Merge "Add controllers and remote servers healthchecks"

5 files changed, 629 insertions, 136 deletions

diff --git a/feature-healthcheck/src/main/feature/config/feature-healthcheck.properties b/feature-healthcheck/src/main/feature/config/feature-healthcheck.properties
index 7739c6e7..de4b8ace 100644
--- a/feature-healthcheck/src/main/feature/config/feature-healthcheck.properties
+++ b/feature-healthcheck/src/main/feature/config/feature-healthcheck.properties
@@ -1,8 +1,8 @@
-###
+#
 # ============LICENSE_START=======================================================
-# feature-healthcheck
+# ONAP
 # ================================================================================
-# Copyright (C) 2017-2019 AT&T Intellectual Property. All rights reserved.
+# Copyright (C) 2017-2019,2022 AT&T Intellectual Property. All rights reserved.
 # ================================================================================
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -16,9 +16,10 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 # ============LICENSE_END=========================================================
-###
+#
+
+http.server.services=HEALTHCHECK,LIVENESS
 
-http.server.services=HEALTHCHECK
 http.server.services.HEALTHCHECK.host=0.0.0.0
 http.server.services.HEALTHCHECK.port=6969
 http.server.services.HEALTHCHECK.restClasses=org.onap.policy.drools.healthcheck.RestHealthCheck
@@ -29,3 +30,31 @@ http.server.services.HEALTHCHECK.password=${envd:HEALTHCHECK_PASSWORD}
 http.server.services.HEALTHCHECK.https=${envd:HTTP_SERVER_HTTPS:false}
 http.server.services.HEALTHCHECK.aaf=${envd:AAF:false}
 http.server.services.HEALTHCHECK.serialization.provider=org.onap.policy.common.gson.JacksonHandler,org.onap.policy.common.endpoints.http.server.YamlJacksonHandler
+
+http.server.services.LIVENESS.host=localhost
+http.server.services.LIVENESS.port=6968
+http.server.services.LIVENESS.restClasses=org.onap.policy.drools.healthcheck.RestHealthCheck
+http.server.services.LIVENESS.managed=false
+http.server.services.LIVENESS.swagger=true
+http.server.services.LIVENESS.serialization.provider=org.onap.policy.common.gson.JacksonHandler,org.onap.policy.common.endpoints.http.server.YamlJacksonHandler
+
+http.client.services=PAP,PDPX
+
+http.client.services.PAP.host=${envd:PAP_HOST}
+http.client.services.PAP.port=6969
+http.client.services.PAP.userName=${envd:PAP_USERNAME}
+http.client.services.PAP.password=${envd:PAP_PASSWORD}
+http.client.services.PAP.https=${envd:HTTP_SERVER_HTTPS:false}
+http.client.services.PAP.managed=true
+http.client.services.PAP.contextUriPath=policy/pap/v1/healthcheck
+
+http.client.services.PDPX.host=${envd:PDP_HOST}
+http.client.services.PDPX.port=6969
+http.client.services.PDPX.userName=${envd:PDP_USERNAME}
+http.client.services.PDPX.password=${envd:PDP_PASSWORD}
+http.client.services.PDPX.https=${envd:HTTP_SERVER_HTTPS}
+http.client.services.PDPX.managed=true
+http.client.services.PDPX.contextUriPath=policy/pdpx/v1/healthcheck"
+
+liveness.controllers=${envd:LIVENESS_CONTROLLERS}
+liveness.controllers.timeout=${envd:LIVENESS_TIMEOUT_SECONDS:10}
\ No newline at end of file
diff --git a/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/HealthCheck.java b/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/HealthCheck.java
index de00df88..06331bcc 100644
--- a/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/HealthCheck.java
+++ b/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/HealthCheck.java
@@ -1,8 +1,8 @@
 /*
  * ============LICENSE_START=======================================================
- * feature-healthcheck
+ * ONAP
  * ================================================================================
- * Copyright (C) 2017-2019, 2021 AT&T Intellectual Property. All rights reserved.
+ * Copyright (C) 2017-2019, 2021-2022 AT&T Intellectual Property. All rights reserved.
  * ================================================================================
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -22,10 +22,11 @@ package org.onap.policy.drools.healthcheck;
 
 import java.util.ArrayList;
 import java.util.List;
-import lombok.Getter;
-import lombok.Setter;
-import lombok.ToString;
+import lombok.Data;
+import lombok.NoArgsConstructor;
 import org.onap.policy.common.capabilities.Startable;
+import org.onap.policy.common.endpoints.http.client.HttpClient;
+import org.onap.policy.drools.system.PolicyController;
 
