szetszwo commented on code in PR #10073: URL: https://github.com/apache/ozone/pull/10073#discussion_r3081134986
########## hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/node/PendingContainerTracker.java: ########## @@ -0,0 +1,364 @@ +/* + * 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.hadoop.hdds.scm.node; + +import com.google.common.annotations.VisibleForTesting; +import java.util.HashSet; +import java.util.List; +import java.util.Set; +import java.util.concurrent.ConcurrentHashMap; +import org.apache.hadoop.hdds.protocol.DatanodeDetails; +import org.apache.hadoop.hdds.protocol.DatanodeID; +import org.apache.hadoop.hdds.protocol.proto.StorageContainerDatanodeProtocolProtos.StorageReportProto; +import org.apache.hadoop.hdds.scm.container.ContainerID; +import org.apache.hadoop.hdds.scm.pipeline.Pipeline; +import org.apache.hadoop.ozone.container.common.volume.VolumeUsage; +import org.apache.hadoop.util.Time; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +/** + * Tracks per-datanode pending container allocations at SCM using a Two Window Tumbling Bucket + * pattern (similar to HDFS HADOOP-3707). + * + * Two Window Tumbling Bucket for automatic aging and cleanup. + * + * How It Works: + * <li>Each DataNode has two sets: <b>currentWindow</b> and <b>previousWindow</b></li> + * <li>New allocations go into <b>currentWindow</b></li> + * <li>Every <b>ROLL_INTERVAL</b> (default 5 minutes): + * <ul> + * <li>previousWindow = currentWindow (shift)</li> + * <li>currentWindow = new empty set (reset)</li> + * <li>Old previousWindow is discarded (automatic aging)</li> + * </ul> + * </li> + * <li>When checking pending: return <b>union</b> of currentWindow + previousWindow</li> + * + * + * Example Timeline: + * <pre> + * Time | Action | CurrentWindow | PreviousWindow | Total Pending + * ------+---------------------------+---------------+----------------+-------------- + * 00:00 | Allocate Container-1 | {C1} | {} | {C1} + * 00:02 | Allocate Container-2 | {C1, C2} | {} | {C1, C2} + * 00:05 | [ROLL] Window tumbles | {} | {C1, C2} | {C1, C2} + * 00:07 | Allocate Container-3 | {C3} | {C1, C2} | {C1, C2, C3} + * 00:08 | Report confirms C1 | {C3} | {C2} | {C2, C3} + * 00:10 | [ROLL] Window tumbles | {} | {C3} | {C3} + * | (C2 aged out if not reported) + * </pre> + * + */ +public class PendingContainerTracker { + + private static final Logger LOG = LoggerFactory.getLogger(PendingContainerTracker.class); + + private final DatanodeBuckets datanodeBuckets; + + /** + * Maximum container size in bytes. + */ + private final long maxContainerSize; + + /** + * Metrics for tracking pending containers (same instance as {@link SCMNodeManager}'s node metrics). + */ + private final SCMNodeMetrics metrics; + + /** + * Two-window bucket for a single DataNode. + * Contains current and previous window sets, plus last roll timestamp. + */ + private static class TwoWindowBucket { + private Set<ContainerID> currentWindow = new HashSet<>(); + private Set<ContainerID> previousWindow = new HashSet<>(); + private long lastRollTime = Time.monotonicNow(); + private final long rollIntervalMs; + + TwoWindowBucket(long rollIntervalMs) { + this.rollIntervalMs = rollIntervalMs; + } + + /** + * Roll one or both windows based on elapsed time. + */ + synchronized void rollIfNeeded() { + long now = Time.monotonicNow(); + long elapsed = now - lastRollTime; + + if (elapsed >= 2 * rollIntervalMs) { + previousWindow.clear(); + currentWindow.clear(); + lastRollTime = now; + } else if (elapsed >= rollIntervalMs) { + previousWindow.clear(); + final Set<ContainerID> tmp = previousWindow; + previousWindow = currentWindow; + currentWindow = tmp; + lastRollTime = now; + LOG.debug("Rolled window. Previous window size: {}, Current window reset to empty", previousWindow.size()); + } + } + + synchronized boolean contains(ContainerID containerID) { + return currentWindow.contains(containerID) || previousWindow.contains(containerID); + } + + /** + * Add