krishan1390 commented on code in PR #16857:
URL: https://github.com/apache/pinot/pull/16857#discussion_r2400759991
##########
pinot-controller/src/main/java/org/apache/pinot/controller/helix/core/minion/PinotTaskManager.java:
##########
@@ -239,40 +254,79 @@ public Map<String, String> createTask(String taskType,
String tableName, @Nullab
// Example usage in BaseTaskGenerator.getNumSubTasks()
taskConfigs.put(MinionConstants.TRIGGERED_BY,
CommonConstants.TaskTriggers.ADHOC_TRIGGER.name());
- List<PinotTaskConfig> pinotTaskConfigs =
taskGenerator.generateTasks(tableConfig, taskConfigs);
- if (pinotTaskConfigs.isEmpty()) {
- LOGGER.warn("No ad-hoc task generated for task type: {}", taskType);
- continue;
+ // Acquire distributed lock before proceeding with ad-hoc task generation
+ // Need locking to protect against:
+ // 1. Race conditions with periodic task generation
+ // 2. Multiple simultaneous ad-hoc requests
+ // 3. Leadership changes during task generation
+ DistributedTaskLockManager.TaskLock lock = null;
+ boolean taskCreationSuccess = true;
+ if (_distributedTaskLockManager != null) {
+ lock = _distributedTaskLockManager.acquireLock(tableNameWithType);
+ if (lock == null) {
+ String message = String.format("Could not acquire table level
distributed lock for ad-hoc task type: %s, "
+ + "table: %s. Another controller is likely generating tasks
for this type. Please try again later.",
+ taskType, tableNameWithType);
+ LOGGER.warn(message);
+ throw new RuntimeException(message);
+ }
+ LOGGER.info("Acquired table level distributed lock for ad-hoc task
type: {} on table: {}", taskType,
+ tableNameWithType);
}
- int maxNumberOfSubTasks = taskGenerator.getMaxAllowedSubTasksPerTask();
- if (pinotTaskConfigs.size() > maxNumberOfSubTasks) {
- String message = String.format(
- "Number of tasks generated for task type: %s for table: %s is %d,
which is greater than the "
- + "maximum number of tasks to schedule: %d. This is "
- + "controlled by the cluster config %s which is set based on
controller's performance.", taskType,
- tableName, pinotTaskConfigs.size(), maxNumberOfSubTasks,
MinionConstants.MAX_ALLOWED_SUB_TASKS_KEY);
- message += "Optimise the task config or reduce tableMaxNumTasks to
avoid the error";
- // We throw an exception to notify the user
- // This is to ensure that the user is aware of the task generation
limit
- throw new RuntimeException(message);
+
+ try {
+ List<PinotTaskConfig> pinotTaskConfigs =
taskGenerator.generateTasks(tableConfig, taskConfigs);
+ if (pinotTaskConfigs.isEmpty()) {
+ LOGGER.warn("No ad-hoc task generated for task type: {}, for table:
{}", taskType, tableNameWithType);
+ continue;
+ }
+ int maxNumberOfSubTasks = taskGenerator.getMaxAllowedSubTasksPerTask();
+ if (pinotTaskConfigs.size() > maxNumberOfSubTasks) {
+ String message = String.format(
+ "Number of tasks generated for task type: %s for table: %s is
%d, which is greater than the "
+ + "maximum number of tasks to schedule: %d. This is
controlled by the cluster config %s which is set "
+ + "based on controller's performance.", taskType,
tableNameWithType, pinotTaskConfigs.size(),
+ maxNumberOfSubTasks, MinionConstants.MAX_ALLOWED_SUB_TASKS_KEY);
+ message += "Optimise the task config or reduce tableMaxNumTasks to
avoid the error";
+ // We throw an exception to notify the user
+ // This is to ensure that the user is aware of the task generation
limit
+ taskCreationSuccess = false;
+ throw new RuntimeException(message);
+ }
+ pinotTaskConfigs.forEach(pinotTaskConfig ->
pinotTaskConfig.getConfigs()
+ .computeIfAbsent(MinionConstants.TRIGGERED_BY, k ->
CommonConstants.TaskTriggers.ADHOC_TRIGGER.name()));
+ addDefaultsToTaskConfig(pinotTaskConfigs);
+ LOGGER.info("Submitting ad-hoc task for task type: {} with task
configs: {}", taskType, pinotTaskConfigs);
+ _controllerMetrics.addMeteredTableValue(taskType,
ControllerMeter.NUMBER_ADHOC_TASKS_SUBMITTED, 1);
+ responseMap.put(tableNameWithType,
+ _helixTaskResourceManager.submitTask(parentTaskName,
pinotTaskConfigs, minionInstanceTag,
+ taskGenerator.getTaskTimeoutMs(minionInstanceTag),
+
taskGenerator.getNumConcurrentTasksPerInstance(minionInstanceTag),
+ taskGenerator.getMaxAttemptsPerTask(minionInstanceTag)));
+ } finally {
+ if (!responseMap.containsKey(tableNameWithType)) {
+ LOGGER.warn("No task submitted for tableNameWithType: {}",
tableNameWithType);
+ taskCreationSuccess = false;
+ }
+ if (lock != null) {
+ _distributedTaskLockManager.releaseLock(lock, taskCreationSuccess);
Review Comment:
why do we require taskCreationSuccess ? I don't see any use of it in
releaseLock() other than logging too. so if its just logging, maybe we log it
here rather than add it in releaseLock() ?
