Usage


galera-doctor audit [flags]
galera-doctor checks
galera-doctor version

Where the nodes come from

Either a config file: 


{
  "clusters": {
    "compress": {
      "nodes": [
        {"name": "sg-01", "dsn": "audit:${GALERA_DOCTOR_PASS}@tcp(10.11.1.5:3306)/"},
        {"name": "cl-02", "dsn": "audit:${GALERA_DOCTOR_PASS}@tcp(10.21.1.5:3306)/"},
        {"name": "ov-03", "dsn": "audit:${GALERA_DOCTOR_PASS}@tcp(10.35.1.5:3306)/"}
      ],
      "proxysql_dsn": "admin:${PROXYSQL_ADMIN_PASS}@tcp(10.11.1.9:6032)/",
      "expect_nodes": 3
    }
  }
}

...or the command line, with --node repeated: 


galera-doctor audit \
  --node "sg-01=audit:$PASS@tcp(10.11.1.5:3306)/" \
  --node "cl-02=audit:$PASS@tcp(10.21.1.5:3306)/"

${ENV_VAR} is expanded in every DSN. An unset variable is an error, not an empty string: a DSN with an empty password fails with "access denied", which is a bad way to learn that a variable is missing. Unknown keys in the file are rejected too — a typo in "noeds" would otherwise produce a cluster with no nodes.

A DSN is never printed. Findings identify nodes by name, and a driver error that quotes a DSN is redacted first.

Flags

Flag Default What it does
--config FILE — cluster definitions (JSON)
--cluster NAME all audit one cluster from the config
--node name=DSN — a node, repeatable
--proxysql DSN — ProxySQL admin interface to compare against
--state FILE — remember counters between runs so rates can be graded
--expect-nodes N nodes given membership the cluster should report
--timeout D 10s per-node timeout — one dead node does not decide the run's length
--no-systables off skip the system table drift comparison
--no-schema off skip the primary key check
--flow-warn F / --flow-bad F 0.01 / 0.10 flow-control share of the interval
--ist-warn D 30m gcache window below which a restart means a full SST
--json / --findings — full report / flat findings array
--min-severity S — hide findings below S
--exit-on S never exit 1 when a finding reaches S
--watch D off re-audit every D and print only the transitions

Exit status

Code Meaning
0 the audit ran — findings are output, not an error
1 --exit-on threshold reached
2 usage error, or no node could be resolved

While you are repairing it


$ galera-doctor audit --config clusters.json --watch 10s

The first report in full, so you know where you are starting from, and after that only what moved: 


16:34:48  OK    cluster/uuid   compress/compress  all nodes report one cluster  [BAD → OK]
16:35:18  WARN  node/state     compress/ov-03     local state is Donor/Desynced  [OK → WARN]
16:35:48  ERROR                compress/node/read@ov-03 no longer reported  [ERROR → gone]

A tick that changed nothing prints nothing — reprinting twenty OK lines every ten seconds buries the one that matters. A finding that disappeared is reported too, because during a repair that is usually the line you are waiting for; an OK that stopped being reported is not, since the check simply did not run this time.

--watch refuses an interval under a second (a busy loop against a cluster that is already having a bad day) and refuses --json and --findings: those emit one document per run, and a stream of documents is not a document.

With --state, each tick writes the file and reads back what it wrote, so the rate checks grade the watch interval rather than the whole session.

It is not a daemon and not a monitoring system: it runs in the foreground, holds its baseline in memory, and stops when you do.

In a scheduled job


galera-doctor audit --config /etc/galera-doctor/clusters.json \
  --state /var/lib/galera-doctor/state.json \
  --findings --min-severity WARN --exit-on BAD > findings.json

The state file is written after the output, so a failure to persist a baseline costs the next run its rates and nothing else.