@Rocky Does StarRocks expose table level query latency metrics ?

StarRocks does not provide a direct "table-level query latency" metric in its standard Prometheus monitoring out of the box. Query latency metrics in Prometheus (

starrocks_fe_query_latency_ms

) are typically exposed at the cluster or resource group level. However, you can obtain table-level performance data or derive query latency through the following methods: 1. Audit Logs (Recommended for Latency) The most common way to track latency per table is by using the AuditLoader plugin to import

fe.audit.log

into a StarRocks table (usually

starrocks_audit_db*.starrocks_audit_tbl*

). * Metric available:

queryTime

(in milliseconds). * Granularity: Each record includes the SQL statement (

stmt

), user, and database. * Usage: You can run SQL queries against this audit table to aggregate average or P95 latency. Since a single query can join multiple tables, StarRocks doesn't always map a query to a "single" table automatically in the logs, but you can filter or parse the

stmt

field to attribute latency to specific tables. 2. Table-level Metrics via

/metrics

You can enable detailed table metrics by adding the

with_table_metrics=all

parameter to the FE metrics endpoint:

http://<FE_IP>:<FE_HTTP_PORT>/metrics?with_table_metrics=all

* What it provides: This exposes table-level throughput and resource usage (e.g.,

table_scan_rows

,

table_scan_bytes

,

table_load_rows

). * Limitation: It does not currently include a specific "query latency" histogram for each table, as latency is measured at the query execution level, not the individual table level. 3. Resource Groups If you need to monitor latency for a specific set of tables used by a particular business line, you can isolate those queries into a Resource Group. StarRocks provides Prometheus metrics for latency per resource group: *

starrocks_fe_query_resource_group_latency

(Histogram/Percentiles) Summary Table Method | Metrics Provided | Granularity | Recommended For Prometheus | Throughput (Scan/Load) | Table-level | Capacity planning, finding "hot" tables Audit Logs | Latency (

queryTime

) | Query-level | Analyzing actual query performance per table Resource Groups | Latency | Group-level | SLA monitoring for specific workloads To enable table-level metric collection in your FE configuration, ensure

enable_table_metrics_collect = true

(which is the default). References • administration/management/monitoring/metrics.md • administration/management/FE_configuration.md