Cherry-pick branch 'owls-112700-cohchk-stopscript' into 'main'

rjeberhard · rjeberhard · commit 7e5c7234aeab · 2023-11-01T22:44:55.000Z
diff --git a/documentation/domains/Cluster.json b/documentation/domains/Cluster.json
@@ -376,6 +376,11 @@
     "Shutdown": {
       "type": "object",
       "properties": {
+        "skipWaitingCohEndangeredState": {
+          "default": false,
+          "description": "For graceful shutdown only, set to true to skip waiting for Coherence Cache Cluster service MBean HAStatus in safe state before shutdown. By default, the operator will wait until it is safe to shutdown the Coherence Cache Cluster. Defaults to false.",
+          "type": "boolean"
+        },
         "ignoreSessions": {
           "default": false,
           "description": "For graceful shutdown only, indicates to ignore pending HTTP sessions during in-flight work handling. Defaults to false.",
diff --git a/documentation/domains/Cluster.md b/documentation/domains/Cluster.md
@@ -115,6 +115,7 @@ The specification of the operation of the WebLogic cluster. Required.
 | --- | --- | --- |
 | `ignoreSessions` | Boolean | For graceful shutdown only, indicates to ignore pending HTTP sessions during in-flight work handling. Defaults to false. |
 | `shutdownType` | string | Specifies how the operator will shut down server instances. Legal values are `Graceful` and `Forced`. Defaults to `Graceful`. |
+| `skipWaitingCohEndangeredState` | Boolean | For graceful shutdown only, set to true to skip waiting for Coherence Cache Cluster service MBean HAStatus in safe state before shutdown. By default, the operator will wait until it is safe to shutdown the Coherence Cache Cluster. Defaults to false. |
 | `timeoutSeconds` | integer | For graceful shutdown only, number of seconds to wait before aborting in-flight work and shutting down the server. Defaults to 30 seconds. |
 | `waitForAllSessions` | Boolean | For graceful shutdown only, set to true to wait for all HTTP sessions during in-flight work handling; false to wait for non-persisted HTTP sessions only. Defaults to false. |
 
diff --git a/documentation/domains/Domain.json b/documentation/domains/Domain.json
@@ -1187,6 +1187,11 @@
     "Shutdown": {
       "type": "object",
       "properties": {
+        "skipWaitingCohEndangeredState": {
+          "default": false,
+          "description": "For graceful shutdown only, set to true to skip waiting for Coherence Cache Cluster service MBean HAStatus in safe state before shutdown. By default, the operator will wait until it is safe to shutdown the Coherence Cache Cluster. Defaults to false.",
+          "type": "boolean"
+        },
         "ignoreSessions": {
           "default": false,
           "description": "For graceful shutdown only, indicates to ignore pending HTTP sessions during in-flight work handling. Defaults to false.",
diff --git a/documentation/domains/Domain.md b/documentation/domains/Domain.md
@@ -278,6 +278,7 @@ The current status of the operation of the WebLogic domain. Updated automaticall
 | --- | --- | --- |
 | `ignoreSessions` | Boolean | For graceful shutdown only, indicates to ignore pending HTTP sessions during in-flight work handling. Defaults to false. |
 | `shutdownType` | string | Specifies how the operator will shut down server instances. Legal values are `Graceful` and `Forced`. Defaults to `Graceful`. |
+| `skipWaitingCohEndangeredState` | Boolean | For graceful shutdown only, set to true to skip waiting for Coherence Cache Cluster service MBean HAStatus in safe state before shutdown. By default, the operator will wait until it is safe to shutdown the Coherence Cache Cluster. Defaults to false. |
 | `timeoutSeconds` | integer | For graceful shutdown only, number of seconds to wait before aborting in-flight work and shutting down the server. Defaults to 30 seconds. |
 | `waitForAllSessions` | Boolean | For graceful shutdown only, set to true to wait for all HTTP sessions during in-flight work handling; false to wait for non-persisted HTTP sessions only. Defaults to false. |
 
