> `HDFS_DELEGATION_TOKEN` 这个BUG在很多文章中都出现着,讲了很多原理,但是只给出了官方引用地扯,完全没有给出如何解决,我们线上的业务就有着这样的问题,7天一到马上出现这问题了,官方明明说这个bug修复了呀,因为我们使用的版本是比较新的,理论上不会有这样的问题才对,可是偏偏就有了,没办法,只能硬上了,花了两天的时间找到了解决这个问题的办法,下面会还原这个错误及给出解决方案。
## 版本列表
![](https://upload-images.jianshu.io/upload_images/9028759-35c1bf0606261dc5.png?imageMogr2/auto-orient/strip%7CimageView2/2/w/1240)
## 测试用例配置
添加 hdfs-site.xml 配置
```
dfs.namenode.delegation.key.update-interval=60000 #1分钟
dfs.namenode.delegation.token.max-lifetime=180000 #3分钟
dfs.namenode.delegation.token.renew-interval=60000 #1分钟
```
修改 /etc/krb5.conf ticket过期为1小时
```
...
ticket_lifetime = 1h
...
```
代码内有`kerberos`认证
```
class App {
System.setProperty("java.security.krb5.conf", "/etc/krb5.conf")
System.setProperty("sun.security.krb5.debug", "false")
val hConf = new Configuration
hConf.addResource("hbase-site.xml")
UserGroupInformation.setConfiguration(hConf)
UserGroupInformation.loginUserFromKeytab("[email protected]", "/etc/security/keytabs/hbase.headless.keytab")
val sparkConf = new SparkConf()
// .setMaster("local[12]")
.setAppName("HDFS_DELEGATION_TOKEN")
val spark = SparkSession
.builder
.config(sparkConf)
.getOrCreate()
hConf.set("hbase.mapreduce.inputtable", "test_log")
def run(args: Array[String]): Unit = {
val sc = spark.sparkContext
import spark.implicits._
val userRDD: RDD[Log] = sc.newAPIHadoopRDD(
hConf,
classOf[TableInputFormat],
classOf[ImmutableBytesWritable],
classOf[Result]
).flatMap {
rdd => {
val map = HbaseUtil.result2Map(rdd._2)
val log = Log(
map.get("uid")
)
Array(log)
}
}
userRDD.toDS().cache().createTempView("log")
spark.sql(
"""select * from log""".stripMargin)
.show(false)
spark.catalog.dropTempView("log")
userRDD.unpersist()
}
}
case class Log(uid: String)
object App {
def main(args: Array[String]): Unit = {
val app = new App()
while (true) {
app.run(args)
TimeUnit.MINUTES.sleep(3)
}
}
}
```
## 测试百度跟谷歌中最最最多出现的解决方案
```
--conf spark.hadoop.fs.hdfs.impl.disable.cache=true
--conf mapreduce.job.complete.cancel.delegation.tokens=false
```
1. 测试提交
```
spark-submit --master yarn \
--class com.dounine.hbase.App \
--executor-memory 1g \
--driver-memory 1g \
--keytab /etc/security/keytabs/hbase.headless.keytab \
--principal [email protected] \
build/libs/hdfs-token-1.0.0-SNAPSHOT-all.jar
```
2. 测试提交
```
...
--conf spark.hadoop.fs.hdfs.impl.disable.cache=true \
--conf mapreduce.job.complete.cancel.delegation.tokens=false \
...
```
3. 测试提交
```
...
--conf mapreduce.job.complete.cancel.delegation.tokens=false \
...
```
4. 测试提交
```
...
--conf spark.hadoop.fs.hdfs.impl.disable.cache=true \
...
```
1,2,3,4 测试结果
时间观察3分钟 => **正常**
时间观察10分钟 => **正常**
时间观察30分钟 => **正常**
时间观察60分钟 => **正常**
时间观察120分钟 => **正常**
**测试结论 => 与1、2、3、4 --conf 配置无关**
*好吧,我已经怀疑人生、可能是我打开的方式不对*
## 继续测试
将认证代码放入run方法内
```
def run(args: Array[String]): Unit = {
System.setProperty("java.security.krb5.conf", "/etc/krb5.conf")
System.setProperty("sun.security.krb5.debug", "false")
val hConf = new Configuration
hConf.addResource("hbase-site.xml")
UserGroupInformation.setConfiguration(hConf)
UserGroupInformation.loginUserFromKeytab("[email protected]", "/etc/security/keytabs/hbase.headless.keytab")
val sparkConf = new SparkConf()
// .setMaster("local[12]")
.setAppName("HDFS_DELEGATION_TOKEN")
val spark = SparkSession
.builder
.config(sparkConf)
.getOrCreate()
hConf.set("hbase.mapreduce.inputtable", "test_log")
....
