upgrade小结：
1.dfsadmin -upgradeProgress status 在cdh5.2.0中没有，在4.6.0有（见源码org.apache.hadoop.hdfs.tools.DFSAdmin）
升级的时候不能通过这个看到升级状态
rollingUpgrade这个参数在4.6.0中不存在，在5.2.0中有，可以用于滚动升级
2.在cdh5.2.0中执行upgrade,nn中调用的命令是
hadoop-daemon.sh start namenode -upgrade
最终调用org.apache.hadoop.hdfs.server.namenode.NameNode类，并传入upgrade参数
3.yarn的设置有些变动，下面两个参数会影响nm是否启动正常
<property> <name>yarn.nodemanager.aux-services</name> <value>mapreduce.shuffle</value> </property> <property> <name>yarn.nodemanager.aux-services.mapreduce.shuffle.class</name> <value>org.apache.hadoop.mapred.ShuffleHandler</value> </property>
修改为：
<property> <name>yarn.nodemanager.aux-services</name> <value>mapreduce_shuffle</value> </property> <property> <name>yarn.nodemanager.aux-services.mapreduce_shuffle.class</name> <value>org.apache.hadoop.mapred.ShuffleHandler</value> </property>
4.impala1.1.1和cdh5.2.0有兼容性问题
cdh5.x开始使用PB做通信，需要升级impala到2.0.0（impala的可用性/稳定性/性能需要测试）
rollback小结：
1.rollback要在4.6.0版本上运行
在4.6.0上运行rollback,finalize,upgrade时，都会判断时否为ha模式，如果没有关闭ha的配置会报如下错误：
14/11/19 15:25:47 FATAL namenode.NameNode: Exception in namenode join org.apache.hadoop.HadoopIllegalArgumentException: Invalid startup option. Cannot perform DFS upgrade with HA enabled. at org.apache.hadoop.hdfs.server.namenode.NameNode.createNameNode(NameNode.java:1130) at org.apache.hadoop.hdfs.server.namenode.NameNode.main(NameNode.java:1241)
namenode类main方法中通过createNameNode方法创建一个NameNode的实例
public static void main(String argv[]) throws Exception {
if (DFSUtil.parseHelpArgument(argv, NameNode.USAGE, System.out, true)) {
System.exit(0);
}
try {
StringUtils.startupShutdownMessage(NameNode.class, argv, LOG);
NameNode namenode = createNameNode(argv, null);
if (namenode != null)
namenode.join();
} catch (Throwable e) {
LOG.fatal("Exception in namenode join", e);
terminate(1, e);
}
}
而在createNameNode方法中会通过下面的代码检测是否为ha的配置，而在5.2.0是没有这个限制的。
if (HAUtil.isHAEnabled(conf, DFSUtil.getNamenodeNameServiceId(conf)) &&
(startOpt == StartupOption.UPGRADE ||
startOpt == StartupOption.ROLLBACK ||
startOpt == StartupOption.FINALIZE)) {
throw new HadoopIllegalArgumentException("Invalid startup option. " +
"Cannot perform DFS upgrade with HA enabled.");
}
这里会涉及到两个判断方法：
1）
org.apache.hadoop.hdfs.HAUti类的isHAEnabled方法：
public static boolean isHAEnabled(Configuration conf, String nsId) {
Map<String, Map<String, InetSocketAddress>> addresses =
DFSUtil.getHaNnRpcAddresses(conf);
if (addresses == null) return false;
Map<String, InetSocketAddress> nnMap = addresses.get(nsId);
return nnMap != null && nnMap.size() > 1;
}
这里会依次调用org.apache.hadoop.hdfs.DFSUtil类的
getHaNnRpcAddresses/getAddresses/getNameServiceIds/getAddressesForNameserviceId/getNameNodeIds
方法，解析dfs.nameservices/dfs.ha.namenodes.xxxx/dfs.namenode.rpc-address.xxxx的设置来获取每个nameserviceid到对应的nn rpc地址的对应关系（
Map<String, Map<String, InetSocketAddress>>
）并判断map value的size(如果dfs.ha.namenodes.x设置超过1个就算ha)，这里只要把配置改成下面即可
<property> <name>dfs.ha.namenodes.bipcluster</name> <value>nn1</value> </property>
2）如果设置了jn会报如下错误：
