rakeshadr commented on code in PR #10073: URL: https://github.com/apache/ozone/pull/10073#discussion_r3090448324
########## hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/node/PendingContainerTracker.java: ########## @@ -0,0 +1,341 @@ +/* + * 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.Objects; +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; + LOG.debug("Double roll interval elapsed: ({}ms) dropped: {} pending containers from both windows", + elapsed, getCount()); + } else if (elapsed >= rollIntervalMs) { + previousWindow.clear(); + final Set<ContainerID> tmp = previousWindow; + previousWindow = currentWindow; + currentWindow = tmp; + lastRollTime = now; + LOG.debug("Rolled window. Previous window size: {} elapsed: ({}ms), Current window reset to empty", + previousWindow.size(), elapsed); + } + } + + synchronized boolean contains(ContainerID containerID) { + return currentWindow.contains(containerID) || previousWindow.contains(containerID); + } + + /** + * Add container to current window. + */ + synchronized boolean add(ContainerID containerID, DatanodeID dnID) { + boolean added = currentWindow.add(containerID); + LOG.debug("Recorded pending container {} on DataNode {}. Added={}, Total pending={}", + containerID, dnID, added, getCount()); + return added; + } + + /** + * Remove container from both windows. + */ + synchronized boolean remove(ContainerID containerID, DatanodeID dnID) { + boolean removedFromCurrent = currentWindow.remove(containerID); + boolean removedFromPrevious = previousWindow.remove(containerID); + boolean removed = removedFromCurrent || removedFromPrevious; + LOG.debug("Removed pending container {} from DataNode {}. Removed={}, Remaining={}", + containerID, dnID, removed, getCount()); + return removed; + } + + /** + * 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) { + Objects.requireNonNull(dn, "dn == null"); + return get(dn.getID()); + } + } + + 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) { + Objects.requireNonNull(node, "node == null"); + 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) { + Objects.requireNonNull(node, "node == null"); + Objects.requireNonNull(datanodeInfo, "datanodeInfo == null"); + if (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(); + Objects.requireNonNull(storageReports, "storageReports == null"); + if (storageReports.isEmpty()) { + return false; + } + for (StorageReportProto report : storageReports) { Review Comment: @ashishkumar50 Point-1) Can you add tests for this logic with multiple volumes in a datanode. Test scenario: ``` pendingAllocationBytes = 15GB Volume-0: capacity=100GB, usableSpace=20GB <---- (20-15 >= 5) <--- return true Volume-1: capacity=100GB, usableSpace=1GB Volume-2: capacity=100GB, usableSpace=1GB ``` Point-2) There is a corner case. Say, all 3 volumes have 15GB free and `pendingAllocationBytes` is 15GB. A 5GB container fits easily. But SCM wrongly rejects the entire DN because it applied 15GB of pending (which in reality may all be on one volume) to every volume. Test scenario: False negative case, since `pendingAllocationBytes` is the total pending across the entire DN, not per-volume and causing the trouble. Good thing is it won't result into write failure but it will result into unused space eventhough volumes has space. ``` pendingAllocationBytes = 15GB Volume-0: capacity=100GB, usableSpace=15GB <---- (15-15 >= 0) <--- return false Volume-1: capacity=100GB, usableSpace=15GB <---- (15-15 >= 0) <--- return false Volume-2: capacity=100GB, usableSpace=15GB <---- (15-15 >= 0) <--- return false ``` -- 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]
