Refactor freeze mode naming and getStatus default value (#1818)
Rename CLUSTER_PAUSE with CLUSTER_FREEZE mode.
Change default return value for LiveInstance#getStatus(): null -> NORMAL. And fix tests accordingly.
diff --git a/helix-core/src/main/java/org/apache/helix/api/status/ClusterManagementMode.java b/helix-core/src/main/java/org/apache/helix/api/status/ClusterManagementMode.java
index 86817d1..a342a8c 100644
--- a/helix-core/src/main/java/org/apache/helix/api/status/ClusterManagementMode.java
+++ b/helix-core/src/main/java/org/apache/helix/api/status/ClusterManagementMode.java
@@ -33,10 +33,10 @@
NORMAL,
/**
- * Puts a cluster into pause mode, which will pause controller and participants.
+ * Puts a cluster into freeze mode, which will freeze controller and participants.
* This can be used to retain the cluster state.
*/
- CLUSTER_PAUSE,
+ CLUSTER_FREEZE,
/** Pause controller only, but not participants. */
CONTROLLER_PAUSE,
@@ -88,7 +88,7 @@
*/
public LiveInstance.LiveInstanceStatus getDesiredParticipantStatus() {
switch (mode) {
- case CLUSTER_PAUSE:
+ case CLUSTER_FREEZE:
return LiveInstance.LiveInstanceStatus.PAUSED;
case NORMAL:
return LiveInstance.LiveInstanceStatus.NORMAL;
diff --git a/helix-core/src/main/java/org/apache/helix/controller/stages/ManagementMessageGenerationPhase.java b/helix-core/src/main/java/org/apache/helix/controller/stages/ManagementMessageGenerationPhase.java
index c795deb..4479d50 100644
--- a/helix-core/src/main/java/org/apache/helix/controller/stages/ManagementMessageGenerationPhase.java
+++ b/helix-core/src/main/java/org/apache/helix/controller/stages/ManagementMessageGenerationPhase.java
@@ -96,9 +96,7 @@
LiveInstance liveInstance = liveInstanceMap.get(instanceName);
Collection<Message> pendingMessages = allInstanceMessages.get(instanceName);
String sessionId = liveInstance.getEphemeralOwner();
- LiveInstanceStatus liveInstanceStatus = liveInstance.getStatus();
- LiveInstanceStatus currentStatus = (liveInstanceStatus == null
- ? LiveInstanceStatus.NORMAL : liveInstanceStatus);
+ LiveInstanceStatus currentStatus = liveInstance.getStatus();
if (needStatusChangeMessage(pendingMessages, currentStatus, desiredStatus)) {
Message statusChangeMessage = MessageUtil.createStatusChangeMessage(currentStatus,
diff --git a/helix-core/src/main/java/org/apache/helix/controller/stages/ManagementModeStage.java b/helix-core/src/main/java/org/apache/helix/controller/stages/ManagementModeStage.java
index b639685..55a320a 100644
--- a/helix-core/src/main/java/org/apache/helix/controller/stages/ManagementModeStage.java
+++ b/helix-core/src/main/java/org/apache/helix/controller/stages/ManagementModeStage.java
@@ -30,7 +30,6 @@
import org.apache.helix.PropertyKey;
import org.apache.helix.PropertyType;
import org.apache.helix.api.status.ClusterManagementMode;
-import org.apache.helix.controller.LogUtil;
import org.apache.helix.controller.dataproviders.ManagementControllerDataProvider;
import org.apache.helix.controller.pipeline.AbstractBaseStage;
import org.apache.helix.controller.pipeline.StageException;
@@ -40,7 +39,6 @@
import org.apache.helix.model.LiveInstance.LiveInstanceStatus;
import org.apache.helix.model.Message;
import org.apache.helix.model.Message.MessageType;
-import org.apache.helix.model.Partition;
import org.apache.helix.model.PauseSignal;
import org.apache.helix.model.Resource;
import org.apache.helix.util.HelixUtil;
@@ -104,7 +102,7 @@
status = ClusterManagementMode.Status.IN_PROGRESS;
}
} else if (pauseSignal.isClusterPause()) {
- type = ClusterManagementMode.Type.CLUSTER_PAUSE;
+ type = ClusterManagementMode.Type.CLUSTER_FREEZE;
if (!instancesFullyFrozen(enabledLiveInstances, liveInstanceMap, allInstanceMessages)) {
status = ClusterManagementMode.Status.IN_PROGRESS;
}
@@ -144,9 +142,9 @@
// If there is any pending message sent by users, status could be computed as in progress.