##########
pinot-controller/src/main/java/org/apache/pinot/controller/helix/core/minion/PinotTaskManager.java:
##########
@@ -210,6 +224,7 @@ public Map<String, String> createTask(String taskType,
String tableName, @Nullab
if (tableNameWithTypes.isEmpty()) {
throw new TableNotFoundException("'tableName' " + tableName + " is not
found");
}
+ LOGGER.info("Generating tasks for {} tables, list: {}",
tableNameWithTypes.size(), tableNameWithTypes);
Review Comment:
looks like createTask() isn't synchronized. we should synchronize this to
reduce failures while acquiring lock if the flow is happening within the leader
controller.
synchronization is fine because its already being done for the other method
scheduleTasks() which is more frequently called.
##########
pinot-controller/src/main/java/org/apache/pinot/controller/helix/core/minion/PinotTaskManager.java:
##########
@@ -239,40 +254,79 @@ public Map<String, String> createTask(String taskType,
String tableName, @Nullab
// Example usage in BaseTaskGenerator.getNumSubTasks()
taskConfigs.put(MinionConstants.TRIGGERED_BY,
CommonConstants.TaskTriggers.ADHOC_TRIGGER.name());
- List<PinotTaskConfig> pinotTaskConfigs =
taskGenerator.generateTasks(tableConfig, taskConfigs);
- if (pinotTaskConfigs.isEmpty()) {
- LOGGER.warn("No ad-hoc task generated for task type: {}", taskType);
- continue;
+ // Acquire distributed lock before proceeding with ad-hoc task generation
+ // Need locking to protect against:
+ // 1. Race conditions with periodic task generation
+ // 2. Multiple simultaneous ad-hoc requests
+ // 3. Leadership changes during task generation
+ DistributedTaskLockManager.TaskLock lock = null;
+ boolean taskCreationSuccess = true;
+ if (_distributedTaskLockManager != null) {
+ lock = _distributedTaskLockManager.acquireLock(tableNameWithType);
+ if (lock == null) {
+ String message = String.format("Could not acquire table level
distributed lock for ad-hoc task type: %s, "
+ + "table: %s. Another controller is likely generating tasks
for this type. Please try again later.",
+ taskType, tableNameWithType);
+ LOGGER.warn(message);
+ throw new RuntimeException(message);
+ }
+ LOGGER.info("Acquired table level distributed lock for ad-hoc task
type: {} on table: {}", taskType,
+ tableNameWithType);
}
- int maxNumberOfSubTasks = taskGenerator.getMaxAllowedSubTasksPerTask();
- if (pinotTaskConfigs.size() > maxNumberOfSubTasks) {
- String message = String.format(
- "Number of tasks generated for task type: %s for table: %s is %d,
which is greater than the "
- + "maximum number of tasks to schedule: %d. This is "
- + "controlled by the cluster config %s which is set based on
controller's performance.", taskType,
- tableName, pinotTaskConfigs.size(), maxNumberOfSubTasks,
MinionConstants.MAX_ALLOWED_SUB_TASKS_KEY);
- message += "Optimise the task config or reduce tableMaxNumTasks to
avoid the error";
- // We throw an exception to notify the user
- // This is to ensure that the user is aware of the task generation
limit
- throw new RuntimeException(message);
+
+ try {
+ List<PinotTaskConfig> pinotTaskConfigs =
taskGenerator.generateTasks(tableConfig, taskConfigs);
+ if (pinotTaskConfigs.isEmpty()) {
+ LOGGER.warn("No ad-hoc task generated for task type: {}, for table:
{}", taskType, tableNameWithType);
+ continue;
+ }
+ int maxNumberOfSubTasks = taskGenerator.getMaxAllowedSubTasksPerTask();
+ if (pinotTaskConfigs.size() > maxNumberOfSubTasks) {
+ String message = String.format(
+ "Number of tasks generated for task type: %s for table: %s is
%d, which is greater than the "
+ + "maximum number of tasks to schedule: %d. This is
controlled by the cluster config %s which is set "
+ + "based on controller's performance.", taskType,
tableNameWithType, pinotTaskConfigs.size(),
+ maxNumberOfSubTasks, MinionConstants.MAX_ALLOWED_SUB_TASKS_KEY);
+ message += "Optimise the task config or reduce tableMaxNumTasks to
avoid the error";
+ // We throw an exception to notify the user
+ // This is to ensure that the user is aware of the task generation
limit
+ taskCreationSuccess = false;
+ throw new RuntimeException(message);
+ }
+ pinotTaskConfigs.forEach(pinotTaskConfig ->
pinotTaskConfig.getConfigs()
+ .computeIfAbsent(MinionConstants.TRIGGERED_BY, k ->
CommonConstants.TaskTriggers.ADHOC_TRIGGER.name()));
+ addDefaultsToTaskConfig(pinotTaskConfigs);
