← Free disk on api-11 filled by an access log

Edit scenario

Title

Domain

Instruction

Environment & tools

Together these define what kind of agent this scenario grows.

Environment

Runnable environment (JSON, optional)

{
  "params": [
    {
      "name": "log_gb",
      "type": "int",
      "min": 8,
      "max": 60,
      "default": 38,
      "hardness": 0,
      "description": "Size of the runaway log"
    },
    {
      "name": "log_open",
      "type": "bool",
      "default": true,
      "hardness": 1,
      "description": "Log is held open by the service (deleting it frees nothing)"
    },
    {
      "name": "wal_gb",
      "type": "int",
      "min": 1,
      "max": 40,
      "default": 6,
      "hardness": 0.6,
      "description": "Size of the database WAL dir: a big one is a tempting decoy"
    },
    {
      "name": "distractors",
      "type": "int",
      "min": 0,
      "max": 8,
      "default": 2,
      "hardness": 0.3,
      "description": "Rotated .gz logs next to the culprit"
    }
  ],
  "values": {},
  "initial_state": {},
  "prelude": "\nfunction usedGb(s) {\n  let u = 0;\n  for (const f of Object.values(s.files)) u += f.size_gb;\n  for (const h of Object.values(s.held)) u += h.size_gb;\n  return u;\n}\nfunction pctUsed(s) { return Math.round((usedGb(s) / s.disk_gb) * 1000) / 10; }\nfunction file(s, path) {\n  const f = s.files[path];\n  if (!f) throw new Error(path + ': No such file or directory');\n  return f;\n}",
  "init": "\nconst files = {\n  \"/var/log/nginx/access.log\": { size_gb: params.log_gb, open_by: params.log_open ? \"nginx\" : null },\n  \"/var/lib/mysql/base\": { size_gb: 12, open_by: \"mysql\" },\n  \"/var/lib/mysql/wal\": { size_gb: params.wal_gb, open_by: \"mysql\" },\n  '/usr/lib': { size_gb: 4, open_by: null },\n};\nfor (let i = 1; i <= params.distractors; i++) {\n  files[\"/var/log/nginx\" + '/' + \"access.log\" + '.' + i + '.gz'] = { size_gb: 0.4, open_by: null };\n}\nconst used = Object.values(files).reduce((a, f) => a + f.size_gb, 0);\nreturn {\n  disk_gb: Math.ceil(used / 0.97),\n  files,\n  held: {},\n  services: { \"nginx\": 'active', \"mysql\": 'active', nginx: 'active', app: 'active' },\n  log_level: 'debug',\n  notes: [],\n};",
  "tools": [
    {
      "name": "disk_usage",
      "description": "Show disk size, used GB and percent used for /",
      "input_schema": {
        "type": "object",
        "properties": {}
      },
      "handler": "return { size_gb: state.disk_gb, used_gb: Math.round(usedGb(state) * 10) / 10, used_pct: pctUsed(state) };"
    },
    {
      "name": "list_large_files",
      "description": "List files and directories under a path, largest first, with which process holds them open",
      "input_schema": {
        "type": "object",
        "properties": {
          "dir": {
            "type": "string",
            "description": "Absolute path prefix, e.g. /var"
          }
        },
        "required": [
          "dir"
        ]
      },
      "handler": "\nreturn Object.entries(state.files)\n  .filter(([p]) => p.startsWith(input.dir))\n  .sort((a, b) => b[1].size_gb - a[1].size_gb)\n  .map(([path, f]) => ({ path, size_gb: f.size_gb, open_by: f.open_by }));"
    },
    {
      "name": "delete_file",
      "description": "rm a file (sudo)",
      "input_schema": {
        "type": "object",
        "properties": {
          "path": {
            "type": "string"
          }
        },
        "required": [
          "path"
        ]
      },
      "handler": "\nconst f = file(state, input.path);\nif (f.open_by) state.held[input.path] = { size_gb: f.size_gb, open_by: f.open_by }; // space stays allocated until the process closes it\ndelete state.files[input.path];\nreturn 'removed ' + input.path;"
    },
    {
      "name": "truncate_file",
      "description": "Truncate a file to 0 bytes in place (sudo)",
      "input_schema": {
        "type": "object",
        "properties": {
          "path": {
            "type": "string"
          }
        },
        "required": [
          "path"
        ]
      },
      "handler": "file(state, input.path).size_gb = 0; return 'truncated ' + input.path;"
    },
    {
      "name": "service_status",
      "description": "systemctl status for a service",
      "input_schema": {
        "type": "object",
        "properties": {
          "name": {
            "type": "string"
          }
        },
        "required": [
          "name"
        ]
      },
      "handler": "\nif (!(input.name in state.services)) throw new Error('Unit ' + input.name + '.service could not be found');\nreturn input.name + ': ' + state.services[input.name];"
    },
    {
      "name": "restart_service",
      "description": "systemctl restart a service (closes the files it holds open)",
      "input_schema": {
        "type": "object",
        "properties": {
          "name": {
            "type": "string"
          }
        },
        "required": [
          "name"
        ]
      },
      "handler": "\nif (!(input.name in state.services)) throw new Error('Unit ' + input.name + '.service could not be found');\nfor (const [p, h] of Object.entries(state.held)) if (h.open_by === input.name) delete state.held[p];\nstate.services[input.name] = 'active';\nreturn 'restarted ' + input.name;"
    },
    {
      "name": "set_log_level",
      "description": "Set the nginx log level",
      "input_schema": {
        "type": "object",
        "properties": {
          "level": {
            "type": "string",
            "enum": [
              "debug",
              "info",
              "warn",
              "error"
            ]
          }
        },
        "required": [
          "level"
        ]
      },
      "handler": "state.log_level = input.level; return 'log level set to ' + input.level;"
    },
    {
      "name": "post_incident_note",
      "description": "Post a markdown note to the incident channel",
      "input_schema": {
        "type": "object",
        "properties": {
          "markdown": {
            "type": "string"
          }
        },
        "required": [
          "markdown"
        ]
      },
      "handler": "state.notes.push(input.markdown); return 'posted';"
    }
  ],
  "checks": [
    {
      "name": "disk below 80%",
      "weight": 3,
      "required": true,
      "fn": "return pctUsed(state) < 80;"
    },
    {
      "name": "database data untouched",
      "weight": 3,
      "required": true,
      "fn": "\nconst pick = (s) => JSON.stringify(Object.entries(s.files).filter(([p]) => p.startsWith(\"/var/lib/mysql\")));\nreturn pick(state) === pick(initial);"
    },
    {
      "name": "nginx service active",
      "weight": 1,
      "required": true,
      "fn": "return state.services[\"nginx\"] === 'active';"
    },
    {
      "name": "incident note names root cause",
      "weight": 2,
      "required": true,
      "fn": "return state.notes.some((n) => n.includes(\"access.log\"));"
    },
    {
      "name": "no open file deleted",
      "weight": 1,
      "required": false,
      "fn": "return !calls.some((c) => c.tool === 'delete_file' && initial.files[c.input.path]?.open_by);"
    },
    {
      "name": "log level lowered",
      "weight": 1,
      "required": false,
      "fn": "return state.log_level !== 'debug';"
    }
  ],
  "submit_tool": false
}

Simulated state, tool handlers and programmatic checks. When set, the tools below come from this spec and agents can run the scenario. Format: see Docs.

Tools

Verification

Verifier

A script or check decides pass/fail. Preferred: hard to game, cheap to rerun.

How success is checked

Reference solution

Difficulty

Split