 /**
  * Healthcheck.
@@ -35,10 +36,9 @@ public interface HealthCheck extends Startable {
     /**
      * Healthcheck Report.
      */
-    @Getter
-    @Setter
-    @ToString
-    public static class Report {
+    @Data
+    @NoArgsConstructor
+    class Report {
         /**
          * Named Entity in the report.
          */
@@ -57,29 +57,98 @@ public interface HealthCheck extends Startable {
         /**
          * return code.
          */
-        private int code;
+        private long code;
+
+        /**
+         * start time.
+         */
+        private long startTime = System.currentTimeMillis();
+
+        /**
+         * end time.
+         */
+        private long endTime;
+
+        /**
+         * elapsed time.
+         */
+        private long elapsedTime;
 
         /**
          * Message from remote entity.
          */
         private String message;
+
+
+        public Report(Report report) {
+            this.startTime = report.startTime;
+            this.code = report.code;
+            this.elapsedTime = report.elapsedTime;
+            this.endTime = report.endTime;
+            this.healthy = report.healthy;
+            this.message = report.message;
+            this.name = report.name;
+            this.url = report.url;
+        }
+
+        public Report setEndTime() {
+            setEndTime(System.currentTimeMillis());
+            setElapsedTime(endTime - startTime);
+            return this;
+        }
     }
 
     /**
      * Report aggregation.
      */
-    @Getter
-    @Setter
-    @ToString
-    public static class Reports {
+    @Data
+    class Reports {
         private boolean healthy;
+        private final long startTime = System.currentTimeMillis();
+        private long endTime;
+        private long elapsedTime;
         private List<Report> details = new ArrayList<>();
+
+        public Reports setEndTime() {
+            this.endTime = System.currentTimeMillis();
+            this.elapsedTime = this.endTime - this.startTime;
+            return this;
+        }
     }
 
     /**
-     * Perform a healthcheck.
-     *
-     * @return a report
+     * Process engine open status.
+     */
+    void open();
+
+    /**
+     *  System healthcheck.
      */
     Reports healthCheck();
+
+    /**
+     * Engine only healthcheck.
+     */
+    Reports engineHealthcheck();
+
+    /**
+     * Controllers only healthcheck.
+     */
+    Reports controllerHealthcheck();
+
+    /**
+     * Healthcheck on a controller.
+     */
+    Reports controllerHealthcheck(PolicyController controller);
+
+    /**
+     * HTTP Clients only healthcheck.
+     */
+    Reports clientHealthcheck();
+
+    /**
+     * Healthcheck on an HTTP Client.
+     */
+    Reports clientHealthcheck(HttpClient client);
+
 }
diff --git a/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/HealthCheckFeature.java b/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/HealthCheckFeature.java
index 961f4a1d..5da134da 100644
--- a/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/HealthCheckFeature.java
+++ b/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/HealthCheckFeature.java
@@ -1,8 +1,8 @@
 /*-
  * ============LICENSE_START=======================================================
- * feature-healthcheck
+ * ONAP
  * ================================================================================
- * Copyright (C) 2017-2019 AT&T Intellectual Property. All rights reserved.
+ * Copyright (C) 2017-2019, 2022 AT&T Intellectual Property. All rights reserved.
  * ================================================================================
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -33,7 +33,7 @@ public class HealthCheckFeature implements PolicyEngineFeatureApi {
     /**
      * Logger.
      */
-    private static Logger logger = LoggerFactory.getLogger(HealthCheckFeature.class);
+    private static final Logger logger = LoggerFactory.getLogger(HealthCheckFeature.class);
 
     /**
      * Properties Configuration Name.
@@ -57,6 +57,17 @@ public class HealthCheckFeature implements PolicyEngineFeatureApi {
     }
 
     @Override
+    public boolean afterOpen(PolicyEngine engine) {
+        try {
+            getManager().open();
+        } catch (IllegalStateException e) {
+            logger.error("Healthcheck Monitor cannot be opened", e);
+        }
+
+        return false;
+    }
+
+    @Override
     public boolean afterShutdown(PolicyEngine engine) {
         try {
             getManager().stop();
diff --git a/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/HealthCheckManager.java b/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/HealthCheckManager.java
index 8bbd19ad..b310168d 100644
--- a/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/HealthCheckManager.java
+++ b/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/HealthCheckManager.java
@@ -2,7 +2,7 @@
  * ============LICENSE_START=======================================================
  * ONAP
  * ================================================================================
- * Copyright (C) 2019-2021 AT&T Intellectual Property. All rights reserved.
+ * Copyright (C) 2019-2022 AT&T Intellectual Property. All rights reserved.
  * ================================================================================
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -20,20 +20,36 @@
 
 package org.onap.policy.drools.healthcheck;
 