diff --git a/kubernetes/crd/cluster-crd.yaml b/kubernetes/crd/cluster-crd.yaml
@@ -5,7 +5,7 @@ apiVersion: apiextensions.k8s.io/v1
 kind: CustomResourceDefinition
 metadata:
   annotations:
-    weblogic.sha256: 9ac551460201264f4eb69fabd7c54120c2e30492143ad81f47965567f2a7ae79
+    weblogic.sha256: 7146de067298f75bf803e2c8acf9f88980d1d5770e375122bcbb1fe4e6f9a64b
   name: clusters.weblogic.oracle
 spec:
   group: weblogic.oracle
@@ -2390,6 +2390,14 @@ spec:
                     description: Configures how the operator should shut down the
                       server instance.
                     properties:
+                      skipWaitingCohEndangeredState:
+                        default: false
+                        description: For graceful shutdown only, set to true to skip
+                          waiting for Coherence Cache Cluster service MBean HAStatus
+                          in safe state before shutdown. By default, the operator
+                          will wait until it is safe to shutdown the Coherence Cache
+                          Cluster. Defaults to false.
+                        type: boolean
                       ignoreSessions:
                         default: false
                         description: For graceful shutdown only, indicates to ignore
diff --git a/kubernetes/crd/domain-crd.yaml b/kubernetes/crd/domain-crd.yaml
@@ -5,7 +5,7 @@ apiVersion: apiextensions.k8s.io/v1
 kind: CustomResourceDefinition
 metadata:
   annotations:
-    weblogic.sha256: 1b5fbb6128eb44897c07ef1304cf64bc8682f7f4ef2454dde6e861a04941b635
+    weblogic.sha256: d33dc80911f1901a703fe95438c2e4fc39271775bfecddd35893f8422d97caa3
   name: domains.weblogic.oracle
 spec:
   group: weblogic.oracle
@@ -3331,6 +3331,14 @@ spec:
                         description: Configures how the operator should shut down
                           the server instance.
                         properties:
+                          skipWaitingCohEndangeredState:
+                            default: false
+                            description: For graceful shutdown only, set to true to
+                              skip waiting for Coherence Cache Cluster service MBean
+                              HAStatus in safe state before shutdown. By default,
+                              the operator will wait until it is safe to shutdown
+                              the Coherence Cache Cluster. Defaults to false.
+                            type: boolean
                           ignoreSessions:
                             default: false
                             description: For graceful shutdown only, indicates to
@@ -6414,6 +6422,14 @@ spec:
                     description: Configures how the operator should shut down the
                       server instance.
                     properties:
+                      skipWaitingCohEndangeredState:
+                        default: false
+                        description: For graceful shutdown only, set to true to skip
+                          waiting for Coherence Cache Cluster service MBean HAStatus
+                          in safe state before shutdown. By default, the operator
+                          will wait until it is safe to shutdown the Coherence Cache
+                          Cluster. Defaults to false.
+                        type: boolean
                       ignoreSessions:
                         default: false
                         description: For graceful shutdown only, indicates to ignore
@@ -9155,6 +9171,14 @@ spec:
                             the server instance.
                           type: object
                           properties:
+                            skipWaitingCohEndangeredState:
+                              default: false
+                              description: For graceful shutdown only, set to true
+                                to skip waiting for Coherence Cache Cluster service
+                                MBean HAStatus in safe state before shutdown. By default,
+                                the operator will wait until it is safe to shutdown
+                                the Coherence Cache Cluster. Defaults to false.
+                              type: boolean
                             ignoreSessions:
                               default: false
                               description: For graceful shutdown only, indicates to
diff --git a/operator/src/main/java/oracle/kubernetes/operator/helpers/PodStepContext.java b/operator/src/main/java/oracle/kubernetes/operator/helpers/PodStepContext.java
@@ -613,6 +613,11 @@ private void updateEnvForShutdown(List<V1EnvVar> env) {
       addDefaultEnvVarIfMissing(env, "SHUTDOWN_WAIT_FOR_ALL_SESSIONS",
               String.valueOf(shutdown.getWaitForAllSessions()));
     }
+    if (!shutdown.getSkipWaitingCohEndangeredState()
+            .equals(Shutdown.DEFAULT_SKIP_WAIT_COH_ENDANGERED_STATE)) {
+      addDefaultEnvVarIfMissing(env, "SHUTDOWN_SKIP_WAIT_COH_ENDANGERED_STATE",
+              String.valueOf(shutdown.getSkipWaitingCohEndangeredState()));
+    }
   }
 
