The Complete Production SOP: How to Deploy and Run Apache Cassandra Nodetool & Compaction Strategy in Autonomous AI Agents

Autonomous runtimes like Claude Code and Cursor scan defined directories (`.claude/skills`, `.cursor/rules`, or `~/.agents/skills`) to dynamically augment agent system prompts. By placing this skill into the target directory, the agent loads procedural bounds for Cassandra maintenance operations and registers triggers like 'run cassandra nodetool repair' without requiring an IDE restart.

---
name: "cassandra-nodetool-compaction-tuning"
description: "Maintain cluster health, repair anti-entropy, and tune SSTable compactions in Apache Cassandra."
metadata:
  author: "Apache Software Foundation"
  version: "1.0.0"
  runtime_compatibility:
    - "codex"
    - "claude-code"
    - "cursor"
    - "windsurf"
triggers:
  - "run cassandra nodetool repair"
  - "fix cassandra tombstones"
  - "choose cassandra compaction strategy"
permissions:
  fs_read: true
  fs_write: "workspace"
  shell_exec: false
  net_access: false
---

Autonomous agents should interact with cluster parameters via structured configuration templates rather than querying production clusters blindly. We establish a local profile schema that informs the agent of the target keyspaces, replication factor, and gc_grace_seconds so its generated nodetool routines match production settings.

{
  "clusterName": "prod-analytics-cluster",
  "jmxHost": "127.0.0.1",
  "jmxPort": 7199,
  "keyspaces": [
    {
      "name": "telemetry",
      "tables": ["events_by_device", "device_heartbeats"],
      "defaultGcGraceSeconds": 864000,
      "currentCompaction": "SizeTieredCompactionStrategy"
    }
  ]
}

When prompted with a trigger like 'run cassandra nodetool repair', the agent activates the skill SOP. Instead of executing an unthrottled global repair (`nodetool repair`), the agent follows Rule 1 of the skill to construct a partitioner-range repair command with parallelism constraints, avoiding token-range collision and node saturation.

# Generated by cassandra-nodetool-compaction-tuning agent SOP
nodetool repair \
  --partitioner-range \
  --sequential \
  --full \
  --in-hosts 10.0.1.12,10.0.1.13,10.0.1.14 \
  telemetry events_by_device

Before executing compaction strategy updates (e.g., converting an STCS table to LCS), the agent must audit table statistics and calculate whether the node has sufficient disk space (LCS requires ~10% headroom, while STCS requires up to 50%). The agent outputs an ALTER TABLE statement with explicit `sstable_size_in_mb` and `tombstone_compaction_interval` configurations.

ALTER TABLE telemetry.events_by_device
WITH compaction = {
    'class': 'LeveledCompactionStrategy',
    'sstable_size_in_mb': 160,
    'tombstone_threshold': 0.2,
    'tombstone_compaction_interval': 86400,
    'unchecked_tombstone_compaction': 'true'
};