container to current window. + */ + synchronized boolean add(ContainerID containerID) { + return currentWindow.add(containerID); + } + + /** + * Remove container from both windows. + */ + synchronized boolean remove(ContainerID containerID) { + boolean removedFromCurrent = currentWindow.remove(containerID); + boolean removedFromPrevious = previousWindow.remove(containerID); + return removedFromCurrent || removedFromPrevious; + } + + /** + * Count of pending containers in both windows. + */ + synchronized int getCount() { + return currentWindow.size() + previousWindow.size(); + } + } + + /** + * Per-datanode two-window buckets. + */ + private static class DatanodeBuckets { + private final ConcurrentHashMap<DatanodeID, TwoWindowBucket> map = new ConcurrentHashMap<>(); + private final long rollIntervalMs; + + DatanodeBuckets(long rollIntervalMs) { + this.rollIntervalMs = rollIntervalMs; + } + + TwoWindowBucket get(DatanodeID id) { + final TwoWindowBucket bucket = map.compute(id, (k, b) -> b != null ? b : new TwoWindowBucket(rollIntervalMs)); + bucket.rollIfNeeded(); + return bucket; + } + + TwoWindowBucket get(DatanodeDetails dn) { + if (dn == null) { + return null; + } Review Comment: It seems like a bug for `dn == null`. Is there a legitimate case to call get(null)? ########## hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/node/PendingContainerTracker.java: ########## @@ -0,0 +1,364 @@ +/* + * 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.hadoop.hdds.scm.node; + +import com.google.common.annotations.VisibleForTesting; +import java.util.HashSet; +import java.util.List; +import java.util.Set; +import java.util.concurrent.ConcurrentHashMap; +import org.apache.hadoop.hdds.protocol.DatanodeDetails; +import org.apache.hadoop.hdds.protocol.DatanodeID; +import org.apache.hadoop.hdds.protocol.proto.StorageContainerDatanodeProtocolProtos.StorageReportProto; +import org.apache.hadoop.hdds.scm.container.ContainerID; +import org.apache.hadoop.hdds.scm.pipeline.Pipeline; +import org.apache.hadoop.ozone.container.common.volume.VolumeUsage; +import org.apache.hadoop.util.Time; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +/** + * Tracks per-datanode pending container allocations at SCM using a Two Window Tumbling Bucket + * pattern (similar to HDFS HADOOP-3707). + * + * Two Window Tumbling Bucket for automatic aging and cleanup. + * + * How It Works: + * <li>Each DataNode has two sets: <b>currentWindow</b> and <b>previousWindow</b></li> + * <li>New allocations go into <b>currentWindow</b></li> + * <li>Every <b>ROLL_INTERVAL</b> (default 5 minutes): + * <ul> + * <li>previousWindow = currentWindow (shift)</li> + * <li>currentWindow = new empty set (reset)</li> + * <li>Old previousWindow is discarded (automatic aging)</li> + * </ul> + * </li> + * <li>When checking pending: return <b>union</b> of currentWindow + previousWindow</li> + * + * + * Example Timeline: + * <pre> + * Time | Action | CurrentWindow | PreviousWindow | Total Pending + * ------+---------------------------+---------------+----------------+-------------- + * 00:00 | Allocate Container-1 | {C1} | {} | {C1} + * 00:02 | Allocate Container-2 | {C1, C2} | {} | {C1, C2} + * 00:05 | [ROLL] Window tumbles | {} | {C1, C2} | {C1, C2} + * 00:07 | Allocate Container-3 | {C3} | {C1, C2} | {C1, C2, C3} + * 00:08 | Report confirms C1 | {C3} | {C2} | {C2, C3} + * 00:10 | [ROLL] Window tumbles | {} | {C3} | {C3} + * | (C2 aged out if not reported) + * </pre> + * + */ +public class PendingContainerTracker { + + private static final Logger LOG = LoggerFactory.getLogger(PendingContainerTracker.class); + + private final DatanodeBuckets datanodeBuckets; + + /** + * Maximum container size in bytes. + */ + private final long maxContainerSize; + + /** + * Metrics for tracking pending containers (same instance as {@link SCMNodeManager}'s node metrics). + */ + private final SCMNodeMetrics metrics; + + /** + * Two-window