+import com.google.common.base.Strings;
 import java.util.ArrayList;
+import java.util.Arrays;
 import java.util.List;
+import java.util.Map;
 import java.util.Properties;
+import java.util.concurrent.CompletableFuture;
+import java.util.concurrent.TimeUnit;
+import java.util.function.Function;
+import java.util.function.Supplier;
+import java.util.regex.Pattern;
+import java.util.stream.Collectors;
 import javax.ws.rs.core.Response;
 import lombok.AccessLevel;
 import lombok.Getter;
-import lombok.ToString;
+import lombok.NonNull;
+import lombok.Setter;
+import org.apache.commons.lang3.ArrayUtils;
+import org.eclipse.jetty.http.HttpStatus;
 import org.onap.policy.common.endpoints.http.client.HttpClient;
 import org.onap.policy.common.endpoints.http.client.HttpClientFactory;
 import org.onap.policy.common.endpoints.http.client.HttpClientFactoryInstance;
 import org.onap.policy.common.endpoints.http.server.HttpServletServer;
 import org.onap.policy.common.endpoints.http.server.HttpServletServerFactory;
 import org.onap.policy.common.endpoints.http.server.HttpServletServerFactoryInstance;
+import org.onap.policy.drools.controller.DroolsController;
 import org.onap.policy.drools.persistence.SystemPersistenceConstants;
+import org.onap.policy.drools.system.PolicyController;
+import org.onap.policy.drools.system.PolicyControllerConstants;
+import org.onap.policy.drools.system.PolicyControllerFactory;
 import org.onap.policy.drools.system.PolicyEngine;
 import org.onap.policy.drools.system.PolicyEngineConstants;
 import org.slf4j.Logger;
@@ -43,19 +59,52 @@ import org.slf4j.LoggerFactory;
 /**
  * Healthcheck Monitor.
  */
+
 @Getter
-@ToString
 public class HealthCheckManager implements HealthCheck {
 
+    protected static final Logger logger = LoggerFactory.getLogger(HealthCheckManager.class);
+
+    protected static final Pattern COMMA_SPACE_PATTERN = Pattern.compile("\\s*,\\s*");
+
+    protected static final String ENGINE_NAME = "PDP-D";
+    protected static final String HEALTHCHECK_SERVER = "HEALTHCHECK";  // expected healthcheck server name in config
+    protected static final String LIVENESS_SERVER = "LIVENESS";        // expected liveness server name in config
+
+    protected static final int SUCCESS_CODE = 200;
+
+    protected static final int BRAINLESS_CODE = 201;
+    protected static final String BRAINLESS_MESSAGE = "no rules configured";
+
+    protected static final String ENABLED_MESSAGE = "enabled";
+
+    protected static final int DISABLED_CODE = 400;
+    protected static final String DISABLED_MESSAGE = "disabled";
+
+    protected static final int INPROG_CODE = 100;
+    protected static final String INPROG_MESSAGE = "test in progress";
+
+    protected static final int TIMEOUT_CODE = 3000;
+    protected static final String TIMEOUT_MESSAGE = "healthcheck timeout";
+
+    protected static final int UNREACHABLE_CODE = 9000;
+    protected static final String UNREACHABLE_MESSAGE = "cannot reach component";
+
+    public static final String  UNKNOWN_ENTITY = "unknown";
+    public static final int UNKNOWN_ENTITY_CODE = 9010;
+    public static final String UNKNOWN_ENTITY_MESSAGE = "unknown entity";
+
+    protected static final long DEFAULT_TIMEOUT_SECONDS = 10;
+
     /**
-     * Logger.
+     * Healthcheck Server.
      */
-    private static Logger logger = LoggerFactory.getLogger(HealthCheckManager.class);
+    protected HttpServletServer healthcheckServer;
 
     /**
-     * Attached http servers.
+     * Liveness Server.
      */
-    protected List<HttpServletServer> servers = new ArrayList<>();
+    protected HttpServletServer livenessServer;
 