```
时间观察3分钟 => 正常
时间观察6分钟 => 异常
```
18/12/29 16:50:31 ERROR AsyncEventQueue: Listener EventLoggingListener threw an exception
org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.security.token.SecretManager$InvalidToken): token (token for hbase: HDFS_DELEGATION_TOKEN owner
[email protected], renewer=yarn, realUser=, issueDate=1546072104965, maxDate=1546072704965, sequenceNumber=15985, masterKeyId=748) is expired, curr
ent time: 2018-12-29 16:32:29,829+0800 expected renewal time: 2018-12-29 16:31:24,965+0800
at org.apache.hadoop.ipc.Client.getRpcResponse(Client.java:1497)
at org.apache.hadoop.ipc.Client.call(Client.java:1443)
at org.apache.hadoop.ipc.Client.call(Client.java:1353)
at org.apache.hadoop.ipc.ProtobufRpcEngine$Invoker.invoke(ProtobufRpcEngine.java:228)
at org.apache.hadoop.ipc.ProtobufRpcEngine$Invoker.invoke(ProtobufRpcEngine.java:116)
at com.sun.proxy.$Proxy11.fsync(Unknown Source)
at org.apache.hadoop.hdfs.protocolPB.ClientNamenodeProtocolTranslatorPB.fsync(ClientNamenodeProtocolTranslatorPB.java:980)
at sun.reflect.GeneratedMethodAccessor11.invoke(Unknown Source)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
at java.lang.reflect.Method.invoke(Method.java:498)
at org.apache.hadoop.io.retry.RetryInvocationHandler.invokeMethod(RetryInvocationHandler.java:422)
at org.apache.hadoop.io.retry.RetryInvocationHandler$Call.invokeMethod(RetryInvocationHandler.java:165)
at org.apache.hadoop.io.retry.RetryInvocationHandler$Call.invoke(RetryInvocationHandler.java:157)
at org.apache.hadoop.io.retry.RetryInvocationHandler$Call.invokeOnce(RetryInvocationHandler.java:95)
at org.apache.hadoop.io.retry.RetryInvocationHandler.invoke(RetryInvocationHandler.java:359)
at com.sun.proxy.$Proxy12.fsync(Unknown Source)
...
```
## 问题发现
通过几十遍不断的调整位置、最终确认的问题所在
还有Exector的问题
```
UserGroupInformation.setConfiguration(hConf)
UserGroupInformation.loginUserFromKeytab("[email protected]", "/etc/security/keytabs/hbase.headless.keytab")
```
是由于以上两句kerberos认证代码导致的结果
跟下面的配置冲突了
```
--principal $principal --keytab $keytab
```
## 解决方案(一)
删除掉下面代码中的这两句认证即可,使用`--principal $principal --keytab $keytab`
```
UserGroupInformation.setConfiguration(hConf)
UserGroupInformation.loginUserFromKeytab("[email protected]", "/etc/security/keytabs/hbase.headless.keytab")
```
因为Spark的`--principal --keytab`会在令牌即将过期的时候帮我们重新续定,如果代码里面加上之后,Spark会读取到ApplicationMaster中用户已经认证了,没有过期是不会续定NodeManager中的Exector的。
如果是开发环境模式,可以加一个判断使用以上两句代码,**简单粗暴**
## 解决方案(二)
使用UserGroupInformation的进程认证方式
```
spark.sparkContext
.parallelize(0 to 1000)
.repartition(10)
.foreachPartition {
iter => {
val hConf = new Configuration
hConf.addResource("hbase-site.xml")
val ugi = UserGroupInformation.loginUserFromKeytabAndReturnUGI("[email protected]", "/etc/security/keytabs/hbase.headless.keytab")
ugi.doAs(new PrivilegedAction[Unit] {//在每个Partition认证
override def run(): Unit = {
val logDir = new Path(args(0))
val fs = FileSystem.get(hConf)
if (!fs.exists(logDir)) throw new Exception(logDir.toUri.getPath + " director not exist.")
while (iter.hasNext) {
iter.next()
val logPaths = fs.listFiles(logDir, false)
TimeUnit.MILLISECONDS.sleep(10)
}
}
})
}
}
```
## BUG 7 天后再次出现
上面推导还是有问题,还有望知道BUG解决的小伙伴告知一下。
## 临时解决方案
就是加大过期的时间
---
![](https://upload-images.jianshu.io/upload_images/9028759-a3ea1ec0784f9946.png?imageMogr2/auto-orient/strip%7CimageView2/2/w/1240)