+ LOGGER.info("Submitting ad-hoc task for task type: {} with task
configs: {}", taskType, pinotTaskConfigs);
+ _controllerMetrics.addMeteredTableValue(taskType,
ControllerMeter.NUMBER_ADHOC_TASKS_SUBMITTED, 1);
+ responseMap.put(tableNameWithType,
+ _helixTaskResourceManager.submitTask(parentTaskName,
pinotTaskConfigs, minionInstanceTag,
+ taskGenerator.getTaskTimeoutMs(minionInstanceTag),
+
taskGenerator.getNumConcurrentTasksPerInstance(minionInstanceTag),
+ taskGenerator.getMaxAttemptsPerTask(minionInstanceTag)));
+ } finally {
+ if (!responseMap.containsKey(tableNameWithType)) {
+ LOGGER.warn("No task submitted for tableNameWithType: {}",
tableNameWithType);
+ taskCreationSuccess = false;
+ }
+ if (lock != null) {
+ _distributedTaskLockManager.releaseLock(lock, taskCreationSuccess);
Review Comment:
same question applicable for taskGenerationSuccesses in scheduleTasks()
##########
pinot-controller/src/main/java/org/apache/pinot/controller/helix/core/minion/DistributedTaskLockManager.java:
##########
@@ -0,0 +1,563 @@
+/**
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+package org.apache.pinot.controller.helix.core.minion;
+
+import com.google.common.annotations.VisibleForTesting;
+import java.util.ArrayList;
+import java.util.Comparator;
+import java.util.List;
+import java.util.UUID;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
+import javax.annotation.Nullable;
+import org.apache.helix.AccessOption;
+import org.apache.helix.store.zk.ZkHelixPropertyStore;
+import org.apache.helix.zookeeper.datamodel.ZNRecord;
+import org.apache.pinot.common.metadata.ZKMetadataProvider;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+
+/**
+ * Manages distributed locks for minion task generation using ZooKeeper
ephemeral sequential nodes.
+ * Uses ephemeral nodes that automatically disappear when the controller
session ends.
+ * This approach provides automatic cleanup and is suitable for long-running
task generation.
+ * Locks are held until explicitly released or the controller session
terminates.
+ * Locks are at the table level, to ensure that only one type of task can be
generated per table at any given time.
+ */
+public class DistributedTaskLockManager {
Review Comment:
>to handle scenarios where multiple controllers / threads within the same
controller do the check, find no lock, and proceed to try to create the lock.
I think the _propertyStore.create() API avoids this problem. only 1 thread
will be able to create the node with the unique path.
https://gemini.google.com/gem/coding-partner/6e4225b9746e0363
And if this can work, this will simplify this code further ?
##########
pinot-controller/src/main/java/org/apache/pinot/controller/helix/core/minion/DistributedTaskLockManager.java:
##########
@@ -0,0 +1,423 @@
+/**
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+package org.apache.pinot.controller.helix.core.minion;
+
+import com.google.common.annotations.VisibleForTesting;
+import java.util.ArrayList;
+import java.util.Comparator;
+import java.util.List;
+import java.util.UUID;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
+import javax.annotation.Nullable;
+import org.apache.helix.AccessOption;
+import org.apache.helix.store.zk.ZkHelixPropertyStore;
+import org.apache.helix.zookeeper.datamodel.ZNRecord;
+import org.apache.pinot.common.metadata.ZKMetadataProvider;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+
+/**
+ * Manages distributed locks for minion task generation using ZooKeeper
ephemeral sequential nodes.
+ * Uses ephemeral nodes that automatically disappear when the controller
session ends.
+ * This approach provides automatic cleanup and is suitable for long-running
task generation.
+ * Locks are held until explicitly released or the controller session
terminates.
+ * Locks are at the table level, to ensure that only one type of task can be
generated per table at any given time.
+ * <p>
+ * ZK EPHEMERAL_SEQUENTIAL Locks (see <a
href="https://zookeeper.apache.org/doc/current/recipes.html#sc_recipes_Locks";>
+ * ZooKeeper Lock Recipe.</a> for more details):
+ * <ul>
+ * <li>Every lock is created with a lock prefix. Lock prefix used:
[controllerName]-lock-[UUID]. The UUID helps
+ * differentiate between requests originating from the same controller at
the same time
+ * <li>When ZK creates the ZNode, it appends a sequence number at the end.