     /**
      * Attached http clients.
@@ -63,150 +112,362 @@ public class HealthCheckManager implements HealthCheck {
     protected List<HttpClient> clients = new ArrayList<>();
 
     /**
+     * Attached controllers.
+     */
+    protected List<PolicyController> controllers = new ArrayList<>();
+
+    /**
      * Healthcheck configuration.
      */
     @Getter(AccessLevel.NONE)
-    @ToString.Exclude
     protected Properties healthCheckProperties = null;
 
-    /**
-     * {@inheritDoc}.
-     */
+    @Setter
+    @Getter
+    protected Long timeoutSeconds = DEFAULT_TIMEOUT_SECONDS;
+
     @Override
     public Reports healthCheck() {
-        var reports = new Reports();
-        boolean thisEngineIsAlive = getEngineManager().isAlive();
-        reports.setHealthy(thisEngineIsAlive);
-
-        var engineReport = new Report();
-        engineReport.setHealthy(thisEngineIsAlive);
-        engineReport.setName("PDP-D");
-        engineReport.setUrl("self");
-        engineReport.setCode(thisEngineIsAlive ? 200 : 500);
-        engineReport.setMessage(thisEngineIsAlive ? "alive" : "not alive");
-        reports.getDetails().add(engineReport);
-
-        for (HttpClient client : clients) {
-            var report = new Report();
-            report.setName(client.getName());
-            report.setUrl(client.getBaseUrl());
+        // get first the engine summary report for setting start time
+        var engineSummary = engineHealthcheck();
+        if (!isEngineAlive()) {
+            logger.info("controller healthchecks ignored as engine is not active");
+            return engineSummary;
+        }
+
+        CompletableFuture<Report>[] reportFutures =
+            ArrayUtils.addAll(futures(getControllers()), futures(getClients()));
+
+        return summary(engineSummary, reportFutures);
+    }
+
+    @Override
+    public Reports engineHealthcheck() {
+        /*
+         * An engine report is special as there always should be 1
+         * report at each system or component healthcheck, since it
+         * is the umbrella component.  Since it does not do IO,
+         * it is generated synchronously which is different from
+         * HTTP clients or Policy Controllers which are asynchronous
+         * with timeout safeties.
+         */
+        var summary = new Reports();
+
+        var engineReport = reportOnEngine();
+        summary.getDetails().add(engineReport);
+        summary.setHealthy(engineReport.isHealthy());
+
+        return summary.setEndTime();
+    }
+
+    @Override
+    public Reports controllerHealthcheck() {
+        if (!isEngineAlive()) {
+            logger.info("controller healthchecks ignored as engine is not active");
+            return engineHealthcheck();
+        }
+
+        CompletableFuture<Report>[] reportFutures = futures(getControllers());
+        return summary(engineHealthcheck(), reportFutures);
+    }
+
+    @Override
+    public Reports controllerHealthcheck(@NonNull PolicyController controller) {
+        /*
+         * allow individual healthchecks without consulting engine state,
+         * it could be useful for troubleshooting.
+         */
+        CompletableFuture<Report>[] reportFutures = futures(List.of(controller));
+        return summary(engineHealthcheck(), reportFutures);
+    }
+
+    @Override
+    public Reports clientHealthcheck() {
+        if (!isEngineAlive()) {
+            logger.info("client healthchecks ignored as engine is not active");
+            return engineHealthcheck();
+        }
+
+        CompletableFuture<Report>[] reportFutures = futures(getClients());
+        return summary(engineHealthcheck(), reportFutures);
+    }
+
+    @Override
+    public Reports clientHealthcheck(@NonNull HttpClient client) {
+        /*
+         * allow individual healthchecks without consulting engine state,
+         * it could be useful for troubleshooting.
+         */
+        CompletableFuture<Report>[] reportFutures = futures(List.of(client));
+        return summary(engineHealthcheck(), reportFutures);
+    }
+
+    protected Reports summary(@NonNull Reports summary, @NonNull CompletableFuture<Report>[] futures) {
+        CompletableFuture.allOf(futures).join();
+
+        Arrays.stream(futures)
+                .map(CompletableFuture::join)
+                .forEach(summary.getDetails()::add);
+
+        summary.setHealthy(summary.getDetails()
+                .stream()
+                .map(this::timeout)
+                .map(Report::isHealthy)
+                .reduce(true, Boolean::logicalAnd));
+
+        return summary.setEndTime();
+    }
+
+    protected Report timeout(Report report) {
+        if (report.getCode() == INPROG_CODE && INPROG_MESSAGE.equals(report.getMessage())) {
+            report.setHealthy(false);
+            report.setCode(TIMEOUT_CODE);
+            report.setMessage(TIMEOUT_MESSAGE);
+        }
+
+        return report;
+    }
+
+    protected <T> CompletableFuture<Report>[] futures(List<T> entities) {
+        return entities.stream()
+                .map(this::supplier)
+                .toArray(CompletableFuture[]::new);
+    }
+
+    protected Report reportOnEngine() {
+        var report = new Report();
+
+        report.setName(ENGINE_NAME);