// Skip recording status change to avoid confusion after cluster is already fully frozen.
- if (ClusterManagementMode.Type.CLUSTER_PAUSE.equals(recordedType)
+ if (ClusterManagementMode.Type.CLUSTER_FREEZE.equals(recordedType)
&& ClusterManagementMode.Status.COMPLETED.equals(recordedStatus)
- && ClusterManagementMode.Type.CLUSTER_PAUSE.equals(mode.getMode())
+ && ClusterManagementMode.Type.CLUSTER_FREEZE.equals(mode.getMode())
&& ClusterManagementMode.Status.IN_PROGRESS.equals(mode.getStatus())) {
LOG.info("Skip recording status mode={}, status={}, because cluster is fully frozen",
mode.getMode(), mode.getStatus());
diff --git a/helix-core/src/main/java/org/apache/helix/manager/zk/ZKHelixAdmin.java b/helix-core/src/main/java/org/apache/helix/manager/zk/ZKHelixAdmin.java
index 73c699e..05bef08 100644
--- a/helix-core/src/main/java/org/apache/helix/manager/zk/ZKHelixAdmin.java
+++ b/helix-core/src/main/java/org/apache/helix/manager/zk/ZKHelixAdmin.java
@@ -509,7 +509,7 @@
// TODO: support other modes
switch (mode) {
- case CLUSTER_PAUSE:
+ case CLUSTER_FREEZE:
enableClusterPauseMode(clusterName, request.isCancelPendingST(), reason);
break;
case NORMAL:
diff --git a/helix-core/src/main/java/org/apache/helix/messaging/handling/HelixTaskExecutor.java b/helix-core/src/main/java/org/apache/helix/messaging/handling/HelixTaskExecutor.java
index ff970c2..11badac 100644
--- a/helix-core/src/main/java/org/apache/helix/messaging/handling/HelixTaskExecutor.java
+++ b/helix-core/src/main/java/org/apache/helix/messaging/handling/HelixTaskExecutor.java
@@ -187,6 +187,7 @@
_timer = new Timer("HelixTaskExecutor_Timer", true);
_isShuttingDown = false;
+ _liveInstanceStatus = LiveInstanceStatus.NORMAL;
startMonitorThread();
}
@@ -1370,7 +1371,7 @@
manager.getStateMachineEngine(), false);
}
_freezeSessionId = null;
- _liveInstanceStatus = null;
+ _liveInstanceStatus = toStatus;
success = accessor.getBaseDataAccessor().update(path, record -> {
// Remove the status field for backwards compatibility
record.getSimpleFields().remove(LiveInstance.LiveInstanceProperty.STATUS.name());
diff --git a/helix-core/src/main/java/org/apache/helix/model/LiveInstance.java b/helix-core/src/main/java/org/apache/helix/model/LiveInstance.java
index fed2940..e711fea 100644
--- a/helix-core/src/main/java/org/apache/helix/model/LiveInstance.java
+++ b/helix-core/src/main/java/org/apache/helix/model/LiveInstance.java
@@ -142,10 +142,12 @@
}
/**
- * Gets the live instance's status. Returns null if the status field is not set.
+ * Gets the live instance's status. Returns {@link LiveInstanceStatus#NORMAL} if the status
+ * field is not set.