E.g.
+ * [controllerName]-lock-[UUID]-00000001
+ * <li>The sequence number is used to identify the lock winner in case
more than one lock node is created at the
+ * same time. The smallest sequence number always wins
+ * <li>The locks are EPHEMERAL in nature, meaning that once the session
with ZK is lost, the lock is automatically
+ * cleaned up. Scenarios when the ZK session can be lost: a) controller
shutdown, b) controller crash, c) ZK session
+ * expiry (e.g. long GC pauses can cause this)
+ * <li>This implementation does not set up watches as described in the
recipe as the task lock is released whenever
+ * we identify that the lock is already acquired. Do not expect lock
ownership to automatically change for the
+ * time being. If such support is needed in the future, this can be
enhanced to add a watch on the neighboring
+ * lock node
+ * </ul>
+ * <p>
+ * Example of how the locks will work:
+ * <p>
+ * Say we have two controllers, and one controller happens to run 2 threads at
the same time, all of which need to take
+ * the distributed lock. Each thread will create a distributed lock node, and
the "-Lock" ZNode getChildren will return:
+ * <ul>
+ * <li>controller2-lock-xyzwx-00000002
+ * <li>controller1-lock-abcde-00000001
+ * <li>controller1-lock-ab345-00000003
+ * </ul>
+ * <p>
+ * In the above, the controller1 with UUID abcde will win the lock as it has
the smallest sequence number. The other
+ * two threads will clean up their locks and return error that the distributed
lock could not be acquired. Controller1
Review Comment:
hypothetically, what happens if the cleanup is missed due to a bug in the
code ? I am assuming controller crash will cleanup the ephemeral node so
crashes shouldn't be a problem.
but if due to a bug, we don't cleanup the lock with the sequence number
00000003, I guess no one will be able to acquire the lock in future right ?
##########
pinot-controller/src/main/java/org/apache/pinot/controller/helix/core/minion/DistributedTaskLockManager.java:
##########
@@ -0,0 +1,423 @@
+/**
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+package org.apache.pinot.controller.helix.core.minion;
+
+import com.google.common.annotations.VisibleForTesting;
+import java.util.ArrayList;
+import java.util.Comparator;
+import java.util.List;
+import java.util.UUID;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
+import javax.annotation.Nullable;
+import org.apache.helix.AccessOption;
+import org.apache.helix.store.zk.ZkHelixPropertyStore;
+import org.apache.helix.zookeeper.datamodel.ZNRecord;
+import org.apache.pinot.common.metadata.ZKMetadataProvider;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+
+/**
+ * Manages distributed locks for minion task generation using ZooKeeper
ephemeral sequential nodes.
+ * Uses ephemeral nodes that automatically disappear when the controller
session ends.
+ * This approach provides automatic cleanup and is suitable for long-running
task generation.
+ * Locks are held until explicitly released or the controller session
terminates.
+ * Locks are at the table level, to ensure that only one type of task can be
generated per table at any given time.
+ * <p>
+ * ZK EPHEMERAL_SEQUENTIAL Locks (see <a
href="https://zookeeper.apache.org/doc/current/recipes.html#sc_recipes_Locks";>
+ * ZooKeeper Lock Recipe.</a> for more details):
+ * <ul>
+ * <li>Every lock is created with a lock prefix. Lock prefix used:
[controllerName]-lock-[UUID]. The UUID helps
+ * differentiate between requests originating from the same controller at
the same time
+ * <li>When ZK creates the ZNode, it appends a sequence number at the end.
E.g.
+ * [controllerName]-lock-[UUID]-00000001
+ * <li>The sequence number is used to identify the lock winner in case
more than one lock node is created at the
+ * same time. The smallest sequence number always wins
+ * <li>The locks are EPHEMERAL in nature, meaning that once the session
with ZK is lost, the lock is automatically
+ * cleaned up. Scenarios when the ZK session can be lost: a) controller
shutdown, b) controller crash, c) ZK session
+ * expiry (e.g. long GC pauses can cause this)
+ * <li>This implementation does not set up watches as described in the
recipe as the task lock is released whenever
+ * we identify that the lock is already acquired. Do not expect lock
ownership to automatically change for the
+ * time being. If such support is needed in the future, this can be
enhanced to add a watch on the neighboring
+ * lock node
+ * </ul>
+ * <p>
+ * Example of how the locks will work:
+ * <p>
+ * Say we have two controllers, and one controller happens to run 2 threads at
the same time, all of which need to take
+ * the distributed lock. Each thread will create a distributed lock node, and
the "-Lock" ZNode getChildren will return:
+ * <ul>
+ * <li>controller2-lock-xyzwx-00000002
+ * <li>controller1-lock-abcde-00000001
+ * <li>controller1-lock-ab345-00000003
+ * </ul>
+ * <p>
+ * In the above, the controller1 with UUID abcde will win the lock as it has
the smallest sequence number. The other
+ * two threads will clean up their locks and return error that the distributed
lock could not be acquired. Controller1
+ * will proceed with performing its tasks, and when done will release the lock.