+        report.setUrl("engine");
+        report.setHealthy(isEngineAlive());
+
+        if (isEngineAlive()) {
+            report.setCode(SUCCESS_CODE);
+            report.setMessage(ENABLED_MESSAGE);
+        } else {
+            report.setCode(DISABLED_CODE);
+            report.setMessage(DISABLED_MESSAGE);
+        }
+
+        return report.setEndTime();
+    }
+
+    protected Report reportOn(@NonNull PolicyController controller, @NonNull Report report) {
+        report.setName(controller.getName());
+
+        if (!controller.isAlive()) {
+            report.setUrl(controller.getName());
+            report.setCode(DISABLED_CODE);
+            report.setMessage(DISABLED_MESSAGE);
+            report.setHealthy(false);
+            return report.setEndTime();
+        }
+
+        DroolsController drools = controller.getDrools();
+        report.setUrl(getControllerCoordinates(drools));
+
+        if (!drools.isBrained()) {
+            report.setCode(BRAINLESS_CODE);
+            report.setMessage(BRAINLESS_MESSAGE);
             report.setHealthy(true);
-            try {
-                var response = client.get();
-                report.setCode(response.getStatus());
-                if (report.getCode() != 200) {
-                    report.setHealthy(false);
-                    reports.setHealthy(false);
-                }
-
-                report.setMessage(getHttpBody(response, client));
-            } catch (Exception e) {
-                logger.warn("{}: cannot contact http-client {}", this, client, e);
-
-                report.setHealthy(false);
-                reports.setHealthy(false);
-            }
-            reports.getDetails().add(report);
+            return report.setEndTime();
         }
-        return reports;
+
+        /*
+         * potentially blocking drools application operation
+         */
+
+        return reportOn(controller.getDrools(), report).setEndTime();
+    }
+
+    private Report reportOn(@NonNull DroolsController drools, @NonNull Report report) {
+        if (!drools.isAlive()) {
+            report.setCode(DISABLED_CODE);
+            report.setMessage(DISABLED_MESSAGE);
+            return report;
+        }
+
+        /*
+         * The code below will block in unresponsive applications.
+         */
+        long factCount = 0;
+        StringBuilder message = new StringBuilder();
+        for (String sessionName: drools.getSessionNames()) {
+            message.append("[").append(sessionName).append(":").append(getFactTypes(drools, sessionName)).append("]");
+            factCount += getFactCount(drools, sessionName);
+        }
+
+        /* success */
+
+        report.setHealthy(true);
+        report.setCode(factCount);
+        report.setMessage("" + message);
+
+        return report;
+    }
+
+    protected Report reportOn(@NonNull HttpClient client, @NonNull Report report) {
+        report.setName(client.getName());
+        report.setUrl(client.getBaseUrl());
+
+        try {
+            Response response = client.get();
+            report.setHealthy(response.getStatus() == HttpStatus.OK_200);
+            report.setCode(response.getStatus());
+            report.setMessage(response.getStatusInfo().getReasonPhrase());
+        } catch (Exception e) {
+            report.setHealthy(false);
+            report.setCode(UNREACHABLE_CODE);
+            report.setMessage(UNREACHABLE_MESSAGE);
+            logger.info("{}: cannot contact http-client {}", this, client.getName(), e);
+        }
+
+        return report.setEndTime();
+    }
+
+    private <T> Supplier<Report> createSupplier(@NonNull T entity, @NonNull Report report) {
+        if (entity instanceof PolicyController) {
+            return () -> reportOn((PolicyController) entity, report);
+        } else if (entity instanceof HttpClient) {
+            return () -> reportOn((HttpClient) entity, report);
+        } else {
+            return () -> reportOnUnknown(entity, report);
+        }
+    }
+
+    private Report reportOnUnknown(Object o, Report report) {
+        report.setName(UNKNOWN_ENTITY);
+        report.setCode(UNKNOWN_ENTITY_CODE);
+        report.setMessage(UNKNOWN_ENTITY_MESSAGE);
+        report.setHealthy(false);
+        report.setUrl(o.getClass().getName());
+        return report.setEndTime();
+    }
+
+    protected <T> CompletableFuture<Report> supplier(T entity) {
+        var report = new Report();
+        report.setHealthy(false);
+        report.setCode(INPROG_CODE);
+        report.setMessage(INPROG_MESSAGE);
+
+        return CompletableFuture
+                       .supplyAsync(createSupplier(entity, report))
+                       .completeOnTimeout(report, getTimeoutSeconds(), TimeUnit.SECONDS)
+                       .thenApply(HealthCheck.Report::setEndTime);
+    }
+
+    private String getControllerCoordinates(DroolsController drools) {
+        return drools.getGroupId() + ":" + drools.getArtifactId() + ":" + drools.getVersion();
+    }
+
+    protected long getFactCount(DroolsController drools, String sessionName) {
+        return drools.factCount(sessionName);
+    }
+
+    protected Map<String, Integer> getFactTypes(DroolsController drools, String sessionName) {
+        return drools.factClassNames(sessionName);
     }
 