*/
public LiveInstanceStatus getStatus() {
- return _record.getEnumField(LiveInstanceProperty.STATUS.name(), LiveInstanceStatus.class, null);
+ return _record.getEnumField(LiveInstanceProperty.STATUS.name(), LiveInstanceStatus.class,
+ LiveInstanceStatus.NORMAL);
}
/**
diff --git a/helix-core/src/test/java/org/apache/helix/controller/stages/TestManagementMessageGeneration.java b/helix-core/src/test/java/org/apache/helix/controller/stages/TestManagementMessageGeneration.java
index c16b503..7c20b02 100644
--- a/helix-core/src/test/java/org/apache/helix/controller/stages/TestManagementMessageGeneration.java
+++ b/helix-core/src/test/java/org/apache/helix/controller/stages/TestManagementMessageGeneration.java
@@ -128,7 +128,7 @@
RebalanceUtil.buildBestPossibleState(resourceMap.keySet(), currentStateOutput);
// Process the event
- ClusterManagementMode mode = new ClusterManagementMode(ClusterManagementMode.Type.CLUSTER_PAUSE,
+ ClusterManagementMode mode = new ClusterManagementMode(ClusterManagementMode.Type.CLUSTER_FREEZE,
ClusterManagementMode.Status.IN_PROGRESS);
event.addAttribute(AttributeName.BEST_POSSIBLE_STATE.name(), bestPossibleStateOutput);
event.addAttribute(AttributeName.CLUSTER_STATUS.name(), mode);
diff --git a/helix-core/src/test/java/org/apache/helix/controller/stages/TestManagementModeStage.java b/helix-core/src/test/java/org/apache/helix/controller/stages/TestManagementModeStage.java
index 2737bf7..1be5c6b 100644
--- a/helix-core/src/test/java/org/apache/helix/controller/stages/TestManagementModeStage.java
+++ b/helix-core/src/test/java/org/apache/helix/controller/stages/TestManagementModeStage.java
@@ -76,7 +76,7 @@
// Freeze cluster
ClusterManagementModeRequest request = ClusterManagementModeRequest.newBuilder()
.withClusterName(_clusterName)
- .withMode(ClusterManagementMode.Type.CLUSTER_PAUSE)
+ .withMode(ClusterManagementMode.Type.CLUSTER_FREEZE)
.withReason("test")
.build();
_gSetupTool.getClusterManagementTool().setClusterManagementMode(request);
@@ -91,7 +91,7 @@
// In frozen mode
ClusterStatus clusterStatus = _accessor.getProperty(_accessor.keyBuilder().clusterStatus());
- Assert.assertEquals(clusterStatus.getManagementMode(), ClusterManagementMode.Type.CLUSTER_PAUSE);
+ Assert.assertEquals(clusterStatus.getManagementMode(), ClusterManagementMode.Type.CLUSTER_FREEZE);
// Mark a live instance to be pause state
diff --git a/helix-core/src/test/java/org/apache/helix/integration/controller/TestClusterFreezeMode.java b/helix-core/src/test/java/org/apache/helix/integration/controller/TestClusterFreezeMode.java
index cbc8535..afaea91 100644
--- a/helix-core/src/test/java/org/apache/helix/integration/controller/TestClusterFreezeMode.java
+++ b/helix-core/src/test/java/org/apache/helix/integration/controller/TestClusterFreezeMode.java
@@ -153,7 +153,7 @@
// Freeze cluster
ClusterManagementModeRequest request = ClusterManagementModeRequest.newBuilder()
.withClusterName(_clusterName)
- .withMode(ClusterManagementMode.Type.CLUSTER_PAUSE)
+ .withMode(ClusterManagementMode.Type.CLUSTER_FREEZE)
.withReason(methodName)
.build();
_gSetupTool.getClusterManagementTool().setClusterManagementMode(request);
@@ -164,7 +164,7 @@
// Cluster is in progress to cluster pause because there is a pending state transition message
ClusterStatus expectedClusterStatus = new ClusterStatus();
- expectedClusterStatus.setManagementMode(ClusterManagementMode.Type.CLUSTER_PAUSE);
+ expectedClusterStatus.setManagementMode(ClusterManagementMode.Type.CLUSTER_FREEZE);
expectedClusterStatus.setManagementModeStatus(ClusterManagementMode.Status.IN_PROGRESS);