bucket for a single DataNode. + * Contains current and previous window sets, plus last roll timestamp. + */ + private static class TwoWindowBucket { + private Set<ContainerID> currentWindow = new HashSet<>(); + private Set<ContainerID> previousWindow = new HashSet<>(); + private long lastRollTime = Time.monotonicNow(); + private final long rollIntervalMs; + + TwoWindowBucket(long rollIntervalMs) { + this.rollIntervalMs = rollIntervalMs; + } + + /** + * Roll one or both windows based on elapsed time. + */ + synchronized void rollIfNeeded() { + long now = Time.monotonicNow(); + long elapsed = now - lastRollTime; + + if (elapsed >= 2 * rollIntervalMs) { + previousWindow.clear(); + currentWindow.clear(); + lastRollTime = now; + } else if (elapsed >= rollIntervalMs) { + previousWindow.clear(); + final Set<ContainerID> tmp = previousWindow; + previousWindow = currentWindow; + currentWindow = tmp; + lastRollTime = now; + LOG.debug("Rolled window. Previous window size: {}, Current window reset to empty", previousWindow.size()); + } + } + + synchronized boolean contains(ContainerID containerID) { + return currentWindow.contains(containerID) || previousWindow.contains(containerID); + } + + /** + * Add container to current window. + */ + synchronized boolean add(ContainerID containerID) { + return currentWindow.add(containerID); + } + + /** + * Remove container from both windows. + */ + synchronized boolean remove(ContainerID containerID) { + boolean removedFromCurrent = currentWindow.remove(containerID); + boolean removedFromPrevious = previousWindow.remove(containerID); + return removedFromCurrent || removedFromPrevious; + } + + /** + * Count of pending containers in both windows. + */ + synchronized int getCount() { + return currentWindow.size() + previousWindow.size(); + } + } + + /** + * Per-datanode two-window buckets. + */ + private static class DatanodeBuckets { + private final ConcurrentHashMap<DatanodeID, TwoWindowBucket> map = new ConcurrentHashMap<>(); + private final long rollIntervalMs; + + DatanodeBuckets(long rollIntervalMs) { + this.rollIntervalMs = rollIntervalMs; + } + + TwoWindowBucket get(DatanodeID id) { + final TwoWindowBucket bucket = map.compute(id, (k, b) -> b != null ? b : new TwoWindowBucket(rollIntervalMs)); + bucket.rollIfNeeded(); + return bucket; + } + + TwoWindowBucket get(DatanodeDetails dn) { + if (dn == null) { + return null; + } + return get(dn.getID()); + } + } + + public PendingContainerTracker(long maxContainerSize) { + this(maxContainerSize, 5 * 60 * 1000, null); // Default 5 minutes + } Review Comment: This is only used in test. Let's use the other constructor and remove this one. ########## hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/node/PendingContainerTracker.java: ########## @@ -0,0 +1,364 @@ +/* + * 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.hadoop.hdds.scm.node; + +import com.google.common.annotations.VisibleForTesting; +import java.util.HashSet; +import java.util.List; +import java.util.Set; +import java.util.concurrent.ConcurrentHashMap; +import org.apache.hadoop.hdds.protocol.DatanodeDetails; +import org.apache.hadoop.hdds.protocol.DatanodeID; +import org.apache.hadoop.hdds.protocol.proto.StorageContainerDatanodeProtocolProtos.StorageReportProto; +import org.apache.hadoop.hdds.scm.container.ContainerID; +import org.apache.hadoop.hdds.scm.pipeline.Pipeline; +import org.apache.hadoop.ozone.container.common.volume.VolumeUsage; +import org.apache.hadoop.util.Time; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +/** + * Tracks per-datanode pending container allocations at SCM using a Two Window Tumbling Bucket + * pattern (similar to HDFS HADOOP-3707). + * + * Two Window Tumbling Bucket for automatic aging and cleanup. + * + * How It Works: + * <li>Each DataNode has two sets: <b>currentWindow</b> and <b>previousWindow</b></li> + * <li>New allocations go into <b>currentWindow</b></li> + * <li>Every <b>ROLL_INTERVAL</b> (default 