   private Shutdown getShutdownSpec() {
diff --git a/operator/src/main/java/oracle/kubernetes/operator/steps/ShutdownManagedServerStep.java b/operator/src/main/java/oracle/kubernetes/operator/steps/ShutdownManagedServerStep.java
@@ -221,7 +221,16 @@ protected PortDetails getPortDetails() {
     private Integer getWlsServerPort() {
       Integer listenPort = Optional.ofNullable(getWlsServerConfig()).map(WlsServerConfig::getListenPort)
           .orElse(null);
-
+      Integer adminPort = Optional.ofNullable(getWlsServerConfig()).map(WlsServerConfig::getAdminPort)
+              .orElse(null);
+      Integer sslListenPort = Optional.ofNullable(getWlsServerConfig()).map(WlsServerConfig::getSslListenPort)
+              .orElse(null);
+      if (adminPort != null) {
+        return adminPort;
+      }
+      if (sslListenPort != null) {
+        return sslListenPort;
+      }
       if (listenPort == null) {
         // This can only happen if the running server pod does not exist in the WLS Domain.
         // This is a rare case where the server was deleted from the WLS Domain config.
diff --git a/operator/src/main/java/oracle/kubernetes/weblogic/domain/model/BaseConfiguration.java b/operator/src/main/java/oracle/kubernetes/weblogic/domain/model/BaseConfiguration.java
@@ -274,6 +274,12 @@ void addLimitRequirement(String resource, String quantity) {
     serverPod.addLimitRequirement(resource, quantity);
   }
 
+  void setShutdown(Shutdown shutdown) {
+    serverPod.setShutdown(shutdown.getShutdownType(),
+            shutdown.getTimeoutSeconds(), shutdown.getIgnoreSessions(), shutdown.getWaitForAllSessions(),
+            shutdown.getSkipWaitingCohEndangeredState());
+  }
+
   V1PodSecurityContext getPodSecurityContext() {
     return Optional.ofNullable(serverPod.getPodSecurityContext()).orElse(getDefaultPodSecurityContext());
   }
diff --git a/operator/src/main/java/oracle/kubernetes/weblogic/domain/model/ServerPod.java b/operator/src/main/java/oracle/kubernetes/weblogic/domain/model/ServerPod.java
@@ -362,12 +362,14 @@ Shutdown getShutdown() {
     return this.shutdown;
   }
 
-  void setShutdown(ShutdownType shutdownType, Long timeoutSeconds, Boolean ignoreSessions, Boolean waitForAllSessions) {
+  void setShutdown(ShutdownType shutdownType, Long timeoutSeconds, Boolean ignoreSessions, Boolean waitForAllSessions,
+                   Boolean skipWaitingCohEndangeredState) {
     this.shutdown
         .shutdownType(shutdownType)
         .timeoutSeconds(timeoutSeconds)
         .ignoreSessions(ignoreSessions)
-        .waitForAllSessions(waitForAllSessions);
+        .waitForAllSessions(waitForAllSessions)
+        .skipWaitingCohEndangeredState(skipWaitingCohEndangeredState);
   }
 
   ProbeTuning getReadinessProbeTuning() {
diff --git a/operator/src/main/java/oracle/kubernetes/weblogic/domain/model/Shutdown.java b/operator/src/main/java/oracle/kubernetes/weblogic/domain/model/Shutdown.java
@@ -17,6 +17,7 @@ public class Shutdown {
   public static final Long DEFAULT_TIMEOUT = 30L;
   public static final Boolean DEFAULT_IGNORESESSIONS = Boolean.FALSE;
   public static final Boolean DEFAULT_WAIT_FOR_ALL_SESSIONS = Boolean.FALSE;
+  public static final Boolean DEFAULT_SKIP_WAIT_COH_ENDANGERED_STATE = Boolean.FALSE;
 
