aswinshakil commented on code in PR #10000:
URL: https://github.com/apache/ozone/pull/10000#discussion_r3048152699
##########
hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/pipeline/PipelineManager.java:
##########
@@ -227,4 +228,11 @@ void reinitialize(Table<PipelineID, Pipeline>
pipelineStore)
* Get the pipeline metrics.
*/
SCMPipelineMetrics getMetrics();
+
+ /**
+ * Get DatanodeInfo for a specific DataNode which includes per-volume
storage reports.
+ * @param datanodeDetails The datanode to get info for
+ * @return DatanodeInfo containing detailed node information including
per-disk stats, or null if not available
+ */
+ DatanodeInfo getDatanodeInfo(DatanodeDetails datanodeDetails);
Review Comment:
Looks like this is not used, It can be removed.
##########
hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/node/PendingContainerTracker.java:
##########
@@ -0,0 +1,425 @@
+/*
+ * 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.Collections;
+import java.util.HashSet;
+import java.util.List;
+import java.util.Set;
+import java.util.UUID;
+import java.util.concurrent.ConcurrentHashMap;
+import org.apache.hadoop.hdds.protocol.DatanodeDetails;
+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>
+ *
+ * <p>Thread-safety: all mutations and reads for a given datanode run under
+ * {@code synchronized (bucket)} on that datanode's {@link TwoWindowBucket},
including
+ * map insert/remove, so concurrent record/remove/report paths cannot
interleave or orphan
+ * a bucket after it was dropped from the map.</p>
+ */
+public class PendingContainerTracker {
+
+ private static final Logger LOG =
LoggerFactory.getLogger(PendingContainerTracker.class);
+
+ /**
+ * Roll interval in milliseconds.
+ * Configurable via hdds.scm.container.pending.allocation.roll.interval.
+ * Default: 5 minutes.
+ * Containers automatically age out after 2 × rollIntervalMs.
+ */
+ private final long rollIntervalMs;
+
+ /**
+ * Map of DataNode UUID to TwoWindowBucket.
+ */
+ private final ConcurrentHashMap<UUID, TwoWindowBucket> 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 the windows: previous = current, current = empty.
+ * Called when current time exceeds lastRollTime + rollIntervalMs.
+ */
+ synchronized void rollIfNeeded() {
+ long now = Time.monotonicNow();
+ if (now - lastRollTime >= rollIntervalMs) {
+ // Shift: current becomes previous
+ previousWindow = currentWindow;
+ // Reset: new empty current window
+ currentWindow = new HashSet<>();
+ lastRollTime = now;
+ LOG.debug("Rolled window. Previous window size: {}, Current window
reset to empty", previousWindow.size());
+ }
+ }
+
+ /**
+ * Get union of both windows (all pending containers).
+ */
+ synchronized Set<ContainerID> getAllPending() {
+ Set<ContainerID> all = new HashSet<>();
+ all.addAll(currentWindow);
+ all.addAll(previousWindow);
+ return all;
+ }
+
+ /**
+ * 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;
+ }
+
+ /**
+ * Check if either window is non-empty.
+ */
+ synchronized boolean isEmpty() {
+ return currentWindow.isEmpty() && previousWindow.isEmpty();
+ }
+
+ /**
+ * Count of pending containers in both windows.
+ */
+ synchronized int getCount() {
+ return currentWindow.size() + previousWindow.size();
+ }
+ }
+
+ public PendingContainerTracker(long maxContainerSize) {
+ this(maxContainerSize, 10 * 60 * 1000, null); // Default 10 minutes
+ }
+
+ public PendingContainerTracker(long maxContainerSize, long rollIntervalMs,
+ SCMNodeMetrics metrics) {
+ this.datanodeBuckets = new ConcurrentHashMap<>();
+ this.maxContainerSize = maxContainerSize;
+ this.rollIntervalMs = rollIntervalMs;
+ this.metrics = metrics;
+ LOG.info("PendingContainerTracker initialized with maxContainerSize={}B,
rollInterval={}ms",
+ maxContainerSize, rollIntervalMs);
+ }
+
+ /**
+ * 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)
+ */
+ /**
+ * Advances the two-window tumbling bucket for this datanode when the roll
interval has elapsed.