5 minutes): + * <ul> + * <li>previousWindow = currentWindow (shift)</li> + * <li>currentWindow = new empty set (reset)</li> + * <li>Old previousWindow is discarded (automatic aging)</li> + * </ul> + * </li> + * <li>When checking pending: return <b>union</b> of currentWindow + previousWindow</li> + * + * + * Example Timeline: + * <pre> + * Time | Action | CurrentWindow | PreviousWindow | Total Pending + * ------+---------------------------+---------------+----------------+-------------- + * 00:00 | Allocate Container-1 | {C1} | {} | {C1} + * 00:02 | Allocate Container-2 | {C1, C2} | {} | {C1, C2} + * 00:05 | [ROLL] Window tumbles | {} | {C1, C2} | {C1, C2} + * 00:07 | Allocate Container-3 | {C3} | {C1, C2} | {C1, C2, C3} + * 00:08 | Report confirms C1 | {C3} | {C2} | {C2, C3} + * 00:10 | [ROLL] Window tumbles | {} | {C3} | {C3} + * | (C2 aged out if not reported) + * </pre> + * + */ +public class PendingContainerTracker { + + private static final Logger LOG = LoggerFactory.getLogger(PendingContainerTracker.class); + + private final DatanodeBuckets datanodeBuckets; + + /** + * Maximum container size in bytes. + */ + private final long maxContainerSize; + + /** + * Metrics for tracking pending containers (same instance as {@link SCMNodeManager}'s node metrics). + */ + private final SCMNodeMetrics metrics; + + /** + * Two-window bucket for a single DataNode. + * Contains current and previous window sets, plus last roll timestamp. + */ + private static class TwoWindowBucket { + private Set<ContainerID> currentWindow = new HashSet<>(); + private Set<ContainerID> previousWindow = new HashSet<>(); + private long lastRollTime = Time.monotonicNow(); + private final long rollIntervalMs; + + TwoWindowBucket(long rollIntervalMs) { + this.rollIntervalMs = rollIntervalMs; + } + + /** + * Roll one or both windows based on elapsed time. + */ + synchronized void rollIfNeeded() { + long now = Time.monotonicNow(); + long elapsed = now - lastRollTime; + + if (elapsed >= 2 * rollIntervalMs) { + previousWindow.clear(); + currentWindow.clear(); + lastRollTime = now; + } else if (elapsed >= rollIntervalMs) { + previousWindow.clear(); + final Set<ContainerID> tmp = previousWindow; + previousWindow = currentWindow; + currentWindow = tmp; + lastRollTime = now; + LOG.debug("Rolled window. Previous window size: {}, Current window reset to empty", previousWindow.size()); + } + } + + synchronized boolean contains(ContainerID containerID) { + return currentWindow.contains(containerID) || previousWindow.contains(containerID); + } + + /** + * Add container to current window. + */ + synchronized boolean add(ContainerID containerID) { + return currentWindow.add(containerID); + } + + /** + * Remove container from both windows. + */ + synchronized boolean remove(ContainerID containerID) { + boolean removedFromCurrent = currentWindow.remove(containerID); + boolean removedFromPrevious = previousWindow.remove(containerID); + return removedFromCurrent || removedFromPrevious; + } + + /** + * Count of pending containers in both windows. + */ + synchronized int getCount() { + return currentWindow.size() + previousWindow.size(); + } + } + + /** + * Per-datanode two-window buckets. + */ + private static class DatanodeBuckets { + private final ConcurrentHashMap<DatanodeID, TwoWindowBucket> map = new ConcurrentHashMap<>(); + private final long rollIntervalMs; + + DatanodeBuckets(long rollIntervalMs) { + this.rollIntervalMs = rollIntervalMs; + } + + TwoWindowBucket get(DatanodeID id) { + final TwoWindowBucket bucket = map.compute(id, (k, b) -> b != null ? b : new TwoWindowBucket(rollIntervalMs)); + bucket.rollIfNeeded(); + return bucket; + } + + TwoWindowBucket get(DatanodeDetails dn) { + if (dn == null) { + return null; + } + return get(dn.getID()); + } + } + + public PendingContainerTracker(long maxContainerSize) { + this(maxContainerSize, 5 * 60 * 1000, null); // Default 5 minutes + } + + public PendingContainerTracker(long