-    /**
-     * {@inheritDoc}.
-     */
     @Override
     public boolean start() {
-
         try {
-            this.healthCheckProperties = getPersistentProperties(HealthCheckFeature.CONFIGURATION_PROPERTIES_NAME);
-            this.servers = getServerFactory().build(healthCheckProperties);
+            this.healthCheckProperties = getPersistentProperties();
+            Map<String, HttpServletServer> servers =
+                    getServerFactory().build(healthCheckProperties).stream()
+                        .collect(Collectors.toMap(HttpServletServer::getName, Function.identity()));
+
+            this.healthcheckServer = servers.get(HEALTHCHECK_SERVER);
+            this.livenessServer = servers.get(LIVENESS_SERVER);
+
+            setTimeoutSeconds(
+                    Long.valueOf(this.healthCheckProperties
+                            .getProperty("liveness.controllers.timeout",
+                                    "" + DEFAULT_TIMEOUT_SECONDS)));
+
             this.clients = getClientFactory().build(healthCheckProperties);
 
-            for (HttpServletServer server : servers) {
-                if (server.isAaf()) {
-                    server.addFilterClass(null, AafHealthCheckFilter.class.getName());
-                }
-                startServer(server);
-            }
+            return startHealthcheckServer();
         } catch (Exception e) {
-            logger.warn("{}: cannot start", this, e);
+            logger.warn("{}: cannot start", HEALTHCHECK_SERVER, e);
             return false;
         }
-
-        return true;
     }
 
-    /**
-     * {@inheritDoc}.
-     */
     @Override
-    public boolean stop() {
+    public void open() {
+        if (this.livenessServer != null) {
+            startServer(this.livenessServer);
+        }
 
-        for (HttpServletServer server : servers) {
-            try {
-                server.stop();
-            } catch (Exception e) {
-                logger.warn("{}: cannot stop http-server {}", this, server, e);
-            }
+        String controllerNames = this.healthCheckProperties.getProperty("liveness.controllers");
+        if (Strings.isNullOrEmpty(controllerNames)) {
+            logger.info("no controllers to live check");
+            return;
+        }
+
+        if ("*".equals(controllerNames)) {
+            // monitor all controllers
+            this.controllers = getControllerFactory().inventory();
+            return;
         }
 
-        for (HttpClient client : clients) {
+        for (String controllerName : COMMA_SPACE_PATTERN.split(controllerNames)) {
             try {
-                client.stop();
-            } catch (Exception e) {
-                logger.warn("{}: cannot stop http-client {}", this, client, e);
+                this.controllers.add(getControllerFactory().get(controllerName));
+            } catch (RuntimeException rex) {
+                logger.warn("cannot get controller {}", controllerName);
             }
         }
+    }
+
+    @Override
+    public boolean stop() {
+        if (this.healthcheckServer != null) {
+            this.healthcheckServer.stop();
+        }
+
+        if (this.livenessServer != null) {
+            this.livenessServer.stop();
+        }
+
+        for (HttpClient client : getClients()) {
+            logger.warn("{}: cannot stop http-client", client.getName());
+            client.stop();
+        }
 
         return true;
     }
 
-    /**
-     * {@inheritDoc}.
-     */
     @Override
     public void shutdown() {
         this.stop();
     }
 
