Skip to main content

Health checks

Health checks examine how a cluster is configured. They read the cluster and warehouse definitions and report settings likely to cause trouble later.

They are not metrics: they report what is configured, not what is happening right now.

Most checks read only those definitions. Three do not: two read license state from the cluster, and one reports on Anywhere's connection to Prometheus. These three can fail or be skipped when the cluster is unreachable, even if every other check passed.

Reading a finding

Each finding names the check that produced it, the component it applies to, a severity, a status, and a suggested fix.

Severity:

SeverityMeans
criticalFix this
warningFix this
infoA judgment call — not following the advice is an acceptable trade-off

Info findings are not defects. Checks such as topology spread and serial pod management are recommendations. A cluster that ignores them can still be correctly configured.

Status. skipped is not passed:

StatusMeans
passedEvaluated, and the configuration is fine
failedEvaluated, and found a problem
skippedCould not be evaluated — the finding says why

Findings explain themselves. Each one names the check that produced it and what it looked for, and a finding that failed or was skipped also carries why, and what to do about it. There is no separate catalogue to consult.

A warehouse is checked against a subset of the list. A warehouse runs a single component, so the checks that compare one component against another do not apply, and a warehouse reports fewer findings than a cluster.

Where to see them

In the console:

  • Health checks — every cluster in one view, with a cluster selector
  • On a single cluster, and on a single warehouse — the same findings, scoped

Or through the REST API, with an administrator session:

RequestReturns
GET /api/v1/admin/namespaces/{namespace}/clusters/{cluster}/inspectionsOne cluster
GET /api/v1/admin/namespaces/{namespace}/clusters/{cluster}/warehouses/{warehouse}/inspectionsOne warehouse
GET /api/v1/admin/inspectionsEvery cluster in view; add ?namespace= to narrow it

Turning checks off

Which checks run is set in Anywhere's configuration, under inspection: enabledRules turns the catalogue into an allow-list, and disabledRules removes checks from whatever is left. Both are read once at startup, so a change needs a restart.

The setting applies to the whole deployment, not to one cluster, and covers warehouse findings too. Disabling a check silences it everywhere.

An id in either list that no longer exists in the catalogue is logged and skipped, so a configuration written for an older build still starts.