maxContainerSize, long rollIntervalMs, + SCMNodeMetrics metrics) { + this.datanodeBuckets = new DatanodeBuckets(rollIntervalMs); + this.maxContainerSize = maxContainerSize; + this.metrics = metrics; + LOG.info("PendingContainerTracker initialized with maxContainerSize={}B, rollInterval={}ms", + maxContainerSize, rollIntervalMs); + } + + /** + * Advances the two-window tumbling bucket for this datanode when the roll interval has elapsed. + * Call on periodic paths (node report) so windows age even when there are no new + * allocations or container reports touching this tracker. + */ + public void rollWindowsIfNeeded(DatanodeDetails node) { + if (node == null) { + return; + } + datanodeBuckets.get(node.getID()); + } + + /** + * Whether the datanode can fit another container of {@code containerSize} after accounting for + * SCM pending allocations for {@code node} (this tracker) and usable space on {@code datanodeInfo}. + * Combines {@link #getPendingAllocationSize} with the per-disk slot check in one call. + * + * @param node identity used to look up pending allocations (same DN as {@code datanodeInfo}) + * @param datanodeInfo storage reports for the datanode + * @param containerSize required container size in bytes (typically SCM max container size) + */ + public boolean hasEffectiveAllocatableSpaceForNewContainer( + DatanodeDetails node, DatanodeInfo datanodeInfo, long containerSize) { + if (node == null || datanodeInfo == null || containerSize <= 0) { + return false; + } + long pendingBytes = getPendingAllocationSize(node); + return hasAllocatableSpaceAfterPending(datanodeInfo, containerSize, pendingBytes); + } + + private boolean hasAllocatableSpaceAfterPending( + DatanodeInfo datanodeInfo, long containerSize, long pendingAllocationBytes) { + List<StorageReportProto> storageReports = datanodeInfo.getStorageReports(); + if (storageReports == null || storageReports.isEmpty()) { + return false; + } + for (StorageReportProto report : storageReports) { + long usableSpace = VolumeUsage.getUsableSpace(report); + if (usableSpace - pendingAllocationBytes >= containerSize) { + return true; + } + } + if (metrics != null) { + metrics.incNumSkippedFullNodeContainerAllocation(); + } + return false; + } + + /** + * Record a pending container allocation for all DataNodes in the pipeline. + * Container is added to the current window. + * + * @param pipeline The pipeline where container is allocated + * @param containerID The container being allocated + */ + public void recordPendingAllocation(Pipeline pipeline, ContainerID containerID) { + if (pipeline == null || containerID == null) { + LOG.warn("Ignoring null pipeline or containerID"); Review Comment: It seems like a bug for null. Is there a legitimate case to pass null? ########## hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/node/PendingContainerTracker.java: ########## @@ -0,0 +1,364 @@ +/* + * 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.hadoop.hdds.scm.node; + +import com.google.common.annotations.VisibleForTesting; +import java.util.HashSet; +import java.util.List; +import java.util.Set; +import java.util.concurrent.ConcurrentHashMap; +import org.apache.hadoop.hdds.protocol.DatanodeDetails; +import org.apache.hadoop.hdds.protocol.DatanodeID; +import org.apache.hadoop.hdds.protocol.proto.StorageContainerDatanodeProtocolProtos.StorageReportProto; +import org.apache.hadoop.hdds.scm.container.ContainerID; +import org.apache.hadoop.hdds.scm.pipeline.Pipeline; +import org.apache.hadoop.ozone.container.common.volume.VolumeUsage; +import org.apache.hadoop.util.Time; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +/** + * Tracks per-datanode pending container allocations at SCM using a Two Window Tumbling Bucket + * pattern (similar to HDFS HADOOP-3707). + * + * Two Window Tumbling Bucket for automatic aging and cleanup. + * + * How It Works: + * <li>Each DataNode has two sets: <b>currentWindow</b> and <b>previousWindow</b></li> + * <li>New