Apache
/
hadoop
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

MAPREDUCE-7150. Optimize collections used by MR JHS to reduce its memory. (Contributed by Misha Dmitriev)

This commit is contained in:
Haibo Chen 2018-10-16 13:44:41 -07:00
parent c2288ac45b
commit babd1449bf
4 changed files with 38 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

56
hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/java/org/apache/hadoop/mapreduce/counters/FileSystemCounterGroup.java
View File

@ -61,8 +61,9 @@ public abstract class FileSystemCounterGroup<C extends Counter>
// C[] would need Array.newInstance which requires a Class<C> reference.
// Just a few local casts probably worth not having to carry it around.
private final Map<String, Object[]> map =
new ConcurrentSkipListMap<String, Object[]>();
// Initialized lazily, since in some situations millions of empty maps can
// waste a substantial (e.g. 4% as we observed) portion of the heap
private Map<String, Object[]> map;
private String displayName;
private static final Joiner NAME_JOINER = Joiner.on('_');
@ -214,6 +215,9 @@ public abstract class FileSystemCounterGroup<C extends Counter>
@SuppressWarnings("unchecked")
public synchronized C findCounter(String scheme, FileSystemCounter key) {
final String canonicalScheme = checkScheme(scheme);
if (map == null) {
map = new ConcurrentSkipListMap<>();
}
Object[] counters = map.get(canonicalScheme);
int ord = key.ordinal();
if (counters == null) {
@ -247,10 +251,12 @@ public abstract class FileSystemCounterGroup<C extends Counter>
protected abstract C newCounter(String scheme, FileSystemCounter key);
@Override
public int size() {
public synchronized int size() {
int n = 0;
for (Object[] counters : map.values()) {
n += numSetCounters(counters);
if (map != null) {
for (Object[] counters : map.values()) {
n += numSetCounters(counters);
}
}
return n;
}
@ -271,19 +277,23 @@ public abstract class FileSystemCounterGroup<C extends Counter>
* FileSystemGroup ::= #scheme (scheme #counter (key value)*)*
*/
@Override
public void write(DataOutput out) throws IOException {
WritableUtils.writeVInt(out, map.size()); // #scheme
for (Map.Entry<String, Object[]> entry : map.entrySet()) {
WritableUtils.writeString(out, entry.getKey()); // scheme
// #counter for the above scheme
WritableUtils.writeVInt(out, numSetCounters(entry.getValue()));
for (Object counter : entry.getValue()) {
if (counter == null) continue;
@SuppressWarnings("unchecked")
FSCounter c = (FSCounter) ((Counter)counter).getUnderlyingCounter();
WritableUtils.writeVInt(out, c.key.ordinal()); // key
WritableUtils.writeVLong(out, c.getValue()); // value
public synchronized void write(DataOutput out) throws IOException {
if (map != null) {
WritableUtils.writeVInt(out, map.size()); // #scheme
for (Map.Entry<String, Object[]> entry : map.entrySet()) {
WritableUtils.writeString(out, entry.getKey()); // scheme
// #counter for the above scheme
WritableUtils.writeVInt(out, numSetCounters(entry.getValue()));
for (Object counter : entry.getValue()) {
if (counter == null) continue;
@SuppressWarnings("unchecked")
FSCounter c = (FSCounter) ((Counter) counter).getUnderlyingCounter();
WritableUtils.writeVInt(out, c.key.ordinal()); // key
WritableUtils.writeVLong(out, c.getValue()); // value
}
}
} else {
WritableUtils.writeVInt(out, 0);
}
}
@ -310,8 +320,8 @@ public abstract class FileSystemCounterGroup<C extends Counter>
@Override
public Iterator<C> iterator() {
return new AbstractIterator<C>() {
Iterator<Object[]> it = map.values().iterator();
Object[] counters = it.hasNext() ? it.next() : null;
Iterator<Object[]> it = map != null ? map.values().iterator() : null;
Object[] counters = (it != null && it.hasNext()) ? it.next() : null;
int i = 0;
@Override
protected C computeNext() {
@ -322,7 +332,7 @@ public abstract class FileSystemCounterGroup<C extends Counter>
if (counter != null) return counter;
}
i = 0;
counters = it.hasNext() ? it.next() : null;
counters = (it != null && it.hasNext()) ? it.next() : null;
}
return endOfData();
}
@ -343,8 +353,10 @@ public abstract class FileSystemCounterGroup<C extends Counter>
public synchronized int hashCode() {
// need to be deep as counters is an array
int hash = FileSystemCounter.class.hashCode();
for (Object[] counters : map.values()) {
if (counters != null) hash ^= Arrays.hashCode(counters);
if (map != null) {
for (Object[] counters : map.values()) {
if (counters != null) hash ^= Arrays.hashCode(counters);
}
}
return hash;
}

2
hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/java/org/apache/hadoop/mapreduce/jobhistory/JobHistoryParser.java
View File

@ -600,7 +600,7 @@ public class JobHistoryParser implements HistoryEventHandler {
public TaskInfo() {
startTime = finishTime = -1;
error = splitLocations = "";
attemptsMap = new HashMap<TaskAttemptID, TaskAttemptInfo>();
attemptsMap = new HashMap<TaskAttemptID, TaskAttemptInfo>(2);
}
public void printAll() {

5
hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-hs/src/main/java/org/apache/hadoop/mapreduce/v2/hs/CompletedTask.java
View File

@ -20,7 +20,6 @@ package org.apache.hadoop.mapreduce.v2.hs;
import java.util.ArrayList;
import java.util.LinkedHashMap;
import java.util.LinkedList;
import java.util.List;
import java.util.Map;
import java.util.concurrent.atomic.AtomicBoolean;
@ -48,11 +47,11 @@ public class CompletedTask implements Task {
private final TaskInfo taskInfo;
private TaskReport report;
private TaskAttemptId successfulAttempt;
private List<String> reportDiagnostics = new LinkedList<String>();
private List<String> reportDiagnostics = new ArrayList<String>(2);
private Lock taskAttemptsLock = new ReentrantLock();
private AtomicBoolean taskAttemptsLoaded = new AtomicBoolean(false);
private final Map<TaskAttemptId, TaskAttempt> attempts =
new LinkedHashMap<TaskAttemptId, TaskAttempt>();
new LinkedHashMap<TaskAttemptId, TaskAttempt>(2);
CompletedTask(TaskId taskId, TaskInfo taskInfo) {
//TODO JobHistoryParser.handleTaskFailedAttempt should use state from the event.

2
hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-hs/src/main/java/org/apache/hadoop/mapreduce/v2/hs/CompletedTaskAttempt.java
View File

@ -39,7 +39,7 @@ public class CompletedTaskAttempt implements TaskAttempt {
private final TaskAttemptInfo attemptInfo;
private final TaskAttemptId attemptId;
private final TaskAttemptState state;
private final List<String> diagnostics = new ArrayList<String>();
private final List<String> diagnostics = new ArrayList<String>(2);
private TaskAttemptReport report;
private String localDiagMessage;