Queue AIops

Governed AI-ops for redis + rabbitmq. queue-aiops is for the team running their own cache and message broker — a redis that "suddenly eats memory", a rabbitmq whose queues quietly grow until publishers block — without an enterprise observability suite. It gives an AI agent (or a human at the CLI) a governed toolset over both: transparent root-cause analyses for memory pressure, latency, queue backlog, and connection churn, plus the handful of writes an operator actually needs (config set, client kill, purge/delete queue, policies) — every call audited, budgeted, risk-tiered, and undo-recorded by the built-in governance harness.

Disclaimer: Community-maintained open-source project, not affiliated with, endorsed by, or sponsored by the Redis or RabbitMQ projects or their respective owners. Redis and RabbitMQ are trademarks of their respective owners.

Verification: behaviour is covered by a mock-based test suite; not yet validated against live brokers. Both redis and rabbitmq are free/self-hostable (one lab container or package install each), so a lab check is easy — queue-aiops doctor is the fastest live probe, and docs/VERIFICATION.md is the checklist.

Quick start

uv tool install queue-aiops

queue-aiops init      # wizard: pick platform (redis/rabbitmq), host/port, encrypted secret
queue-aiops doctor    # config + secret + connectivity check (PING / /api/overview)
queue-aiops overview  # one-shot health summary for the default target

Then the interesting parts:

queue-aiops analyze memory     # redis memory-pressure RCA (maxmemory, eviction, frag, big keys)
queue-aiops analyze latency    # redis latency RCA (slowlog digest, fork/AOF stalls)
queue-aiops analyze backlog    # rabbitmq queue-backlog RCA (consumers, unacked, watermarks)
queue-aiops analyze churn      # connection churn, both platforms
queue-aiops redis bigkeys      # SCAN-budgeted big-key sample (never KEYS *)
queue-aiops rabbitmq queues    # deepest backlog first

Security: read-only mode

This tool is meant to be handed to an AI agent, so its safety story is enforced by the server rather than requested in a prompt:

export QUEUE_READ_ONLY=1

With that set, the 8 write tools are never registered. An MCP client lists 20 tools instead of 28 — the writes are not hidden, not gated behind a flag, and not merely refused when called. They are absent from the session. A model cannot invoke a tool it was never offered, and cannot be argued into one.

That distinction is the whole point. A tool that exists but refuses still invites retry loops and "I'll describe the call instead" behaviour from smaller models, and it leaves a reviewer trusting a promise. An absent tool is a fact you can check: connect, list the tools, and see that the writes are not there.

Enforcement is two layers deep, so the switch cannot be sidestepped by changing entry point:

Read operations are unaffected, and every call is still audited to ~/.queue-aiops/audit.db.

The read/write split is derived from each tool's declared risk_level, and a test asserts that this never disagrees with the [READ]/[WRITE] tag in the tool's own documentation — so a write can't quietly present itself as a read.

Running a smaller / local model? See agent-guardrails.md — it lists the guardrails this tool now enforces for you (so you don't spend prompt budget restating them) and gives a ready-made system prompt for what's left.

Support scope

28 MCP tools — 20 reads (incl. 4 flagship RCAs) + 8 governed writes.

The four RCAs are transparent heuristics that report their numbers — thresholds are named constants, every finding carries its evidence, never a black-box verdict. Big-key sampling walks at most 10,000 keys with SCAN and sizes at most 200 with MEMORY USAGE — never KEYS * — and reports its coverage.

Governance

Every MCP tool runs through the bundled @governed_tool harness (queue_aiops.governance — no external dependency):

• Audit — every call lands in ~/.queue-aiops/audit.db (relocatable via QUEUE_AIOPS_HOME), secret-redacted.
• Budget — call/time ceilings (QUEUE_MAX_TOOL_CALLS, QUEUE_MAX_TOOL_SECONDS) + a runaway-loop breaker.
• Risk tiers & approval — secure by default: with no ~/.queue-aiops/rules.yaml, high-risk writes (purge_queue, delete_queue) are denied unless QUEUE_AUDIT_APPROVED_BY names an approver (set QUEUE_AUDIT_RATIONALE too). queue-aiops init seeds a starter rules.yaml with that dual-control tier; an operator-authored rules file is honoured as-is.
• Undo — reversible writes capture the real before-state first: redis_config_set records the prior value from CONFIG GET; set_policy/delete_policy record the prior policy; delete_queue records the queue's definition and its undo re-declares it (the messages are not restored — the descriptor says so). Irreversible writes (purge_queue, redis_kill_client) record priorState only.
• Dry-run + double-confirm — every write takes dry_run=True (MCP) / --dry-run (CLI); CLI writes double-confirm and execute through the same governed twins, so they land in the audit log too.
• Credentials live encrypted in ~/.queue-aiops/secrets.enc (Fernet + scrypt master password; QUEUE_AIOPS_MASTER_PASSWORD for non-interactive use). Redis passwords are optional — an auth-less lab instance is a supported target.

MCP configuration

{
  "mcpServers": {
    "queue-aiops": {
      "command": "uvx",
      "args": ["--from", "queue-aiops", "queue-aiops-mcp"],
      "env": {
        "QUEUE_AIOPS_MASTER_PASSWORD": "your-master-password"
      }
    }
  }
}

env-block caveat: MCP clients launch the server with a minimal environment — your shell profile is not sourced. Anything the server needs (QUEUE_AIOPS_MASTER_PASSWORD, a relocated QUEUE_AIOPS_HOME, QUEUE_AUDIT_APPROVED_BY for high-risk writes) must be set in the env block above, not just in your terminal.

Or, with the package installed: queue-aiops mcp.

CLI reference (short)

queue-aiops init | doctor | overview | mcp
queue-aiops secret set|list|migrate ...
queue-aiops redis info|memory|clients|slowlog|config-get|keyspace|bigkeys
queue-aiops redis config-set <param> <value> [--dry-run]
queue-aiops redis kill-client --id <id> | --addr <ip:port> [--dry-run]
queue-aiops rabbitmq overview|queues|queue|connections|channels|policies|nodes
queue-aiops rabbitmq purge|delete-queue|declare-queue <name> [--vhost /] [--dry-run]
queue-aiops rabbitmq set-policy|delete-policy <name> ... [--dry-run]
queue-aiops analyze memory|latency|backlog|churn

Verification status

Live-verified against Redis 7.4.9 and RabbitMQ 3.13.7 (2026-07-19/20). Connectivity, every Redis read (cross-checked against redis-cli ground truth), all four analyses, and the full governance loop (real redis_config_set → audit row → undo restoring the prior value) were exercised against a real server. That run found and fixed a real defect: integer quantities — key counts, client counts, byte totals — were rendered as floats (202.0 keys), which equality assertions cannot catch.

The RabbitMQ group is now verified end-to-end too — reads cross-checked against rabbitmqadmin, and set_policy → undo_apply closing on the live broker. Redis cluster/sentinel topologies and AUTH/TLS connections remain unverified.

docs/VERIFICATION.md records exactly what was checked and what is still open. queue-aiops doctor is the fastest live check.

Contributing

缺功能提 issue/PR 欢迎留言 — missing a read you need (streams/consumer groups, quorum-queue specifics, shovel/federation status), another broker platform, or a threshold that doesn't fit your fleet? Open an issue or PR at github.com/AIops-tools/Queue-AIops — platform registry entries are additive and small.

License

MIT