verifyClusterStatus(expectedClusterStatus);
@@ -175,7 +175,7 @@
verifyLiveInstanceStatus(_participants, LiveInstance.LiveInstanceStatus.PAUSED);
expectedClusterStatus = new ClusterStatus();
- expectedClusterStatus.setManagementMode(ClusterManagementMode.Type.CLUSTER_PAUSE);
+ expectedClusterStatus.setManagementMode(ClusterManagementMode.Type.CLUSTER_FREEZE);
expectedClusterStatus.setManagementModeStatus(ClusterManagementMode.Status.COMPLETED);
verifyClusterStatus(expectedClusterStatus);
@@ -187,7 +187,7 @@
return false;
}
String lastHistory = managementHistory.get(managementHistory.size() - 1);
- return lastHistory.contains("MODE=" + ClusterManagementMode.Type.CLUSTER_PAUSE)
+ return lastHistory.contains("MODE=" + ClusterManagementMode.Type.CLUSTER_FREEZE)
&& lastHistory.contains("STATUS=" + ClusterManagementMode.Status.COMPLETED)
&& lastHistory.contains("REASON=" + methodName);
}, TestHelper.WAIT_DURATION));
@@ -303,7 +303,7 @@
.build();
_gSetupTool.getClusterManagementTool().setClusterManagementMode(request);
- verifyLiveInstanceStatus(_participants, null);
+ verifyLiveInstanceStatus(_participants, LiveInstance.LiveInstanceStatus.NORMAL);
ClusterStatus expectedClusterStatus = new ClusterStatus();
expectedClusterStatus.setManagementMode(ClusterManagementMode.Type.NORMAL);
diff --git a/helix-core/src/test/java/org/apache/helix/integration/paticipant/TestParticipantFreeze.java b/helix-core/src/test/java/org/apache/helix/integration/paticipant/TestParticipantFreeze.java
index fedee37..1eb760e 100644
--- a/helix-core/src/test/java/org/apache/helix/integration/paticipant/TestParticipantFreeze.java
+++ b/helix-core/src/test/java/org/apache/helix/integration/paticipant/TestParticipantFreeze.java
@@ -117,7 +117,9 @@
@Test
public void testNormalLiveInstanceStatus() {
LiveInstance liveInstance = _accessor.getProperty(_keyBuilder.liveInstance(_instanceName));
- Assert.assertNull(liveInstance.getStatus());
+ Assert.assertEquals(liveInstance.getStatus(), LiveInstance.LiveInstanceStatus.NORMAL);
+ Assert.assertNull(
+ liveInstance.getRecord().getSimpleField(LiveInstance.LiveInstanceProperty.STATUS.name()));
}
@Test(dependsOnMethods = "testNormalLiveInstanceStatus")
@@ -151,7 +153,7 @@
// New live instance ephemeral node
Assert.assertEquals(liveInstance.getEphemeralOwner(), _participants[1].getSessionId());
// Status is not frozen because controller is not running, no freeze message sent.
- verifyLiveInstanceStatus(_participants[1], null);
+ verifyLiveInstanceStatus(_participants[1], LiveInstance.LiveInstanceStatus.NORMAL);
// Old session current state is deleted because of current state carry-over
Assert.assertTrue(TestHelper.verify(
@@ -221,7 +223,7 @@
// Live instance status is NORMAL, but set to null value in both memory and zk.
// After live instance status is updated, the process is completed.
- verifyLiveInstanceStatus(_participants[0], null);
+ verifyLiveInstanceStatus(_participants[0], LiveInstance.LiveInstanceStatus.NORMAL);
// Unfreeze message is correctly deleted
Assert.assertNull(
_accessor.getProperty(_keyBuilder.message(_instanceName, unfreezeMessage.getId())));
@@ -242,7 +244,7 @@
private void verifyLiveInstanceStatus(MockParticipantManager participant,
LiveInstance.LiveInstanceStatus status) throws Exception {
- // Live instance status is frozen in both memory and zk
+ // Verify live instance status in both memory and zk
Assert.assertTrue(TestHelper.verify(() -> {
LiveInstance.LiveInstanceStatus inMemoryLiveInstanceStatus =
((DefaultMessagingService) participant.getMessagingService()).getExecutor()
@@ -285,7 +287,8 @@