+ */
+public class DistributedTaskLockManager {
+ private static final Logger LOGGER =
LoggerFactory.getLogger(DistributedTaskLockManager.class);
+
+ // Lock paths are constructed using ZKMetadataProvider
+ private static final String LOCK_OWNER_KEY = "lockOwner";
+ private static final String LOCK_UUID_KEY = "lockUuid";
+ private static final String LOCK_TIMESTAMP_MILLIS_KEY =
"lockTimestampMillis";
+
+ // Define a custom comparator to compare strings of format
'<controllerName>-lock-<uuid>-<sequenceNumber>' and sort
+ // them by the sequence number at the end
+ private static final Comparator<String> TASK_LOCK_SEQUENCE_ID_COMPARATOR =
(s1, s2) -> {
+ // Regex to find the trailing sequence of digits
+ Pattern p = Pattern.compile("\\d+$");
+
+ // Extract the number from the first string
+ Matcher m1 = p.matcher(s1);
+ long num1 = m1.find() ? Long.parseLong(m1.group()) : 0;
+
+ // Extract the number from the second string
+ Matcher m2 = p.matcher(s2);
+ long num2 = m2.find() ? Long.parseLong(m2.group()) : 0;
+
+ return Long.compare(num1, num2);
+ };
+
+ private final ZkHelixPropertyStore<ZNRecord> _propertyStore;
+ private final String _controllerInstanceId;
+
+ public DistributedTaskLockManager(ZkHelixPropertyStore<ZNRecord>
propertyStore, String controllerInstanceId) {
+ _propertyStore = propertyStore;
+ _controllerInstanceId = controllerInstanceId;
+
+ // Ensure base paths exist
+ ensureBasePaths();
+ }
+
+ /**
+ * Attempts to acquire a distributed lock at the table level for task
generation using session-based locking.
+ * The lock is held until explicitly released or the controller session ends.
+ *
+ * @param tableNameWithType the table name with type
+ * @return TaskLock object if successful, null if lock could not be acquired
+ */
+ @Nullable
+ public TaskLock acquireLock(String tableNameWithType) {
+ String lockBasePath = getLockBasePath(tableNameWithType);
+
+ LOGGER.info("Attempting to acquire task generation lock: {} by controller:
{}", tableNameWithType,
+ _controllerInstanceId);
+
+ try {
+ // Check if task generation is already in progress
+ if (isTaskGenerationInProgress(tableNameWithType, lockBasePath)) {
+ LOGGER.info("Task generation already in progress for: {} by this or
another controller", tableNameWithType);
+ return null;
+ }
+
+ // Try to acquire the lock using ephemeral sequential node
+ TaskLock lock = tryAcquireSessionBasedLock(lockBasePath,
tableNameWithType);
+ if (lock != null) {
+ LOGGER.info("Successfully acquired task generation lock: {} by
controller: {}", tableNameWithType,
+ _controllerInstanceId);
+ return lock;
+ } else {
+ LOGGER.warn("Could not acquire lock: {} - another controller holds
it", tableNameWithType);
+ return null;
+ }
+ } catch (Exception e) {
+ LOGGER.error("Error while trying to acquire lock: {}",
tableNameWithType, e);
+ return null;
+ }
+ }
+
+ private String getLockBasePath(String tableNameForPath) {
+ return
ZKMetadataProvider.constructPropertyStorePathForMinionTaskGenerationLock(tableNameForPath);
+ }
+
+ private String getBasePath() {
+ return
ZKMetadataProvider.getPropertyStorePathForMinionTaskMetadataPrefix();
+ }
+
+ /**
+ * Releases a lock assuming successful completion.
+ */
+ public boolean releaseLock(TaskLock lock) {
+ return releaseLock(lock, true);
+ }
+
+ /**
+ * Releases a previously acquired session-based lock and marks task
generation as completed.