+ * Call on periodic paths (e.g. 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;
+ }
+ UUID uuid = node.getUuid();
Review Comment:
`node.getUuid()` is being deprecated, We can use `node.getID()` instead. If
we change here, need to change in all the places in the class
##########
hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/node/PendingContainerTracker.java:
##########
@@ -0,0 +1,338 @@
+/*
+ * 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.container;
+
+import com.google.common.annotations.VisibleForTesting;
+import java.util.Collections;
+import java.util.HashSet;
+import java.util.Set;
+import java.util.UUID;
+import java.util.concurrent.ConcurrentHashMap;
+import org.apache.hadoop.hdds.protocol.DatanodeDetails;
+import org.apache.hadoop.hdds.scm.container.metrics.SCMContainerManagerMetrics;
+import org.apache.hadoop.hdds.scm.pipeline.Pipeline;
+import org.apache.hadoop.util.Time;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * Tracks pending container allocations using a Two Window Tumbling Bucket
pattern.
+ * Similar like 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 10 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}
+ * 00:05 | Allocate Container-2 | {C1, C2} | {} | {C1,
C2}
+ * 00:10 | [ROLL] Window tumbles | {} | {C1, C2} | {C1,
C2}
+ * 00:12 | Allocate Container-3 | {C3} | {C1, C2} | {C1,
C2, C3}
+ * 00:15 | Report confirms C1 | {C3} | {C2} | {C2,
C3}
+ * 00:20 | [ROLL] Window tumbles | {} | {C3} | {C3}
+ * | (C2 aged out if not reported)
+ * </pre>
+ *
+ */
+public class PendingContainerTracker {
+
+ private static final Logger LOG =
LoggerFactory.getLogger(PendingContainerTracker.class);
+
+ /**
+ * Roll interval in milliseconds.
+ * Configurable via hdds.scm.container.pending-allocation.roll-interval.
+ * Default: 10 minutes.
+ * Containers automatically age out after 2 × rollIntervalMs.
+ */
+ private final long rollIntervalMs;
+
+ /**
+ * Map of DataNode UUID to TwoWindowBucket.
+ */
+ private final ConcurrentHashMap<UUID, TwoWindowBucket> datanodeBuckets;
+
+ /**
+ * Maximum container size in bytes.
+ */
+ private final long maxContainerSize;
+
+ /**
+ * Metrics for tracking pending containers.
+ */
+ private final SCMContainerManagerMetrics 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 = ConcurrentHashMap.newKeySet();
+ private Set<ContainerID> previousWindow = ConcurrentHashMap.newKeySet();
+ private long lastRollTime = Time.monotonicNow();
+ private final long rollIntervalMs;
+
+ TwoWindowBucket(long rollIntervalMs) {
+ this.rollIntervalMs = rollIntervalMs;
+ }
+
+ /**
+ * Roll the windows: previous = current, current = empty.
+ * Called when current time exceeds lastRollTime + rollIntervalMs.
+ */
+ synchronized void rollIfNeeded() {
Review Comment:
We don't have to roll for every node report. If there are 10,000+ datanodes
then we will try to roll for each report every minute. Even if the node is not
even used for allocation we still roll every time. We are already rolling when
there are some container allocation/removal ops.
Also rolling shouldn't just depend on the timing of the report to roll, The
node might go down and come back. if someone sets the
`ozone.scm.stale.node.interval` and `ozone.scm.stale.node.interval `> 10mins
then this logic wouldn't work. If the node comes back after 10 mins and then
sends a report we will only roll once. We can just roll like mentioned in
https://github.com/apache/ozone/pull/10000/changes#r3044295093 or something
like this
```
long now = Time.monotonicNow();
long elapsed = now - lastRollTime;
if (elapsed >= 2 * rollIntervalMs) {
previousWindow.clear();
currentWindow.clear();
lastRollTime = now;
} else if (elapsed >= rollIntervalMs) {
previousWindow = currentWindow;
currentWindow = new HashSet<>();
lastRollTime = now;
}
```
##########
hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/container/ContainerManagerImpl.java:
##########
Review Comment:
@rakeshadr @ashishkr200 `pipelineManager.hasEnoughSpace` is not a
synchronized operation on the datanodes. We take lock on the pipeline
`synchronized (pipeline.getId())`. Still two different pipelines can still
allocate container on the same datanode.
We record the pending allocation on only Line 289. Until the code reaches
that point multiple pipeline will be allocated to the same space on the
datanode.
Let's say we only have 6GB space available in d1.
Pipeline 1: d1, d2, d3
Pipeline 2: d1, d4, d5
When we allocate container on both the pipeline, will get past
`pipelineManager.hasEnoughSpace` and both will be allocated container 5GB *2 =
10GB. But we only have 5GB on d1
--
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]
