. They help you judge whether a warehouse or cluster is healthy in a given time range, and whether historical or current queries are affecting performance. This is useful input when planning to scale up, scale down, or optimize SQL.
The Scope column shows whether a metric applies at warehouse level, cluster level, or both. The Available from column shows the minimum engine (core) version required. A blank cell means the metric has been available since the first release.
| Metric | Scope | Unit | Available from | What it shows |
|---|---|---|---|---|
| CPU Utilization | Warehouse, Cluster | % | CPU utilization across all nodes. Useful for finding quiet windows for scaling or other resource-heavy operations. | |
| Memory Usage | Warehouse, Cluster | GB | Memory consumed across all nodes. Sustained high usage signals a need to scale or tune workloads. | |
| Memory Utilization | Warehouse, Cluster | % | Memory utilization across all nodes. Helps detect sustained memory pressure. | |
| I/O Utilization | Warehouse, Cluster | % | Disk I/O utilization across nodes. Sustained high values suggest a storage bottleneck. | |
| Disk Cache Utilization | Cluster | % | 4.0.4 | Local cache space utilization. Helps assess whether cache capacity is sufficient. |
| Disk Cache Hit Rate | Cluster | % | Share of reads served from local cache. Low values suggest reviewing cache size or policy. | |
| Disk Cache Read/Write IOPS | Cluster | IOPS | Local cache read and write IOPS. Reflects cache-layer request load. | |
| Disk Cache Read/Write Throughput | Cluster | MB/s | Local cache read and write throughput. Reflects cache-layer data transfer. | |
| Remote Storage Read/Write IOPS | Cluster | IOPS | 4.0.4 | Object storage read and write IOPS. Reflects remote storage request load. |
| Remote Storage Read/Write Throughput | Cluster | MB/s | 4.0.4 | Object storage read and write throughput. |
| Network Inbound/Outbound Throughput | Warehouse, Cluster | MB/s | Inbound and outbound network throughput. Helps identify bandwidth bottlenecks. | |
| Unreachable Node Count | Warehouse, Cluster | Host-level unreachable nodes, based on host availability checks. | ||
| Inactive FE Service Node Count | Warehouse | FE nodes whose service process is not active. Surfaces systemd or process failures. |
Service Metrics
Service metrics track query and workload behavior: how fast queries run, how many succeed, and how write paths behave.
| Metric | Scope | Unit | Available from | What it shows |
|---|---|---|---|---|
| Queries per Second | Warehouse, Cluster | QPS | Query requests processed per second. Peak QPS is a useful input when sizing a cluster. | |
| Query Success Rate | Warehouse, Cluster | % | Share of successful queries. Abnormal drops may indicate a cluster or node failure. | |
| Average Query Latency | Cluster | ms | Average query latency. A rise signals a cluster-wide slowdown to investigate. | |
| P99 Query Latency | Cluster | ms | 99th-percentile query latency. Reflects tail latency and slow-query risk. | |
| Rows Loaded per Second | Warehouse, Cluster | Row/s | Rows written per second by load jobs. Tracks ingestion throughput. | |
| Load Throughput | Warehouse, Cluster | MB/s | Data written per second by load jobs. Tracks ingestion bandwidth. | |
| Active Connections | Warehouse | Active connections to the warehouse, across all clusters. | ||
| Finished Load Job Rate | Warehouse | Recent completion rate of load jobs. Sharp changes may indicate a workload shift. | ||
| Average Load Job Latency | Warehouse | ms | Average execution latency of load jobs. Lower means data becomes queryable sooner. | |
| Stream Load Request Rate | Warehouse | Stream Load request rate. Reflects real-time ingestion traffic. | ||
| Broker Load Job Rate | Warehouse | Broker Load job rate. Reflects batch ingestion traffic. | ||
| Insert Into Job Rate | Warehouse | Insert Into write rate. Reflects SQL-based ingestion traffic. | ||
| Routine Load Jobs by Status | Warehouse | Count of Routine Load jobs in each status. Tracks continuous-ingestion health. | ||
| Compaction Score | Cluster | Data-file merge pressure. Higher scores can affect write or query performance. | ||
| Routine Load Lag | Warehouse | 4.0.5 | Routine Load consumption lag, for example from Kafka. Reflects data freshness. | |
| Routine Load Aborted Transactions | Warehouse | 4.0.5 | Routine Load transactions aborted in the last 5 minutes. Growth points to ingest failures. | |
| Routine Load Jobs in ABNORMAL_PAUSED State | Warehouse | 4.0.5 | Routine Load jobs paused abnormally. Usually needs investigation and a manual resume. | |
| Virtual Cluster Active-Standby Switches | Warehouse | 4.1.7 | Active-standby switch events for virtual clusters, over at least a 10-minute window. | |
| Data Size | Warehouse | GB | Historical trend of the warehouse's data size. | |
| MV Refresh Skipped Tasks | Warehouse | tasks | 4.1.8 | Materialized view refresh tasks skipped in the window. Surfaces scheduling delay or backlog. |
| MV Refresh Queue Length | Warehouse | tasks | 4.1.8 | Running plus pending materialized view refresh tasks. Tracks refresh backlog. |
| MV Refresh Success Rate | Warehouse | % | 4.1.8 | Share of successful materialized view refreshes. Windows with no refreshes show 100%. |
| MV Refresh Duration | Warehouse | ms | 4.1.8 | P75 and P95 materialized view refresh duration. |
Job-level metrics for individual Routine Load and Warmup jobs appear on each job's own page, not here. To alert on them, see Alerts.
Next steps
- Metrics API: scrape these metrics into your own Prometheus or Grafana stack.
- Alerts: get notified when a metric crosses a threshold.