   @Description(
       "Specifies how the operator will shut down server instances."
@@ -44,6 +45,14 @@ public class Shutdown {
   @Default(boolDefault = false)
   private Boolean waitForAllSessions;
 
+  @Description(
+          "For graceful shutdown only, set to true to skip waiting for Coherence Cache Cluster service MBean HAStatus"
+                  + " in safe state before shutdown. By default, the operator will wait until it is"
+                  + " safe to shutdown the Coherence Cache Cluster."
+                  + " Defaults to false.")
+  @Default(boolDefault = false)
+  private Boolean skipWaitingCohEndangeredState;
+
   void copyValues(Shutdown fromShutdown) {
     if (shutdownType == null) {
       shutdownType(fromShutdown.shutdownType);
@@ -58,6 +67,10 @@ void copyValues(Shutdown fromShutdown) {
     if (waitForAllSessions == null) {
       waitForAllSessions(fromShutdown.waitForAllSessions);
     }
+
+    if (skipWaitingCohEndangeredState == null) {
+      skipWaitingCohEndangeredState(fromShutdown.skipWaitingCohEndangeredState);
+    }
   }
 
   public ShutdownType getShutdownType() {
@@ -96,13 +109,24 @@ public Shutdown waitForAllSessions(Boolean waitForAllSessions) {
     return this;
   }
 
+  public Boolean getSkipWaitingCohEndangeredState() {
+    return Optional.ofNullable(skipWaitingCohEndangeredState)
+            .orElse(DEFAULT_SKIP_WAIT_COH_ENDANGERED_STATE);
+  }
+
+  public Shutdown skipWaitingCohEndangeredState(Boolean skipWaitingCohEndangeredState) {
+    this.skipWaitingCohEndangeredState = skipWaitingCohEndangeredState;
+    return this;
+  }
+
   @Override
   public String toString() {
     return new ToStringBuilder(this)
         .append("shutdownType", shutdownType)
         .append("timeoutSeconds", timeoutSeconds)
         .append("ignoreSessions", ignoreSessions)
         .append("waitForAllSessions", waitForAllSessions)
+        .append("skipWaitingCohEndangeredState", skipWaitingCohEndangeredState)
         .toString();
   }
 
@@ -123,6 +147,7 @@ public boolean equals(Object o) {
         .append(timeoutSeconds, that.timeoutSeconds)
         .append(ignoreSessions, that.ignoreSessions)
         .append(waitForAllSessions, that.waitForAllSessions)
+        .append(skipWaitingCohEndangeredState, that.skipWaitingCohEndangeredState)
         .isEquals();
   }
 
@@ -133,6 +158,7 @@ public int hashCode() {
         .append(timeoutSeconds)
         .append(ignoreSessions)
         .append(waitForAllSessions)
+        .append(skipWaitingCohEndangeredState)
         .toHashCode();
   }
 }
diff --git a/operator/src/main/resources/scripts/stop-server.py b/operator/src/main/resources/scripts/stop-server.py
@@ -106,7 +106,9 @@ def checkCoherenceClusterExist(configData):
 def waitUntilCoherenceSafe():
   print ('Shutdown: getting all service Coherence MBeans')
   query='Coherence:type=PartitionAssignment,service=*,*'
-
+  if 'SHUTDOWN_SKIP_WAIT_COH_ENDANGERED_STATE' in os.environ and \
+          os.environ['SHUTDOWN_SKIP_WAIT_COH_ENDANGERED_STATE'] == 'true':
+      return
   # By default, Coherence will use a single WebLogic Runtime MBean server to managed
   # its MBeans.  That server will correspond to the current Coherence senior member,
   # which means that the Coherence MBeans will migrate to the oldest cluster member
@@ -169,6 +171,10 @@ def waitUntilServiceSafeToShutdown(objectName):
       if status != "ENDANGERED":
         break
 
+      safe_status_ha = _checkCacheSafeStatusHA(objectName)
+      if safe_status_ha:
+        break
+
       # Coherence caches are ENDANGERED meaning that we may lose data
       print ('Shutdown: Waiting until it is safe to shutdown Coherence server ...')
       systime.sleep(5)
@@ -180,6 +186,57 @@ def waitUntilServiceSafeToShutdown(objectName):
       systime.sleep(10)
       pass
 