+ *
+ * @param lock the lock to release
+ * @param success whether task generation completed successfully
+ * @return true if successfully released, false otherwise
+ */
+ public boolean releaseLock(TaskLock lock, boolean success) {
+ if (lock == null) {
+ return true;
+ }
+
+ String lockKey = lock.getLockKey();
+
+ try {
+ // Remove the ephemeral lock node
+ if (lock.getLockNodePath() != null) {
+ try {
+ boolean status = _propertyStore.remove(lock.getLockNodePath(),
AccessOption.EPHEMERAL);
+ LOGGER.info("Removed ephemeral lock node: {}, removal success: {}",
lock.getLockNodePath(), status);
+ } catch (Exception e) {
+ // Lock node might have already been removed due to session timeout
- this is OK
+ LOGGER.warn("Ephemeral lock node already removed or session expired:
{}", lock.getLockNodePath(), e);
+ }
+ }
+
+ LOGGER.info("Successfully released task generation lock: {} by
controller: {} (success: {})", lockKey,
+ _controllerInstanceId, success);
+ return true;
+ } catch (Exception e) {
+ LOGGER.error("Error while releasing lock: {}", lockKey, e);
+ return false;
+ }
+ }
+
+ /**
+ * Force release the lock without checking if any tasks are in progress
+ */
+ public boolean forceReleaseLock(String tableNameWithType) {
+ LOGGER.info("Trying to force release the lock for table: {}",
tableNameWithType);
+ String lockBasePath = getLockBasePath(tableNameWithType);
+
+ boolean released = true;
+ if (_propertyStore.exists(lockBasePath, AccessOption.PERSISTENT)) {
+ List<String> lockNodes = _propertyStore.getChildNames(lockBasePath,
AccessOption.PERSISTENT);
+ if (lockNodes != null && !lockNodes.isEmpty()) {
+ // There are active ephemeral lock nodes, check if any are still valid
and delete them
+ for (String nodeName : lockNodes) {
+ String nodePath = lockBasePath + "/" + nodeName;
+ if (_propertyStore.exists(nodePath, AccessOption.EPHEMERAL)) {
+ LOGGER.info("Lock for table: {} found at path: {}, trying to
remove", tableNameWithType, nodePath);
+ boolean result = _propertyStore.remove(nodePath,
AccessOption.EPHEMERAL);
+ if (!result) {
+ LOGGER.warn("Could not force release lock: {}", nodePath);
+ released = false;
+ }
+ }
+ }
+ } else {
+ LOGGER.info("No locks to force release, no child lock ZNodes found for
table: {} under base: {}",
+ tableNameWithType, lockBasePath);
+ }
+ } else {
+ LOGGER.info("No locks to force release, no base lock ZNode: {} found for
table: {}", lockBasePath,
+ tableNameWithType);
+ }
+ return released;
+ }
+
+ /**
+ * Checks if any task generation is currently in progress for the given
table.
+ *
+ * @param tableNameWithType the table name with type
+ * @return true if task generation is in progress, false otherwise
+ */
+ @VisibleForTesting
+ boolean isTaskGenerationInProgress(String tableNameWithType) {
+ String lockBasePath = getLockBasePath(tableNameWithType);
+ return isTaskGenerationInProgress(tableNameWithType, lockBasePath);
+ }
+
+ /**
+ * Internal method to check if task generation is in progress for a lock key.
+ */
+ private boolean isTaskGenerationInProgress(String tableNameWithType, String
lockBasePath) {
+ try {
+ if (!_propertyStore.exists(lockBasePath, AccessOption.PERSISTENT)) {
+ return false;
+ }
+
+ // Check if there are any active ephemeral lock nodes
+ List<String> lockNodes = _propertyStore.getChildNames(lockBasePath,
AccessOption.PERSISTENT);
+ if (lockNodes != null && !lockNodes.isEmpty()) {
+ // There are active ephemeral lock nodes, check if any are still valid
+ for (String nodeName : lockNodes) {
+ String nodePath = lockBasePath + "/" + nodeName;
+ if (_propertyStore.exists(nodePath, AccessOption.EPHEMERAL)) {
+ // Ephemeral node exists, meaning session is still alive and task
should be in progress
+ return true;
+ }
+ }
+ }
+ return false;
+ } catch (Exception e) {
+ LOGGER.error("Error checking task generation status for: {}",
tableNameWithType, e);
+ return false;
+ }
+ }
+
+ /**
+ * Attempts to acquire a lock using ephemeral sequential nodes.
+ * Uses the ZooKeeper recipe for distributed locking with automatic cleanup.