14/11/19 16:47:32 FATAL namenode.NameNode: Exception in namenode join java.io.IOException: Invalid configuration: a shared edits dir must not be specified if HA is not enabled. at org.apache.hadoop.hdfs.server.namenode.FSNamesystem.<init>(FSNamesystem.java:576) at org.apache.hadoop.hdfs.server.namenode.FSNamesystem.loadFromDisk(FSNamesystem.java:513) at org.apache.hadoop.hdfs.server.namenode.NameNode.loadNamesystem(NameNode.java:403) at org.apache.hadoop.hdfs.server.namenode.NameNode.initialize(NameNode.java:445) at org.apache.hadoop.hdfs.server.namenode.NameNode.<init>(NameNode.java:621) at org.apache.hadoop.hdfs.server.namenode.NameNode.<init>(NameNode.java:606) at org.apache.hadoop.hdfs.server.namenode.NameNode.createNameNode(NameNode.java:1177) at org.apache.hadoop.hdfs.server.namenode.NameNode.main(NameNode.java:1241) 14/11/19 16:47:32 INFO util.ExitUtil: Exiting with status 1
错误在FSNamesystem类的构造方法中:
final boolean persistBlocks = conf.getBoolean(DFS_PERSIST_BLOCKS_KEY,
DFS_PERSIST_BLOCKS_DEFAULT); //默认是false
// block allocation has to be persisted in HA using a shared edits directory
// so that the standby has up-to-date namespace information
String nameserviceId = DFSUtil.getNamenodeNameServiceId(conf);
this.haEnabled = HAUtil.isHAEnabled(conf, nameserviceId);
this.persistBlocks = persistBlocks || (haEnabled && HAUtil.usesSharedEditsDir(conf));
// Sanity check the HA-related config.
if (nameserviceId != null) {
LOG.info("Determined nameservice ID: " + nameserviceId); //Determined nameservice ID: bipcluster
}
LOG.info("HA Enabled: " + haEnabled); //HA Enabled: false
if (!haEnabled && HAUtil.usesSharedEditsDir(conf)) { //异常由这里抛出
LOG.warn("Configured NNs:\n" + DFSUtil.nnAddressesAsString(conf)); //Configured NNs: Nameservice <bipcluster>: NN ID nn1 => gd6g12s117-hadoop-lab3.idc.vipshop.com/10.19.111.117:8020
throw new IOException("Invalid configuration: a shared edits dir " +
"must not be specified if HA is not enabled.");
}
HAUtil.usesSharedEditsDir方法：
public static boolean usesSharedEditsDir(Configuration conf) {
return null != conf.get(DFS_NAMENODE_SHARED_EDITS_DIR_KEY);
}
判断jn的edit dir设置，如果设置了dfs.namenode.shared.edits.dir就会抛出异常
去掉下面的设置即可：
<property> <name>dfs.namenode.shared.edits.dir</name> <value>qjournal://xxxxx/bipcluster</value> </property>
2.rollback的时候要先把ha的配置更改为非ha的配置，然后进行rollback
rollback之后，再重做ha
重做ha的步骤简单如下：
1).关闭整个集群(非ha的),更改配置为ha的配置，备份原来的standby nn和jn的数据目录
2).删除旧的jn数据，并单独启动jn
./hadoop-daemon.sh start journalnode
3).在active nn上运行
hdfs namenode -initializeSharedEdits
namenode通过initializeSharedEdits命令来初始化journalnode，把edits文件共享到journalnode上
4).运行active nn
./hadoop-daemon.sh start namenode
5).在standby的nn上运行
hadoop-daemon.sh start namenode -bootstrapStandby hadoop-daemon.sh start namenode
同步元数据并启动standby namenode
6).启动所有的dn
./hadoop-daemons.sh start datanode
7).nn transitionToActive
hdfs haadmin -transitionToActive nn1 hdfs haadmin -getServiceState nn1
3.start-dfs.sh 这个脚本有bug，在传入-rollback时，只能dn可以rollback,nn不能rollback
diff ../../hadoop-2.5.0-cdh5.2.0/sbin/start-dfs.sh start-dfs.sh 50c50 < nameStartOpt="$nameStartOpt [email protected]" --- > nameStartOpt="$nameStartOpts [email protected]"
也可以直接通过
sh -x ./hadoop-daemon.sh start namenode -rollback
命令rollback nn，注意dn还是要rollback的
4.如果rollback或upgrade失败，可以通过之前的备份元数据进行覆盖恢复