Klarsmith Ops Kit

WordPress has no honest health endpoint. The front page, admin-ajax.php and the usual health.php stubs all return 200 while the database is down, the object cache is gone, or WordPress is serving its "database update required" interstitial. Every probe in common use calls that pod ready. Klarsmith Ops Kit gives a containerised WordPress site three things operators actually need:

• Honest readiness — GET /wp-json/ops/v1/readyz returns 200 only when the database answers, the schema matches the running core version, the object cache round-trips, and the uploads directory is writable. Anything else is a 503 with the failing check names. Liveness is deliberately left alone: a database outage must drain pods, never restart them.
• Prometheus metrics — GET /wp-json/ops/v1/metrics serves the text exposition format from a snapshot that wp ops collect writes to the object cache on a schedule. Nothing expensive runs on scrape. wp_ops_snapshot_age_seconds tells you when the collector has stopped. Pod-level series (wp_ops_pod_*) and site-level series (wp_ops_site_*) are split so replicas do not duplicate site numbers.
• Structured JSON logs — with WP_OPS_LOG_JSON=true, PHP errors and fatals go to stderr as one JSON object per line, ready for a log pipeline.

The plugin fails closed. /metrics is disabled until a token is configured, and the anonymous readiness response names failing checks without leaking detail. Configuration is by environment variable, which is how containers are configured:

• WP_OPS_TOKEN — bearer token for /metrics (Authorization: Bearer or X-Ops-Token). Unset disables the endpoint.
• WP_OPS_SITE_NAME — site label on JSON log records (metrics carry no site label; the scraper's own labels identify the site).
• WP_OPS_EXPECT_OBJECT_CACHE — fail readiness when no external object cache is active.
• WP_OPS_REQUIRED_PLUGINS — comma-separated plugin files (dir/plugin.php) that must be active for readiness.
• WP_OPS_REST_BYPASS_AUTH — set to false to stop the plugin allowing anonymous access to its own REST namespace.
• WP_OPS_LOG_JSON — set to true to emit JSON log lines on stderr.

WP-CLI commands: wp ops check (exit 1 on any failing check), wp ops collect (refresh the snapshot), wp ops metrics (print the exposition locally). Source, issue tracker and Kubernetes manifests for probes, the collector CronJob and scraping live at https://github.com/klarsmith/wp-ops-kit.

1. Install from the WordPress plugin directory, or with Composer (composer require klarsmith/wp-ops-kit; the Composer package installs as wp-content/plugins/wp-ops-kit).
2. Activate the plugin.
3. Point your readiness probe at /wp-json/ops/v1/readyz.
4. Run wp ops collect on a schedule (a five-minute CronJob is the reference setup).
5. Set WP_OPS_TOKEN and scrape /wp-json/ops/v1/metrics with that bearer token.

If a security plugin blocks anonymous REST requests, the plugin allows its own ops/v1 namespace through at priority 1 unless WP_OPS_REST_BYPASS_AUTH=false. Add ops to the security plugin's allowlist as well where it has one.

0.1.7

• Directory listing name is "Klarsmith Ops Kit" (slug klarsmith-ops-kit); Composer/GitHub name unchanged.

0.1.6

• Earlier directory listing name "Ops Kit" (superseded by 0.1.7). Tested up to WordPress 7.1.

0.1.3

• Code conforms to the WordPress Coding Standards; PHPStan level 6 clean. No behaviour change beyond sanitising two request inputs used for route detection.
• Copy-paste manifests in examples/ for Kubernetes, stock Prometheus, VictoriaMetrics and a Grafana dashboard.

0.1.2

• /metrics no longer appends its exposition to a response another handler has already served.
• Zero-valued post statuses are dropped at collection time; publish is always exported so a drop to zero stays alertable.

0.1.1

• Allow the plugin's own ops/v1 REST namespace through site-level anonymous-REST lockdowns (hooked at priority 1, never overriding an earlier decision). Escape hatch: WP_OPS_REST_BYPASS_AUTH=false.

0.1.0

• Initial release: readiness endpoint, snapshot-backed Prometheus metrics, JSON logging, wp ops check|collect|metrics commands.