+ */
+ @VisibleForTesting
+ TaskLock tryAcquireSessionBasedLock(String lockBasePath, String lockKey) {
+ try {
+ long currentTimeMs = System.currentTimeMillis();
+
+ // Ensure the base lock directory exists
+ if (!_propertyStore.exists(lockBasePath, AccessOption.PERSISTENT)) {
+ ZNRecord baseRecord = new ZNRecord(lockKey);
+ _propertyStore.create(lockBasePath, baseRecord,
AccessOption.PERSISTENT);
+ }
+
+ // Create ephemeral sequential node for this controller, add an UUID to
ensure that the path is unique in case
+ // multiple controller threads run at the same time
+ UUID uuid = UUID.randomUUID();
+ String lockNodePrefix = lockBasePath + "/" + _controllerInstanceId + "-"
+ uuid + "-lock-";
+ ZNRecord lockRecord = new ZNRecord(_controllerInstanceId);
+ lockRecord.setSimpleField(LOCK_OWNER_KEY, _controllerInstanceId);
+ lockRecord.setSimpleField(LOCK_TIMESTAMP_MILLIS_KEY,
String.valueOf(currentTimeMs));
+ lockRecord.setSimpleField(LOCK_UUID_KEY, uuid.toString());
+
+ // ZK will assign the sequence when creating EPHEMERAL_SEQUENTIAL ZNodes
+ boolean created = _propertyStore.create(lockNodePrefix, lockRecord,
AccessOption.EPHEMERAL_SEQUENTIAL);
+
+ if (created) {
+ // Find our actual node path by listing children and finding the one
we just created, the UUID makes the path
+ // unique, even if we have multiple requests from the same controller
+ List<String> children = _propertyStore.getChildNames(lockBasePath,
AccessOption.PERSISTENT);
+ List<String> allLockNodePathsForController = new ArrayList<>();
+ String lockNodePath = null;
+ if (children != null) {
+ // Find any node that starts with our controller ID and contains
"-lock-"
+ for (String child : children) {
+ if (child.startsWith(_controllerInstanceId) &&
child.contains("-lock-")) {
+ if (child.startsWith(_controllerInstanceId + "-" + uuid)) {
+ // If the node also contains the UUID, it's the lock we created
+ lockNodePath = lockBasePath + "/" + child;
+ }
+ allLockNodePathsForController.add(lockBasePath + "/" + child);
+ }
+ }
+ }
+
+ LOGGER.info("Found {} lockNodePaths for controller instance: {}, list:
{}, first lockNodePath cached: {}",
+ allLockNodePathsForController.size(), _controllerInstanceId,
allLockNodePathsForController, lockNodePath);
+
+ if (lockNodePath != null && allLockNodePathsForController.size() == 1)
{
+ // Check if we got the lowest sequence number (i.e., we're first in
line)
+ List<String> allChildren =
_propertyStore.getChildNames(lockBasePath, AccessOption.PERSISTENT);
+ if (allChildren != null && !allChildren.isEmpty()) {
+ allChildren.sort(TASK_LOCK_SEQUENCE_ID_COMPARATOR); // Sort by
sequence number
+ String ourNode =
lockNodePath.substring(lockNodePath.lastIndexOf('/') + 1);
+ if (ourNode.equals(allChildren.get(0))) {
+ // We have the lock!
+ LOGGER.info("Acquired lock with ephemeral node: {}",
lockNodePath);
+ return new TaskLock(lockKey, _controllerInstanceId,
currentTimeMs, lockNodePath);
+ } else {
+ // Someone else has the lock, clean up our node
+ boolean status = _propertyStore.remove(lockNodePath,
AccessOption.EPHEMERAL);
+ LOGGER.info("Did not get lock, removing ephemeral node: {},
return status: {}", lockNodePath, status);
+ return null;
+ }
+ } else {
+ // No children found, something went wrong, clean up
+ boolean status = _propertyStore.remove(lockNodePath,
AccessOption.EPHEMERAL);
+ LOGGER.warn("No children found under {}. Remove lockNodePath
status: {} for node: {}. Something must have "
+ + "gone wrong", lockBasePath, status, lockNodePath);
+ return null;
+ }
+ } else {
+ // Could not find our node path, or found too many paths for the
same controller, cleanup failed creation
+ LOGGER.warn("Either lockNodePath: {} wasn't found, or too many locks
({}) found for the same controller: {},"
+ + "list of locks: {}", lockNodePath,
allLockNodePathsForController.size(), _controllerInstanceId,
+ allLockNodePathsForController);
+
+ if (lockNodePath != null) {
+ boolean status = _propertyStore.remove(lockNodePath,
AccessOption.EPHEMERAL);
+ LOGGER.warn("Remove lockNodePath status: {} for path: {}", status,
lockNodePath);
+ }
+ return null;
+ }
+ }
+ return null;
+ } catch (Exception e) {
+ LOGGER.error("Error creating ephemeral lock under path: {}, lockKey:
{}", lockBasePath, lockKey, e);
+ return null;
+ }
+ }
+
+ private void ensureBasePaths() {
+ try {
+ // Ensure minion task metadata base path exists
+ String basePath = getBasePath();
+ if (!_propertyStore.exists(basePath, AccessOption.PERSISTENT)) {
+ ZNRecord baseRecord = new ZNRecord("MINION_TASK_METADATA");
Review Comment:
the basepath should already be created in some other code flow right ? I am
not able to find it, so not sure if this is true. but given that the path
exists in ZK, it should be created somewhere.
and if it is already created somewhere else, should we call that code path
here rather than having duplicate code to create the path with required znode
metadata / configs like AccessOption.PERSISTENT and baseRecord, etc ?
