trino-self-healing-platform

What problem does it solve?

It reduces on-call burden by autonomously detecting and recovering from common Trino cluster failures (hung queries, worker loss, OOM-killed queries) and by proactively maintaining query performance and Iceberg health.

Core Features & Use Cases

• Hung Query Killer: Classifies queries by source (interactive vs batch) and kills those exceeding time thresholds.
• Memory Pressure Relief: Detects critical memory pressure and terminates low-priority queries to restore stability.
• Iceberg Auto-Maintenance: Triggers Iceberg compaction and stale statistics ANALYZE based on small-file and staleness signals.
• RCA Generation: Produces structured incident root-cause analysis using a Claude-based workflow when alerts fire.
• Use Case: When Prometheus/AlertManager signals elevated latency or query failures, run this skill to kill stuck queries, relieve memory pressure, compact/ANALYZE impacted Iceberg tables, and generate an RCA report for operators.

Quick Start

Ask the agent to enable the Trino self-healing watchdog and configure it to run every 15 minutes against your Trino coordinator, using AlertManager webhooks and Airflow scheduling.