+# Checking the Cache is safe to shutdown.  The logic is incorporated from Coherence Operator
+#    OperatorRestServer.areCacheServicesHA.
+
+def _checkCacheSafeStatusHA(objectName):
+  try:
+    fields = ["StorageEnabled", "StorageEnabledCount", "BackupCount", "PartitionsAll",
+        "OwnedPartitionsPrimary", "StatusHA", "OutgoingTransferCount", "PartitionsEndangered", "PartitionsVulnerable" ]
+
+    if objectName.getKeyProperty('service') == 'PartitionedCache':
+      partitioned_cache_query = "Coherence:name=PartitionedCache,type=Service,*"
+      partitioned_cache_mbeans = mbs.queryMBeans(ObjectName(partitioned_cache_query), None)
+      for item in partitioned_cache_mbeans:
+        fields_dict = {}
+        attrs = mbs.getAttributes(item.getObjectName(), fields)
+        for attr in attrs:
+          fields_dict.update({ attr.getName(): attr.getValue()})
+
+        storage_enabled = fields_dict['StorageEnabled']
+        storage_enabled_count = fields_dict['StorageEnabledCount']
+        backup_count = fields_dict['BackupCount']
+        status_ha = fields_dict['StatusHA']
+        outgoing_transfer_count = fields_dict['OutgoingTransferCount']
+        partitions_endangered = fields_dict['PartitionsEndangered']
+        partitions_vulnerable = fields_dict['PartitionsVulnerable']
+        partitions_all = fields_dict['PartitionsAll']
+        owned_partitions_primary = fields_dict['OwnedPartitionsPrimary']
+
+        if storage_enabled:
+          if storage_enabled_count > 1 and backup_count > 0 and "ENDANGERED" == status_ha:
+            print("StatusHA check failed. Service %s has HA status of %s" % (objectName, status_ha))
+            return False
+
+          if outgoing_transfer_count > 0:
+            print("StatusHA check failed. Service %s distribution in progress" % (objectName))
+            return False
+
+          persistent_query = "Coherence:service=PartitionedCache,type=Persistence,*"
+          persistent_mbeans = mbs.queryMBeans(ObjectName(persistent_query), None)
+          if persistent_mbeans.size() == 1:
+            idle = mbs.getAttribute(persistent_mbeans[0].getObjectName(), 'Idle')
+            if not idle:
+              print("StatusHA check failed. Service %s persistence is not idle" % (objectName))
+              return False
+
+  except:
+    print("Error _checkCacheSafeStatusHA function:")
+    dumpStack()
+
+  return True
+
+
 
 #----------------------------------
 # Main script
diff --git a/operator/src/test/java/oracle/kubernetes/operator/helpers/ManagedPodHelperTest.java b/operator/src/test/java/oracle/kubernetes/operator/helpers/ManagedPodHelperTest.java
@@ -35,6 +35,7 @@
 import oracle.kubernetes.operator.work.Step.StepAndPacket;
 import oracle.kubernetes.weblogic.domain.DomainConfigurator;
 import oracle.kubernetes.weblogic.domain.ServerConfigurator;
+import oracle.kubernetes.weblogic.domain.model.Shutdown;
 import org.hamcrest.Description;
 import org.hamcrest.TypeSafeDiagnosingMatcher;
 import org.junit.jupiter.api.Disabled;
@@ -1293,6 +1294,15 @@ void whenOnlyAdminAndSslPortsAvailable_monitoringExporterSpecifiesAdminPort() {
                both(hasJavaOption("-DWLS_PORT=8001")).and(hasJavaOption("-DWLS_SECURE=true")));
   }
 
+  @Test
+  void whenDomainSetShutdownSkippingCoherenceEndangeredStateHasEnvSet() {
+    Shutdown shutdown = new Shutdown();
+    shutdown.skipWaitingCohEndangeredState(true);
+    getConfigurator().withServerPodShutdownSpec(shutdown);
+    assertThat(
+            getCreatedPodSpecContainer().getEnv(),
+            allOf(hasEnvVar("SHUTDOWN_SKIP_WAIT_COH_ENDANGERED_STATE", "true")));
+  }
 
