Observe and maintain a running StarRocks cluster: health metrics, compaction, tablet repair, statistics, and routine upkeep.
When to Use
- Checking cluster/BE health, disk usage, or tablet distribution
- Inspecting running queries and killing runaway ones
- Diagnosing or triggering compaction
- Repairing or rebalancing tablets/replicas
- Collecting statistics for the cost-based optimizer (CBO)
- Planning daily/weekly/monthly maintenance
Related skills: [starrocks-cluster-setup] for node topology and config; [starrocks-resource-management] for workload isolation; [starrocks-backup-recovery] for snapshots; [starrocks-query-optimization] for query-level diagnosis.
Key Metrics to Monitor
Cluster Health:
-- Backend status and load
SHOW PROC '/backends';
-- Disk usage per BE
SHOW PROC '/statistic';
-- Tablet distribution
SHOW PROC '/statistic/{be_id}';
Query Performance:
-- Recent queries
SELECT * FROM information_schema.loads ORDER BY StartTime DESC LIMIT 10;
-- Running queries
SHOW PROCESSLIST;
-- Kill slow query
KILL QUERY connection_id;
-- Query profile (after execution)
SHOW QUERY PROFILE '<query_id>';
Data Loading:
-- Recent load jobs (Broker Load / INSERT / Spark Load)
SHOW LOAD ORDER BY CreateTime DESC LIMIT 10;
-- Failed loads
SHOW LOAD WHERE State = 'CANCELLED';
-- Stream Load / Routine Load are not shown by SHOW LOAD; query information_schema instead
SELECT * FROM information_schema.loads ORDER BY create_time DESC LIMIT 10;
Compaction Management
-- Check compaction status
SHOW PROC '/compactions';
-- Trigger manual compaction (v3.1+): whole table, a partition, or cumulative
ALTER TABLE table_name COMPACT;
ALTER TABLE table_name COMPACT partition_name;
ALTER TABLE table_name CUMULATIVE COMPACT;
-- Compaction thresholds are BE configuration, NOT table properties.
-- Tune them in be.conf or dynamically via the be_configs table, e.g.:
UPDATE information_schema.be_configs
SET value = "100"
WHERE name = "max_cumulative_compaction_num_singleton_deltas";
Tablet Repair and Rebalancing
-- Check tablet health
ADMIN SHOW REPLICA STATUS FROM table_name;
-- Check tablet distribution
SHOW TABLET FROM table_name;
-- Trigger tablet repair (if replicas are corrupted)
ADMIN REPAIR TABLE table_name;
-- Manual tablet rebalance (usually automatic)
ADMIN REBALANCE DISK;
Routine Maintenance Tasks
Daily:
- Monitor load job failures
- Check query latencies
- Review disk usage
Weekly:
- Analyze slow queries
- Review partition retention
- Check compaction lag
Monthly:
- Review and optimize bucket counts
- Analyze data skew
- Update statistics (if using CBO)
Statistics Collection (for CBO)
-- Collect statistics for better query planning (manual, synchronous by default)
ANALYZE TABLE table_name;
-- Collect for specific columns
ANALYZE TABLE table_name (col1, col2);
-- Automatic full collection is ON by default (no per-table property needed);
-- it is governed by FE config (e.g. enable_collect_full_statistic,
-- statistic_auto_analyze_start_time / _end_time). Define a recurring custom
-- collection job with CREATE ANALYZE if you need finer control.
-- Check collection status and statistics metadata
SHOW ANALYZE STATUS;
SHOW META;
Troubleshooting Commands
-- Check FE logs
ADMIN SHOW FRONTEND CONFIG LIKE 'log%';
-- Show system variables
SHOW VARIABLES;
-- Show session variables
SHOW SESSION VARIABLES;
-- Check metadata inconsistencies
ADMIN CHECK TABLET (tablet_id);
-- Show query plan and profile
EXPLAIN VERBOSE SELECT ...;
EXPLAIN COSTS SELECT ...;