##########
pinot-controller/src/main/java/org/apache/pinot/controller/helix/core/minion/DistributedTaskLockManager.java:
##########
@@ -0,0 +1,423 @@
+/**
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+package org.apache.pinot.controller.helix.core.minion;
+
+import com.google.common.annotations.VisibleForTesting;
+import java.util.ArrayList;
+import java.util.Comparator;
+import java.util.List;
+import java.util.UUID;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
+import javax.annotation.Nullable;
+import org.apache.helix.AccessOption;
+import org.apache.helix.store.zk.ZkHelixPropertyStore;
+import org.apache.helix.zookeeper.datamodel.ZNRecord;
+import org.apache.pinot.common.metadata.ZKMetadataProvider;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+
+/**
+ * Manages distributed locks for minion task generation using ZooKeeper
ephemeral sequential nodes.
+ * Uses ephemeral nodes that automatically disappear when the controller
session ends.
+ * This approach provides automatic cleanup and is suitable for long-running
task generation.
+ * Locks are held until explicitly released or the controller session
terminates.
+ * Locks are at the table level, to ensure that only one type of task can be
generated per table at any given time.
+ * <p>
+ * ZK EPHEMERAL_SEQUENTIAL Locks (see <a
href="https://zookeeper.apache.org/doc/current/recipes.html#sc_recipes_Locks";>
+ * ZooKeeper Lock Recipe.</a> for more details):
+ * <ul>
+ * <li>Every lock is created with a lock prefix. Lock prefix used:
[controllerName]-lock-[UUID]. The UUID helps
+ * differentiate between requests originating from the same controller at
the same time
+ * <li>When ZK creates the ZNode, it appends a sequence number at the end.
E.g.
+ * [controllerName]-lock-[UUID]-00000001
+ * <li>The sequence number is used to identify the lock winner in case
more than one lock node is created at the
+ * same time. The smallest sequence number always wins
+ * <li>The locks are EPHEMERAL in nature, meaning that once the session
with ZK is lost, the lock is automatically
+ * cleaned up. Scenarios when the ZK session can be lost: a) controller
shutdown, b) controller crash, c) ZK session
+ * expiry (e.g. long GC pauses can cause this)
+ * <li>This implementation does not set up watches as described in the
recipe as the task lock is released whenever
+ * we identify that the lock is already acquired. Do not expect lock
ownership to automatically change for the
+ * time being. If such support is needed in the future, this can be
enhanced to add a watch on the neighboring
+ * lock node
+ * </ul>
+ * <p>
+ * Example of how the locks will work:
+ * <p>
+ * Say we have two controllers, and one controller happens to run 2 threads at
the same time, all of which need to take
+ * the distributed lock. Each thread will create a distributed lock node, and
the "-Lock" ZNode getChildren will return:
+ * <ul>
+ * <li>controller2-lock-xyzwx-00000002
+ * <li>controller1-lock-abcde-00000001
+ * <li>controller1-lock-ab345-00000003
+ * </ul>
+ * <p>
+ * In the above, the controller1 with UUID abcde will win the lock as it has
the smallest sequence number. The other
+ * two threads will clean up their locks and return error that the distributed
lock could not be acquired. Controller1
+ * will proceed with performing its tasks, and when done will release the lock.
+ */
+public class DistributedTaskLockManager {
+ private static final Logger LOGGER =
LoggerFactory.getLogger(DistributedTaskLockManager.class);
+
+ // Lock paths are constructed using ZKMetadataProvider
+ private static final String LOCK_OWNER_KEY = "lockOwner";
+ private static final String LOCK_UUID_KEY = "lockUuid";
+ private static final String LOCK_TIMESTAMP_MILLIS_KEY =
"lockTimestampMillis";
+
+ // Define a custom comparator to compare strings of format
'<controllerName>-lock-<uuid>-<sequenceNumber>' and sort
+ // them by the sequence number at the end
+ private static final Comparator<String> TASK_LOCK_SEQUENCE_ID_COMPARATOR =
(s1, s2) -> {
+ // Regex to find the trailing sequence of digits
+ Pattern p = Pattern.compile("\\d+$");
+
+ // Extract the number from the first string
+ Matcher m1 = p.matcher(s1);
+ long num1 = m1.find() ? Long.parseLong(m1.group()) : 0;
+
+ // Extract the number from the second string
+ Matcher m2 = p.matcher(s2);
+ long num2 = m2.find() ? Long.parseLong(m2.group()) : 0;
+
+ return Long.compare(num1, num2);
+ };
+
+ private final ZkHelixPropertyStore<ZNRecord> _propertyStore;
+ private final String _controllerInstanceId;
+
+ public DistributedTaskLockManager(ZkHelixPropertyStore<ZNRecord>
propertyStore, String controllerInstanceId) {
+ _propertyStore = propertyStore;
+ _controllerInstanceId = controllerInstanceId;
+
+ // Ensure base paths exist
+ ensureBasePaths();
+ }
+
+ /**
+ * Attempts to acquire a distributed lock at the table level for task
generation using session-based locking.
+ * The lock is held until explicitly released or the controller session ends.
Review Comment:
nit - can add a note on why we need to lock at table level rather than table
+ task level.
because some task types depend on other task types to not run parallely. so
serialised execution will guarantee such expectations.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