   @Override
   void setServerPort(int port) {
diff --git a/operator/src/test/java/oracle/kubernetes/weblogic/domain/DomainConfigurator.java b/operator/src/test/java/oracle/kubernetes/weblogic/domain/DomainConfigurator.java
@@ -32,6 +32,7 @@
 import oracle.kubernetes.weblogic.domain.model.DomainSpec;
 import oracle.kubernetes.weblogic.domain.model.InitializeDomainOnPV;
 import oracle.kubernetes.weblogic.domain.model.Model;
+import oracle.kubernetes.weblogic.domain.model.Shutdown;
 
 /**
  * Configures a domain, adding settings independently of the version of the domain representation.
@@ -424,6 +425,8 @@ public abstract DomainConfigurator withFluentdConfiguration(boolean watchIntrosp
                                                               String credentialName, String fluentdConfig,
                                                               List<String> args, List<String> command);
 
+  public abstract DomainConfigurator withServerPodShutdownSpec(Shutdown shutdown);
+
   /**
    * Adds a default server configuration to the domain, if not already present.
    *
diff --git a/operator/src/test/java/oracle/kubernetes/weblogic/domain/model/DomainCommonConfigurator.java b/operator/src/test/java/oracle/kubernetes/weblogic/domain/model/DomainCommonConfigurator.java
@@ -239,6 +239,12 @@ public DomainConfigurator withFluentdConfiguration(boolean watchIntrospectorLog,
     return this;
   }
 
+  @Override
+  public DomainConfigurator withServerPodShutdownSpec(Shutdown shutdown) {
+    getDomainSpec().setShutdown(shutdown);
+    return this;
+  }
+
   private AdminServer getOrCreateAdminServer() {
     return getDomainSpec().getOrCreateAdminServer();
   }
@@ -821,6 +827,7 @@ public ServerConfigurator withPriorityClassName(String priorityClassName) {
       getDomainSpec().setPriorityClassName(priorityClassName);
       return this;
     }
+
   }
 
   class ClusterConfiguratorImpl implements ClusterConfigurator {

Original file line number	Diff line number	Diff line change
`@@ -613,6 +613,11 @@ private void updateEnvForShutdown(List<V1EnvVar> env) {`
`613`	`613`	`addDefaultEnvVarIfMissing(env, "SHUTDOWN_WAIT_FOR_ALL_SESSIONS",`
`614`	`614`	`String.valueOf(shutdown.getWaitForAllSessions()));`
`615`	`615`	`}`
	`616`	`+ if (!shutdown.getSkipWaitingCohEndangeredState()`
	`617`	`+ .equals(Shutdown.DEFAULT_SKIP_WAIT_COH_ENDANGERED_STATE)) {`
	`618`	`+ addDefaultEnvVarIfMissing(env, "SHUTDOWN_SKIP_WAIT_COH_ENDANGERED_STATE",`
	`619`	`+ String.valueOf(shutdown.getSkipWaitingCohEndangeredState()));`
	`620`	`+ }`
`616`	`621`	`}`
`617`	`622`
`618`	`623`	`private Shutdown getShutdownSpec() {`
Original file line number	Diff line number	Diff line change
`@@ -274,6 +274,12 @@ void addLimitRequirement(String resource, String quantity) {`
`274`	`274`	`serverPod.addLimitRequirement(resource, quantity);`
`275`	`275`	`}`
`276`	`276`
	`277`	`+ void setShutdown(Shutdown shutdown) {`
	`278`	`+ serverPod.setShutdown(shutdown.getShutdownType(),`
	`279`	`+ shutdown.getTimeoutSeconds(), shutdown.getIgnoreSessions(), shutdown.getWaitForAllSessions(),`
	`280`	`+ shutdown.getSkipWaitingCohEndangeredState());`
	`281`	`+ }`
	`282`	`+`
`277`	`283`	`V1PodSecurityContext getPodSecurityContext() {`
`278`	`284`	`return Optional.ofNullable(serverPod.getPodSecurityContext()).orElse(getDefaultPodSecurityContext());`
`279`	`285`	`}`