// Live instance status is frozen in both memory and zk
verifyLiveInstanceStatus(participant, LiveInstance.LiveInstanceStatus.PAUSED);
// Freeze message is correctly deleted
- Assert.assertNull(_accessor
- .getProperty(_keyBuilder.message(participant.getInstanceName(), freezeMessage.getId())));
+ Assert.assertTrue(TestHelper.verify(() -> !_gZkClient.exists(
+ _keyBuilder.message(participant.getInstanceName(), freezeMessage.getId()).getPath()),
+ TestHelper.WAIT_DURATION));
}
}
diff --git a/helix-core/src/test/java/org/apache/helix/manager/zk/TestZkHelixAdmin.java b/helix-core/src/test/java/org/apache/helix/manager/zk/TestZkHelixAdmin.java
index 7ddf677..968a5d4 100644
--- a/helix-core/src/test/java/org/apache/helix/manager/zk/TestZkHelixAdmin.java
+++ b/helix-core/src/test/java/org/apache/helix/manager/zk/TestZkHelixAdmin.java
@@ -1087,7 +1087,7 @@
try {
ClusterManagementModeRequest request = ClusterManagementModeRequest.newBuilder()
.withClusterName(clusterName)
- .withMode(ClusterManagementMode.Type.CLUSTER_PAUSE)
+ .withMode(ClusterManagementMode.Type.CLUSTER_FREEZE)
.withCancelPendingST(true)
.withReason(methodName)
.build();
@@ -1100,7 +1100,7 @@
ClusterManagementModeRequest request = ClusterManagementModeRequest.newBuilder()
.withClusterName(clusterName)
- .withMode(ClusterManagementMode.Type.CLUSTER_PAUSE)
+ .withMode(ClusterManagementMode.Type.CLUSTER_FREEZE)
.withReason(methodName)
.build();
_gSetupTool.getClusterManagementTool().setClusterManagementMode(request);
diff --git a/helix-core/src/test/java/org/apache/helix/model/TestControllerHistoryModel.java b/helix-core/src/test/java/org/apache/helix/model/TestControllerHistoryModel.java
index 2e0bab3..d77287a 100644
--- a/helix-core/src/test/java/org/apache/helix/model/TestControllerHistoryModel.java
+++ b/helix-core/src/test/java/org/apache/helix/model/TestControllerHistoryModel.java
@@ -37,7 +37,7 @@
public void testManagementModeHistory() {
ControllerHistory controllerHistory = new ControllerHistory("HISTORY");
String controller = "controller-0";
- ClusterManagementMode mode = new ClusterManagementMode(ClusterManagementMode.Type.CLUSTER_PAUSE,
+ ClusterManagementMode mode = new ClusterManagementMode(ClusterManagementMode.Type.CLUSTER_FREEZE,
ClusterManagementMode.Status.COMPLETED);
long time = System.currentTimeMillis();
String fromHost = NetworkUtil.getLocalhostName();
diff --git a/helix-core/src/test/java/org/apache/helix/model/TestLiveInstance.java b/helix-core/src/test/java/org/apache/helix/model/TestLiveInstance.java
index 18285f5..b8024bb 100644
--- a/helix-core/src/test/java/org/apache/helix/model/TestLiveInstance.java
+++ b/helix-core/src/test/java/org/apache/helix/model/TestLiveInstance.java
@@ -150,7 +150,9 @@
@Test
public void testLiveInstanceStatus() {
LiveInstance testLiveInstance = new LiveInstance("testLiveInstanceStatus");
- Assert.assertNull(testLiveInstance.getStatus());
+ Assert.assertNull(testLiveInstance.getRecord()
+ .getSimpleField(LiveInstance.LiveInstanceProperty.STATUS.name()));
+ Assert.assertEquals(testLiveInstance.getStatus(), LiveInstance.LiveInstanceStatus.NORMAL);
testLiveInstance.setStatus(LiveInstance.LiveInstanceStatus.PAUSED);
Assert.assertEquals(testLiveInstance.getStatus(), LiveInstance.LiveInstanceStatus.PAUSED);
}
diff --git a/helix-rest/src/main/java/org/apache/helix/rest/server/resources/helix/ClusterAccessor.java b/helix-rest/src/main/java/org/apache/helix/rest/server/resources/helix/ClusterAccessor.java
index 9efcdc7..552a530 100644
--- a/helix-rest/src/main/java/org/apache/helix/rest/server/resources/helix/ClusterAccessor.java
