- Status: Notes
- JIRA: KEYCLOAK-8288
- Need to know if Keycloak is running
- Need to find and debug issues in production
Health - a health check endpoint returns a simple binary up or down state of the system. This is very useful in managed environments like Kubernetes and OpenShift where the controller can automatically stop broken instances.
Metrics - metrics provide a more detailed view of the health and state of the system. From metrics it is possible to find out how the system is behaving, not just if the system is up or down. Examples here could be observing a large amount of error responses or response times increasing beyond a given threshold.
Logging - metrics provides the details on how the system is behaving, but will not always be able to provide sufficient information to debug the system. This is where logging comes in. Through well defined logging it is possible to further obtain information on how the system is behaving and resolving potential issues.
Tracing - in a distributed system it can be difficult to find the journey a single transactions takes through the systems. With distributed tracing it is possible to do exactly this.
WildFly 16 comes with built-in support for Observerability, which will make it relatively straightforward to add the same to Keycloak.
Health endpoint is enabled by default in WildFly 16 at:
WildFly implements the MicroProfile Health specification.
Metrics endpoint is enabled by default in WildFly 16 at:
WildFly implements the MicroProfile Metrics specification. Metrics can be exposed in Prometheus or JSON format.
By default only metrics for the base (JVM) is available. Further metrics can be enabled with -Dwildfly.statistics-enabled=true
,
which will enable statistics for all subsystems that support it. It is also possible to enable statistics for specific
subsystems. For example -Dwildfly.undertow.statistics-enabled=true
will enable statistics for Undertow.
It is also possible to change the prefix of metrics from wildfly
to keycloak
by simply changing the prefix
attribute on the microprofile-metrics-smallrye subsystem (or wildfly.metrics.prefix
system property).
To add application (Keycloak) specific metrics WildFly supports the MicroProfile Metrics specification. This is not useful to Keycloak as it requires CDI, which Keycloak by design has chosen not to use. This leaves us with the following options:
-
DMR - WildFly automatically exposes metrics from subsystems that expose attributes with the type
metrics
. One big challenge here is the amount of boilerplate required to work with DMR. -
SmallRye Registry - SmallRye exposes static registries that allows registering additional metrics. One challenge here is that there is an issue in SmallRye 1 that makes it not possible to use tags/labels. This is resolved in SmallRye 2, but WildFly is using SmallRye 1.
-
WildFly Registry - WildFly has its own registry used to expose metrics from subsystems. This could be an option, but would require changes to the API as it is using DMR specific APIs today.
JSON formatted logging can be enabled in WildFLy with the following CLI commands:
/subsystem=logging/json-formatter=json:add(pretty-print=true, exception-output-type=formatted)
/subsystem=logging/console-handler=CONSOLE:write-attribute(name=named-formatter, value=json)
:reload
More info in EAP Documentation.
Unclear if we need to do some additional health checks in addition to what WildFly does.
In addition to metrics exposed by WildFly we will expose some Keycloak specific metrics. For example:
- User login
- User login failures
- User count
- Session count
- Client count
Exactly what additional metrics we should expose in the first round is still to be decided.
We will start small with the aim to expand on the available metrics in the future.
WildFly covers JSON formatted logging. We will simply add some documentation.
Most transactions to Keycloak are directly from the browser and do not invoke other services. Applications invoke Keycloak to exchange code for a token, to refresh tokens and finally services may invoke Keycloak to verify tokens. These are simpler use-cases as we haven't seen much demand for tracing in Keycloak. Hence distributed tracing is a low priority and will be considered in the future.
- http://www.mastertheboss.com/jboss-server/jboss-monitoring/how-to-manage-wildfly-metrics
- http://www.mastertheboss.com/jboss-server/jboss-monitoring/monitoring-wildfly-with-prometheus
- http://www.mastertheboss.com/jboss-server/jboss-monitoring/wildfly-application-monitoring
- https://microprofile.io/project/eclipse/microprofile-health
- https://microprofile.io/project/eclipse/microprofile-metrics
Non-public resources:
- https://docs.google.com/presentation/d/1YU6kGjereV8-tJKXvYtmNIwiLu1TyF24MtXHMx-HFBI
- https://docs.google.com/document/d/1v0cs_RdcX4ZKuOUpM9frbFjs9KG9DMLFfOnvt124YIY
- https://docs.google.com/spreadsheets/d/1uZnsXTu3W0oXxfSxiB78WBcuKf27hEs_txDsxdkbPjY
For example for JMS Queue, you create metric definition for DMR JMSQueueDefinition and you associate a read handler that will read the value for the Artemis objects and return it to DMR JMSQueueReadAttributeHandler
MetricRegistries to access the static Smallrye registries - they have 3 different ones:
- base (for JVM metrics)
- vendor (for vendor / app server metrics)
- application
The base class is MetricCollector, but the API would have to be adapted a bit for our use case.