[
https://issues.apache.org/jira/browse/HDFS-17463?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17837566#comment-17837566
]
ASF GitHub Bot commented on HDFS-17463:
---------------------------------------
hiwangzhihui commented on code in PR #6736:
URL: https://github.com/apache/hadoop/pull/6736#discussion_r1566889645
##########
hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/SerialNumberManager.java:
##########
@@ -115,6 +119,36 @@ public static StringTable getStringTable() {
return map;
}
+ // returns serial snapshot of current values for a save.
+ public static StringTable getSerialStringTable() {
+ // approximate size for capacity.
+ int size = 0;
+ for (final SerialNumberManager snm : values) {
+ size += snm.size();
+ }
+
+ int tableMaskBits = getMaskBits();
+ StringTable map = new StringTable(size, 0);
+ Set<String> entrySet = new HashSet<String>();
+ AtomicInteger index = new AtomicInteger();
+
+ Map <Integer, String> entryMap = new TreeMap<Integer, String>();
+ for (final SerialNumberManager snm : values) {
+ final int mask = snm.getMask(tableMaskBits);
+ for (Entry<Integer, String> entry : snm.entrySet()) {
+ entryMap.put(entry.getKey() | mask, entry.getValue());
+ }
+ }
+
+ entryMap.forEach((key, value) -> {
+ if(entrySet.add(value)){
+ map.put(index.incrementAndGet(), value);
Review Comment:
Why not execute the incrementAndGet method when persistently collecting
information?
##########
hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSImageFormatPBINode.java:
##########
@@ -628,25 +638,59 @@ private void loadRootINode(INodeSection.INode p) {
public final static class Saver {
private long numImageErrors;
- private static long buildPermissionStatus(INodeAttributes n) {
- return n.getPermissionLong();
+ private static long buildPermissionStatus(INodeAttributes n,
+ boolean enableSaveSplitIdStringTable) {
+ if(enableSaveSplitIdStringTable){
+ return n.getPermissionLong();
+ } else {
+ StringTable serialStringTable =
SerialNumberManager.getSerialStringTable();
+ Map<String, Integer> stringMap = StreamSupport
+ .stream(serialStringTable.spliterator(), false)
+ .collect(Collectors.toMap(Map.Entry::getValue, Map.Entry::getKey));
+
+ long permission = n.getPermissionLong();
+ permission = USER.BITS.combine(stringMap.get(n.getUserName()),
permission);
+ permission = GROUP.BITS.combine(stringMap.get(n.getGroupName()),
permission);
+ return permission;
+ }
}
- private static AclFeatureProto.Builder buildAclEntries(AclFeature f) {
+ private static AclFeatureProto.Builder buildAclEntries(AclFeature f,
+ boolean enableSaveSplitIdStringTable) {
AclFeatureProto.Builder b = AclFeatureProto.newBuilder();
+ Map<String, Integer> stringMap = null;
+ if(!enableSaveSplitIdStringTable){
+ StringTable serialStringTable =
SerialNumberManager.getSerialStringTable();
+ stringMap = StreamSupport.stream(serialStringTable.spliterator(),
false)
+ .collect(Collectors.toMap(Map.Entry::getValue,
Map.Entry::getKey));
+ }
+
for (int pos = 0, e; pos < f.getEntriesSize(); pos++) {
e = f.getEntryAt(pos);
+ if(!enableSaveSplitIdStringTable){
+ e = (int)
NAME.BITS.combine(stringMap.get(AclEntryStatusFormat.getName(e)), e);
+ }
b.addEntries(e);
}
return b;
}
- private static XAttrFeatureProto.Builder buildXAttrs(XAttrFeature f) {
+ private static XAttrFeatureProto.Builder buildXAttrs(XAttrFeature f,
boolean enableSaveSplitIdStringTable) {
XAttrFeatureProto.Builder b = XAttrFeatureProto.newBuilder();
+ Map<String, Integer> stringMap = null;
+ if(!enableSaveSplitIdStringTable){
+ StringTable serialStringTable =
SerialNumberManager.getSerialStringTable();
Review Comment:
When enableSaveSplitIdStringTable=false, different StringMaps should be
obtained to store data
##########
hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/AclEntryStatusFormat.java:
##########
@@ -47,7 +47,7 @@ public enum AclEntryStatusFormat implements
LongBitFormat.Enum {
private static final AclEntryType[] ACL_ENTRY_TYPE_VALUES =
AclEntryType.values();
- private final LongBitFormat BITS;
Review Comment:
Why remove the “private”
##########
hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/SerialNumberManager.java:
##########
@@ -115,6 +119,36 @@ public static StringTable getStringTable() {
return map;
}
+ // returns serial snapshot of current values for a save.
+ public static StringTable getSerialStringTable() {
Review Comment:
Please add test cases.
> Support the switch StringTable Split ID feature
> -----------------------------------------------
>
> Key: HDFS-17463
> URL: https://issues.apache.org/jira/browse/HDFS-17463
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: namenode
> Affects Versions: 3.2.0, 3.3.5, 3.3.3, 3.3.4
> Reporter: wangzhihui
> Priority: Major
> Labels: pull-request-available
> Attachments: Image_struct.png, error.png
>
>
> desc:
> *
> Hadoop 3.2 introduced optimization features for HDFS StringTable
> (b60ca37914b22550e3630fa02742d40697decb3), It resulted in lower versions of
> Hadoop upgraded to 3.2 and later versions not supporting downgrade
> operations.
> !error.png!
> * This issue has also been discussed in HDFS-14831, and it is recommended to
> revert the feature, but it cannot fundamentally solve the problem。
> *
> Therefore, we have added an optimization to support downgrading
>
> Solution:
> * First, we will add the "dfs. image. save. splitId. stringTable" conf
> switch "StringTable optimization feature" is enabled
> * When the conf value is false, an Image file compatible with lower versions
> of HDFS is generated to support downgrading.
> *
> The difference in HDFS Image file format between Hadoop 3.1.1 and Hadoop 3.2
> is shown in the following figure.
> * With the sub-sections feature introduced in HDFS-14617, Protobuf can
> support compatible reading.
> *
> The data structure causing incompatible differences is mainly StringTable.
> !Image_struct.png|width=396,height=163!
> * In "dfs.image.save.splitId.stringTable = false " the Id generation order
> of StringTable starts from 0 to Integer.Max. When true, the Id value range
> follows the latest rules.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
