firefly/actuator is LaraFly's production-ready management surface — the Spring-Boot-Actuator analogue. It ships a
HealthIndicator SPI with liveness/readiness probe groups, an ActuatorEndpoint contract + registry, and a
route-registration BootPass that mounts framework endpoints on the illuminate Router under /actuator. Dependency-light,
always-on, and secured entirely by M11 config with zero code edge to firefly/security.
/actuator— HAL index of exposed endpoints/actuator/health(+/actuator/health/{group}, liveness/readiness) — aggregated health, 503 on DOWN/actuator/info— mergedInfoContributorfragments (runtime,app,build)/actuator/env— thefirefly.*config tree, sensitive values masked/actuator/configprops— every#[ConfigProperties]DTO, with the values it actually resolved off the bound instance, masked/actuator/caches(+/actuator/caches/{name}) — the configuredcache.stores(name/driver/default only); read-only, no eviction/actuator/beans,/actuator/conditions,/actuator/mappings,/actuator/loggers(GET/POST),/actuator/scheduledtasks/actuator/metrics,/actuator/prometheus,/actuator/httpexchanges,/actuator/process— supplied byfirefly/observabilitywhen installed/actuator/oauth2clients— supplied byfirefly/security-oauth2-serverwhen installed and enabled (bothfirefly.security.enabledandfirefly.security.oauth2.server.enabled): every registered client with its grants, scopes, redirect URIs, settings and live authorization count, never a secret. The counts are per-process on the defaultmemoryauthorizations driver — a map rebuilt in every PHP worker — which is why the payload states its own storage model inauthorizations.processLocal;authorizations.driver = eloquentis what makes them describe the deployment
!!! tip "A browser view over all of this"
firefly/admin renders these same endpoints as a server-side dashboard, reading them in-process rather
than over HTTP — so it shows pages the exposure model below deliberately keeps unpublished. That inversion is
the whole of its security model: see Admin Dashboard, and read
its access model before enabling it outside app.debug.
HealthIndicator { health(): Health }; Status = UP/DOWN/OUT_OF_SERVICE/UNKNOWN (severity DOWN>OUT_OF_SERVICE>UP>UNKNOWN);
a most-severe StatusAggregator (empty→UP); a bean-scan HealthContributorRegistrar; built-in Ping/DiskSpace/Db
indicators. A throwing indicator degrades to DOWN — never a 500.
Default firefly.management.endpoints.web.exposure.include = "health,info"; sensitive endpoints return 404 until
explicitly exposed. An endpoint body is always a JSON object: /actuator/info with no InfoContributor
registered answers {}, not [], so a typed client deserialising into a map does not break on the default. Lock them
down with firefly.security.http.rules (no second management port — doesn't fit PHP-FPM). These are the exact rules the
framework's own end-to-end lockdown test seeds, written as the settings the test overrides; in an application the same
three entries are the nested security.http.rules array of config/firefly.php, which the shipped reference already
carries commented out:
return [
...parent::configOverrides(),
'firefly.management.endpoints.web.exposure.include' => 'health,info,env',
'firefly.security.enabled' => true,
'firefly.security.http.enabled' => true,
'firefly.security.http.rules' => [
['pattern' => 'actuator/health', 'access' => 'permitAll'],
['pattern' => 'actuator/info', 'access' => 'permitAll'],
['pattern' => 'actuator/*', 'access' => 'hasRole:ACTUATOR'],
],
];firefly/security's HttpSecurityFilter is a global middleware — FilterChainRegistrar pushes it onto Laravel's
HTTP-kernel middleware stack ($kernel->pushMiddleware(...)), so it runs for every request the kernel handles,
including the actuator's directly-$router->match()-registered /actuator/** routes. Securing actuator is therefore
pure configuration: firefly/actuator's composer.json has no dependency on firefly/security, and deptrac.yaml
carries no Actuator → Security edge. An integration test
(packages/actuator/tests/SecurityLockdownIntegrationTest.php, booted via
packages/actuator/tests/Support/SecuredActuatorCapstoneTestCase.php) boots both stacks together and proves it
end-to-end: with the lockdown rules above and env exposed, an anonymous GET /actuator/env is denied 401
(AuthenticationException, no matching rule's expression is satisfied) while GET /actuator/health stays 200
(permitAll).
/actuator/env and /actuator/configprops additionally mask any key matching
password|secret|token|key|credential|passwd|authorization|headers (case-insensitive, substring, recursive — the shared
SensitiveValueMasker) with ******. The key is tested before the value's type, so a sensitive key holding an array
(a JWT keyring, a credentials pair) is replaced wholesale rather than recursed into, independent of whether the URL
lockdown above is configured — defense in depth for an endpoint that is reachable at all only once explicitly exposed.
headers is in the list because a bag of outbound headers is where a client's credential travels
(firefly.observability.tracing.otlp.headers documents authorization=Bearer … as its contents, and a vendor's
x-honeycomb-team leaf matches nothing on its own, so only the bag's key can decide); the accepted cost is that
firefly.security.headers — the response-header filter's enabled/hsts/csp block, nothing an operator cannot read
off any response — renders as ****** too. The singular header is deliberately not matched: the same rule names the
data browser's sensitive columns, and page_header is page furniture, not a credential.
firefly.management.enabled(defaulttrue) — master gatefirefly.management.endpoints.web.exposure.include/.exclude(CSV or*;*is a wildcard in both lists and exclude wins, so.exclude = "*"is the kill switch)firefly.management.endpoints.web.base-path(default/actuator)firefly.management.endpoint.{id}.enabled(per-endpoint)firefly.management.endpoint.health.show-details(defaultnever; only the literalalwaysshows component details — see Known-latent forwhen-authorized)firefly.management.endpoint.health.group.{name}.includefirefly.management.endpoint.health.db.enabled(defaulttrue) — thedbhealth indicator, registered wheneverdatabase.defaultnames a connection with a driver (Spring Boot'sDataSourceHealthIndicatorauto-configuration); a failing or missing database reports DOWN and/actuator/healthanswers 503; an application with no default database gets nodbcomponent at all.falseremoves the indicator. An indicator can decline registration itself by implementingConditionalHealthIndicator::available().firefly.management.info.app.*,firefly.management.info.build.pathfirefly.management.info.runtime.enabled(defaulttrue,#[ConditionalOnProperty(matchIfMissing: true)]) — theruntimefragment of/actuator/info(PHP version/SAPI/OPcache, Laravel version, LaraFly version, current+peak memory). Setting itfalseremoves the contributor bean entirely.
Framework endpoints are #[Component] beans mounted by a BootPass on the illuminate Router (reusing route:list,
URL generation, the HTTP-kernel middleware pipeline) — not app controllers. Health/info reuse Laravel DB/Log/config.
when-authorizedshow-details degrades tonever(no Security code edge; gate details via the lockdown)./refresh,/threaddump,/shutdownare deferred to later SP cycles./cachesis read-only: Spring'sDELETEeviction is deliberately not implemented, becausefirefly/actuatorcarries no code edge tofirefly/securityand so cannot say who asked; aPOSTto it answers 404.- No second management port (an Octane second-listener is an SP-7 option).