allocations go into <b>currentWindow</b></li> + * <li>Every <b>ROLL_INTERVAL</b> (default 5 minutes): + * <ul> + * <li>previousWindow = currentWindow (shift)</li> + * <li>currentWindow = new empty set (reset)</li> + * <li>Old previousWindow is discarded (automatic aging)</li> + * </ul> + * </li> + * <li>When checking pending: return <b>union</b> of currentWindow + previousWindow</li> + * + * + * Example Timeline: + * <pre> + * Time | Action | CurrentWindow | PreviousWindow | Total Pending + * ------+---------------------------+---------------+----------------+-------------- + * 00:00 | Allocate Container-1 | {C1} | {} | {C1} + * 00:02 | Allocate Container-2 | {C1, C2} | {} | {C1, C2} + * 00:05 | [ROLL] Window tumbles | {} | {C1, C2} | {C1, C2} + * 00:07 | Allocate Container-3 | {C3} | {C1, C2} | {C1, C2, C3} + * 00:08 | Report confirms C1 | {C3} | {C2} | {C2, C3} + * 00:10 | [ROLL] Window tumbles | {} | {C3} | {C3} + * | (C2 aged out if not reported) + * </pre> + * + */ +public class PendingContainerTracker { + + private static final Logger LOG = LoggerFactory.getLogger(PendingContainerTracker.class); + + private final DatanodeBuckets datanodeBuckets; + + /** + * Maximum container size in bytes. + */ + private final long maxContainerSize; + + /** + * Metrics for tracking pending containers (same instance as {@link SCMNodeManager}'s node metrics). + */ + private final SCMNodeMetrics metrics; + + /** + * Two-window bucket for a single DataNode. + * Contains current and previous window sets, plus last roll timestamp. + */ + private static class TwoWindowBucket { + private Set<ContainerID> currentWindow = new HashSet<>(); + private Set<ContainerID> previousWindow = new HashSet<>(); + private long lastRollTime = Time.monotonicNow(); + private final long rollIntervalMs; + + TwoWindowBucket(long rollIntervalMs) { + this.rollIntervalMs = rollIntervalMs; + } + + /** + * Roll one or both windows based on elapsed time. + */ + synchronized void rollIfNeeded() { + long now = Time.monotonicNow(); + long elapsed = now - lastRollTime; + + if (elapsed >= 2 * rollIntervalMs) { + previousWindow.clear(); + currentWindow.clear(); + lastRollTime = now; + } else if (elapsed >= rollIntervalMs) { + previousWindow.clear(); + final Set<ContainerID> tmp = previousWindow; + previousWindow = currentWindow; + currentWindow = tmp; + lastRollTime = now; + LOG.debug("Rolled window. Previous window size: {}, Current window reset to empty", previousWindow.size()); + } + } + + synchronized boolean contains(ContainerID containerID) { + return currentWindow.contains(containerID) || previousWindow.contains(containerID); + } + + /** + * Add container to current window. + */ + synchronized boolean add(ContainerID containerID) { + return currentWindow.add(containerID); + } + + /** + * Remove container from both windows. + */ + synchronized boolean remove(ContainerID containerID) { + boolean removedFromCurrent = currentWindow.remove(containerID); + boolean removedFromPrevious = previousWindow.remove(containerID); + return removedFromCurrent || removedFromPrevious; + } + + /** + * Count of pending containers in both windows. + */ + synchronized int getCount() { + return currentWindow.size() + previousWindow.size(); + } + } + + /** + * Per-datanode two-window buckets. + */ + private static class DatanodeBuckets { + private final ConcurrentHashMap<DatanodeID, TwoWindowBucket> map = new ConcurrentHashMap<>(); + private final long rollIntervalMs; + + DatanodeBuckets(long rollIntervalMs) { + this.rollIntervalMs = rollIntervalMs; + } + + TwoWindowBucket get(DatanodeID id) { + final TwoWindowBucket bucket = map.compute(id, (k, b) -> b != null ? b : new TwoWindowBucket(rollIntervalMs)); + bucket.rollIfNeeded(); + return bucket; + } + + TwoWindowBucket get(DatanodeDetails dn) { + if (dn == null) { + return null; + } + return get(dn.getID()); + } + } + + public PendingContainerTracker(long maxContainerSize) { + this(maxContainerSize, 