+++ b/helix-rest/src/main/java/org/apache/helix/rest/server/resources/helix/ClusterAccessor.java
@@ -344,7 +344,7 @@
List<LiveInstance> liveInstances = accessor.getChildValues(keyBuilder.liveInstances());
BaseDataAccessor<ZNRecord> baseAccessor = accessor.getBaseDataAccessor();
- if (ClusterManagementMode.Type.CLUSTER_PAUSE.equals(mode.getMode())) {
+ if (ClusterManagementMode.Type.CLUSTER_FREEZE.equals(mode.getMode())) {
// Entering cluster freeze mode, check live instance freeze status and pending ST
for (LiveInstance liveInstance : liveInstances) {
String instanceName = liveInstance.getInstanceName();
@@ -370,7 +370,7 @@
participantDetails.put("status", status.name());
participantDetails.put("liveInstancesInProgress", liveInstancesInProgress);
- if (ClusterManagementMode.Type.CLUSTER_PAUSE.equals(mode.getMode())) {
+ if (ClusterManagementMode.Type.CLUSTER_FREEZE.equals(mode.getMode())) {
// Add pending ST result for cluster freeze mode
participantDetails.put("hasPendingStateTransition", hasPendingST);
}
diff --git a/helix-rest/src/test/java/org/apache/helix/rest/server/TestClusterAccessor.java b/helix-rest/src/test/java/org/apache/helix/rest/server/TestClusterAccessor.java
index 8d2de70..8b8f38a 100644
--- a/helix-rest/src/test/java/org/apache/helix/rest/server/TestClusterAccessor.java
+++ b/helix-rest/src/test/java/org/apache/helix/rest/server/TestClusterAccessor.java
@@ -1302,7 +1302,7 @@
// Set cluster pause mode
ClusterManagementModeRequest request = ClusterManagementModeRequest.newBuilder()
- .withMode(ClusterManagementMode.Type.CLUSTER_PAUSE)
+ .withMode(ClusterManagementMode.Type.CLUSTER_FREEZE)
.withClusterName(cluster)
.build();
String payload = OBJECT_MAPPER.writeValueAsString(request);
@@ -1321,22 +1321,23 @@
// Verify get cluster status
String body = get(endpoint, null, Response.Status.OK.getStatusCode(), true);
- ClusterManagementMode mode =
- OBJECT_MAPPER.readerFor(ClusterManagementMode.class).readValue(body);
- Assert.assertEquals(mode.getMode(), ClusterManagementMode.Type.CLUSTER_PAUSE);
+ Map<String, Object> responseMap = OBJECT_MAPPER.readerFor(Map.class).readValue(body);
+ Assert.assertEquals(responseMap.get("mode"), ClusterManagementMode.Type.CLUSTER_FREEZE.name());
// Depending on timing, it could IN_PROGRESS or COMPLETED.
// It's just to verify the rest response format is correct
- Assert.assertTrue(ClusterManagementMode.Status.IN_PROGRESS.equals(mode.getStatus())
- || ClusterManagementMode.Status.COMPLETED.equals(mode.getStatus()));
+ String status = (String) responseMap.get("status");
+ Assert.assertTrue(ClusterManagementMode.Status.IN_PROGRESS.name().equals(status)
+ || ClusterManagementMode.Status.COMPLETED.name().equals(status));
body = get(endpoint, ImmutableMap.of("showDetails", "true"), Response.Status.OK.getStatusCode(),
true);
- Map<String, Object> responseMap = OBJECT_MAPPER.readerFor(Map.class).readValue(body);
+ responseMap = OBJECT_MAPPER.readerFor(Map.class).readValue(body);
Map<String, Object> detailsMap = (Map<String, Object>) responseMap.get("details");
+ status = (String) responseMap.get("status");
Assert.assertEquals(responseMap.get("cluster"), cluster);
- Assert.assertEquals(responseMap.get("mode"), mode.getMode().name());
- Assert.assertEquals(responseMap.get("status"), mode.getStatus().name());
+ Assert.assertEquals(responseMap.get("mode"), ClusterManagementMode.Type.CLUSTER_FREEZE.name());
+ Assert.assertEquals(responseMap.get("status"), status);
Assert.assertTrue(responseMap.containsKey("details"));
Assert.assertTrue(detailsMap.containsKey("cluster"));
Assert.assertTrue(detailsMap.containsKey("liveInstances"));