-    /**
-     * {@inheritDoc}.
-     */
     @Override
     public synchronized boolean isAlive() {
         return this.healthCheckProperties != null;
     }
 
-    /**
-     * Gets the body from the response.
-     *
-     * @param response response from which to get the body
-     * @param client HTTP client from which the response was received
-     * @return the response body
-     */
-    public String getHttpBody(Response response, HttpClient client) {
+    protected boolean startHealthcheckServer() {
+        if (this.healthcheckServer == null) {
+            logger.warn("no {} server found", HEALTHCHECK_SERVER);
+            return false;
+        }
 
-        String body = null;
-        try {
-            body = HttpClient.getBody(response, String.class);
-        } catch (Exception e) {
-            logger.info("{}: cannot get body from http-client {}", this, client, e);
+        if (this.healthcheckServer.isAaf()) {
+            this.healthcheckServer.addFilterClass(null, AafHealthCheckFilter.class.getName());
         }
 
-        return body;
+        return startServer(this.healthcheckServer);
     }
 
-    /**
-     * Starts an HTTP server.
-     *
-     * @param server server to be started
-     */
-    public void startServer(HttpServletServer server) {
+    protected boolean startServer(HttpServletServer server) {
         try {
-            server.start();
+            return server.start();
         } catch (Exception e) {
-            logger.warn("{}: cannot start http-server {}", this, server, e);
+            logger.warn("cannot start http-server {}", server.getName(), e);
         }
+        return false;
     }
 
     // the following methods may be overridden by junit tests
@@ -215,6 +476,10 @@ public class HealthCheckManager implements HealthCheck {
         return PolicyEngineConstants.getManager();
     }
 
+    protected boolean isEngineAlive() {
+        return getEngineManager().isAlive();
+    }
+
     protected HttpServletServerFactory getServerFactory() {
         return HttpServletServerFactoryInstance.getServerFactory();
     }
@@ -223,7 +488,11 @@ public class HealthCheckManager implements HealthCheck {
         return HttpClientFactoryInstance.getClientFactory();
     }
 
-    protected Properties getPersistentProperties(String propertyName) {
-        return SystemPersistenceConstants.getManager().getProperties(propertyName);
+    protected PolicyControllerFactory getControllerFactory() {
+        return PolicyControllerConstants.getFactory();
+    }
+
+    protected Properties getPersistentProperties() {
+        return SystemPersistenceConstants.getManager().getProperties(HealthCheckFeature.CONFIGURATION_PROPERTIES_NAME);
     }
 }
diff --git a/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/RestHealthCheck.java b/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/RestHealthCheck.java
index e323d405..5b36c5a3 100644
--- a/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/RestHealthCheck.java
+++ b/feature-healthcheck/src/main/java/org/onap/policy/drools/healthcheck/RestHealthCheck.java
@@ -1,8 +1,8 @@
 /*-
  * ============LICENSE_START=======================================================
- * feature-healthcheck
+ * ONAP
  * ================================================================================
- * Copyright (C) 2017-2019 AT&T Intellectual Property. All rights reserved.
+ * Copyright (C) 2017-2019, 2022 AT&T Intellectual Property. All rights reserved.
  * ================================================================================
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -22,17 +22,26 @@ package org.onap.policy.drools.healthcheck;
 
 import io.swagger.annotations.Api;
 import io.swagger.annotations.ApiOperation;
+import io.swagger.annotations.ApiParam;
 import io.swagger.annotations.Info;
 import io.swagger.annotations.SwaggerDefinition;
 import io.swagger.annotations.Tag;
 import javax.ws.rs.GET;
 import javax.ws.rs.Path;
+import javax.ws.rs.PathParam;
 import javax.ws.rs.Produces;
 import javax.ws.rs.core.MediaType;
 import javax.ws.rs.core.Response;
+import org.onap.policy.common.endpoints.http.client.HttpClientFactory;
+import org.onap.policy.common.endpoints.http.client.HttpClientFactoryInstance;
 import org.onap.policy.common.endpoints.http.server.YamlMessageBodyHandler;
 import org.onap.policy.drools.healthcheck.HealthCheck.Reports;
+import org.onap.policy.drools.system.PolicyControllerConstants;
+import org.onap.policy.drools.system.PolicyControllerFactory;
 