5 * 60 * 1000, null); // Default 5 minutes + } + + public PendingContainerTracker(long maxContainerSize, long rollIntervalMs, + SCMNodeMetrics metrics) { + this.datanodeBuckets = new DatanodeBuckets(rollIntervalMs); + this.maxContainerSize = maxContainerSize; + this.metrics = metrics; + LOG.info("PendingContainerTracker initialized with maxContainerSize={}B, rollInterval={}ms", + maxContainerSize, rollIntervalMs); + } + + /** + * Advances the two-window tumbling bucket for this datanode when the roll interval has elapsed. + * Call on periodic paths (node report) so windows age even when there are no new + * allocations or container reports touching this tracker. + */ + public void rollWindowsIfNeeded(DatanodeDetails node) { + if (node == null) { + return; + } + datanodeBuckets.get(node.getID()); + } + + /** + * Whether the datanode can fit another container of {@code containerSize} after accounting for + * SCM pending allocations for {@code node} (this tracker) and usable space on {@code datanodeInfo}. + * Combines {@link #getPendingAllocationSize} with the per-disk slot check in one call. + * + * @param node identity used to look up pending allocations (same DN as {@code datanodeInfo}) + * @param datanodeInfo storage reports for the datanode + * @param containerSize required container size in bytes (typically SCM max container size) + */ + public boolean hasEffectiveAllocatableSpaceForNewContainer( + DatanodeDetails node, DatanodeInfo datanodeInfo, long containerSize) { + if (node == null || datanodeInfo == null || containerSize <= 0) { + return false; + } + long pendingBytes = getPendingAllocationSize(node); + return hasAllocatableSpaceAfterPending(datanodeInfo, containerSize, pendingBytes); + } + + private boolean hasAllocatableSpaceAfterPending( + DatanodeInfo datanodeInfo, long containerSize, long pendingAllocationBytes) { + List<StorageReportProto> storageReports = datanodeInfo.getStorageReports(); + if (storageReports == null || storageReports.isEmpty()) { + return false; + } + for (StorageReportProto report : storageReports) { + long usableSpace = VolumeUsage.getUsableSpace(report); + if (usableSpace - pendingAllocationBytes >= containerSize) { + return true; + } + } + if (metrics != null) { + metrics.incNumSkippedFullNodeContainerAllocation(); + } + return false; + } + + /** + * Record a pending container allocation for all DataNodes in the pipeline. + * Container is added to the current window. + * + * @param pipeline The pipeline where container is allocated + * @param containerID The container being allocated + */ + public void recordPendingAllocation(Pipeline pipeline, ContainerID containerID) { + if (pipeline == null || containerID == null) { + LOG.warn("Ignoring null pipeline or containerID"); + return; + } + + for (DatanodeDetails node : pipeline.getNodes()) { + recordPendingAllocationForDatanode(node, containerID); + } + } + + /** + * Record a pending container allocation for a single DataNode. + * Container is added to the current window. + * + * @param node The DataNode where container is being allocated/replicated + * @param containerID The container being allocated/replicated + */ + public void recordPendingAllocationForDatanode(DatanodeDetails node, ContainerID containerID) { + if (node == null || containerID == null) { + LOG.warn("Ignoring null node or containerID"); + return; + } + + DatanodeID dnID = node.getID(); + boolean added = addContainerToBucket(containerID, dnID); + + if (added && metrics != null) { + metrics.incNumPendingContainersAdded(); + } + } + + private boolean addContainerToBucket(ContainerID containerID, DatanodeID dnID) { + TwoWindowBucket bucket = datanodeBuckets.get(dnID); + synchronized (bucket) { Review Comment: The synchronized (bucket) is not needed since add(..) is already synchronized. If we want to print the count in the debug message, move it inside add(..). -- 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]