+/**
+ * REST Healthcheck JAX-RS.
+ */
 @Path("/")
 @Api
 @Produces({MediaType.APPLICATION_JSON, YamlMessageBodyHandler.APPLICATION_YAML})
@@ -51,6 +60,10 @@ import org.onap.policy.drools.healthcheck.HealthCheck.Reports;
     )
 public class RestHealthCheck {
 
+    /**
+     * System healthcheck per configuration.
+     */
+
     @GET
     @Path("healthcheck")
     @ApiOperation(
@@ -60,17 +73,119 @@ public class RestHealthCheck {
             response = Reports.class
             )
     public Response healthcheck() {
-        return Response.status(Response.Status.OK).entity(HealthCheckConstants.getManager().healthCheck()).build();
+        var summary = getHealthcheckManager().healthCheck();
+        return getResponse(summary);
     }
 
+    /**
+     * Engine Healthcheck.
+     */
+
     @GET
-    @Path("healthcheck/configuration")
+    @Path("healthcheck/engine")
     @ApiOperation(
-            value = "Configuration",
-            notes = "Provides the Healthcheck server configuration and monitored REST clients",
+            value = "Engine Healthcheck",
+            notes = "Provides a Healthcheck on the engine",
             response = HealthCheck.class
-            )
-    public HealthCheck configuration() {
+    )
+    public Response engine() {
+        var summary = getHealthcheckManager().engineHealthcheck();
+        return getResponse(summary);
+    }
+
+    /**
+     * Healthcheck on the controllers.
+     */
+
+    @GET
+    @Path("healthcheck/controllers")
+    @ApiOperation(
+            value = "Controllers Healthcheck",
+            notes = "Provides a Healthcheck on the configured controllers",
+            response = Reports.class
+    )
+    public Response controllers() {
+        var summary = getHealthcheckManager().controllerHealthcheck();
+        return getResponse(summary);
+    }
+
+    /**
+     * Healthcheck a controller.
+     */
+
+    @GET
+    @Path("healthcheck/controllers/{controllerName}")
+    @ApiOperation(
+            value = "Controller Healthcheck",
+            notes = "Provides a Healthcheck on a configured controller",
+            response = Reports.class
+    )
+    public Response controllers(@ApiParam(value = "Policy Controller Name",
+            required = true) @PathParam("controllerName") String controllerName) {
+        try {
+            var controller = getControllerFactory().get(controllerName);
+            var summary = getHealthcheckManager().controllerHealthcheck(controller);
+            return getResponse(summary);
+        } catch (final IllegalArgumentException e) {
+            return Response.status(Response.Status.NOT_FOUND).build();
+        } catch (final IllegalStateException e) {
+            return Response.status(Response.Status.NOT_ACCEPTABLE).build();
+        }
+    }
+
+    /**
+     * Healthcheck on the Http Clients per configuration.
+     */
+
+    @GET
+    @Path("healthcheck/clients")
+    @ApiOperation(
+            value = "Http Clients Healthcheck",
+            notes = "Provides a Healthcheck on the configured HTTP clients",
+            response = Reports.class
+    )
+    public Response clients() {
+        var summary = getHealthcheckManager().clientHealthcheck();
+        return getResponse(summary);
+    }
+
+    /**
+     * Healthcheck a on a Http Client.
+     */
+
+    @GET
+    @Path("healthcheck/clients/{clientName}")
+    @ApiOperation(
+            value = "Http Client Healthcheck",
+            notes = "Provides a Healthcheck on a configured HTTP client",
+            response = Reports.class
+    )
+    public Response clients(@ApiParam(value = "Http Client Name",
+            required = true) @PathParam("clientName") String clientName) {
+        try {
+            var client = getClientFactory().get(clientName);
+            var summary = getHealthcheckManager().clientHealthcheck(client);
+            return getResponse(summary);
+        } catch (final IllegalArgumentException e) {
+            return Response.status(Response.Status.NOT_FOUND).build();
+        }
+    }
+
+    protected Response getResponse(Reports summary) {
+        return Response.status(summary.isHealthy() ? Response.Status.OK : Response.Status.SERVICE_UNAVAILABLE)
+                .entity(summary).build();
+    }
+
+    protected HttpClientFactory getClientFactory() {
+        return HttpClientFactoryInstance.getClientFactory();
+    }
+
+    protected PolicyControllerFactory getControllerFactory() {
+        return PolicyControllerConstants.getFactory();
+    }
+
+    protected HealthCheck getHealthcheckManager() {
         return HealthCheckConstants.getManager();
     }
+
 }