【背景】
前一段时间总结了hadoop中的token认证、yarn任务运行中的token，其中也都提到了delegation token。而最近也遇到了一个问题，问题现象是：flink任务运行超过七天后，由于宿主机异常导致任务失败，继而触发任务的重试，但接连重试几次都是失败的，并且任务的日志也没有聚合，导致无法分析问题失败的原因。最后发现是和delegation token有关，本文就来总结下相关的原理。
【原理】
1. 什么是delegation token
先简单描述下为什么需要delegation token。在开启kerberos之后，服务之间交互前，都需要先向KDC认证获取对应的票据。而在一个yarn任务运行过程中可能会产生很多任务container，每个这样的任务container都可能会访问hdfs，由于访问前需要先获取票据来进行认证，那么这个时候KDC就很容易成为性能瓶颈。delegation token（委派token）就是为了减少不必要的认证工作而出现的。
2. delegation token在任务提交运行过程中的使用
任务提交运行过程中，delegation token相关的流程如下图所示：
1）首先，RM启动后，内部会创建一个服务线程专门用于处理token的更新
代码语言：javascript复制// ResourceManager.java
protected void serviceInit(Configuration configuration) throws Exception {
...
if (UserGroupInformation.isSecurityEnabled()) {
delegationTokenRenewer = createDelegationTokenRenewer();
rmContext.setDelegationTokenRenewer(delegationTokenRenewer);
}
....
}
protected DelegationTokenRenewer createDelegationTokenRenewer() {
return new DelegationTokenRenewer();
}2）客户端申请delegation token
客户端在提交任务前，通常需要先向hdfs上传资源文件（包括运行所需的jar包等），在此过程中会向nn申请一个delegation token，并放到任务启动上下文中，然后向rm发送提交任务请求（请求中包含任务的启动上下文）。
下面是flink on yarn提交任务时的代码片段：
代码语言：javascript复制// flink YarnClusterDescriptor.java
private ApplicationReport startAppMaster(...){
// 开启kerberos的情况下，获取token
if (UserGroupInformation.isSecurityEnabled()) {
// set HDFS delegation tokens when security is enabled
LOG.info("Adding delegation token to the AM container.");
List<Path> yarnAccessList =
ConfigUtils.decodeListFromConfig(
configuration, YarnConfigOptions.YARN_ACCESS, Path::new);
Utils.setTokensFor(
amContainer,
ListUtils.union(yarnAccessList, fileUploader.getRemotePaths()),
yarnConfiguration);
}
}
public static void setTokensFor(
ContainerLaunchContext amContainer, List<Path> paths, Configuration conf)
throws IOException {
Credentials credentials = new Credentials();
// for HDFS
TokenCache.obtainTokensForNamenodes(credentials, paths.toArray(new Path[0]), conf);
// for HBase
obtainTokenForHBase(credentials, conf);
// for user
UserGroupInformation currUsr = UserGroupInformation.getCurrentUser();
// 获取到的token 放到启动上下文中
Collection<Token<? extends TokenIdentifier>> usrTok = currUsr.getTokens();
for (Token<? extends TokenIdentifier> token : usrTok) {
final Text id = new Text(token.getIdentifier());
LOG.info("Adding user token " id " with " token);
credentials.addToken(id, token);
}
try (DataOutputBuffer dob = new DataOutputBuffer()) {
credentials.writeTokenStorageToStream(dob);
if (LOG.isDebugEnabled()) {
LOG.debug("Wrote tokens. Credentials buffer length: " dob.getLength());
}
ByteBuffer securityTokens = ByteBuffer.wrap(dob.getData(), 0, dob.getLength());
amContainer.setTokens(securityTokens);
}
}
// TokenCache.java
// 调用hadoop的接口 向nn请求token
public static void obtainTokensForNamenodes(
Credentials credentials,
Path[] ps, Configuration conf)
throws IOException {
if (!UserGroupInformation.isSecurityEnabled()) {
return;
}
obtainTokensForNamenodesInternal(credentials, ps, conf);
}
static void obtainTokensForNamenodesInternal(
Credentials credentials,
Path[] ps,
Configuration conf)
throws IOException {
Set<FileSystem> fsSet = new HashSet<FileSystem>();
for (Path p : ps) {
fsSet.add(p.getFileSystem(conf));
}
String masterPrincipal = Master.getMasterPrincipal(conf);
for (FileSystem fs : fsSet) {
obtainTokensForNamenodesInternal(fs, credentials, conf, masterPrincipal);
}
}
static void obtainTokensForNamenodesInternal(
FileSystem fs,
Credentials credentials,
Configuration conf,
String renewer)
throws IOException {
...
final Token<?> tokens[] = fs.addDelegationTokens(delegTokenRenewer, credentials);
...
}
// FileSystem.java
public Token<?>[] addDelegationTokens(
final String renewer, Credentials credentials)
throws IOException {
if (credentials == null) {
credentials = new Credentials();
}
final List<Token<?>> tokens = new ArrayList<>();
collectDelegationTokens(renewer, credentials, tokens);
return tokens.toArray(new Token<?>[tokens.size()]);
}
private void collectDelegationTokens(
final String renewer,
final Credentials credentials,
final List<Token<?>> tokens)
throws IOException {
final String serviceName = getCanonicalServiceName();
// Collect token of the this filesystem and then of its embedded children
if (serviceName != null) { // fs has token, grab it
final Text service = new Text(serviceName);
Token<?> token = credentials.getToken(service);
if (token == null) {
// 向NN 请求delegation token
token = getDelegationToken(renewer);
if (token != null) {
tokens.add(token);
credentials.addToken(service, token);
}
}
}
...
}3）RM将token添加到delegation token更新服务中
RM在处理客户端提交任务请求时，判断是否启用kerberos认证，如果启用则从任务启动上下文中解析出delegation token，并添加到delegation token更新服务中。在该服务中，会启动线程定时对delegation token进行更新。此后，继续向NM发送启动container的请求，delegation token则随启动上下文被带到NM中。
代码语言：javascript复制// RMAppManager.java
protected void submitApplication(
ApplicationSubmissionContext submissionContext,
long submitTime,
String user)
throws YarnException {
...
if (UserGroupInformation.isSecurityEnabled()) {
this.rmContext.getDelegationTokenRenewer().addApplicationAsync(
applicationId,
BuilderUtils.parseCredentials(submissionContext),
submissionContext.getCancelTokensWhenComplete(),
application.getUser(),
BuilderUtils.parseTokensConf(submissionContext));
}
...
}4）NM使用delegation token
NM收到启动container的请求后，从请求（任务启动上下文）中解析出delegation token，并为该container构造一个对应的实例对象，同时将delegation token保存在该实例对象中，然后为该container进行资源本地化，即从hdfs中下载必须的资源文件，这里就会用到传递过来的delegation token。同时在任务结束时，如果需要进行任务日志聚合，仍旧会使用该delegation token将任务的日志上传到hdfs的指定路径。
另外，delegation token还会写入到持久化文件中，一方面用于NM的异常恢复，另一方面是将token传递给任务container进程以供使用。
3. delegation token的更新与生命周期
1）申请token时已经指定了token的最大生命周期
代码语言：javascript复制// FSNamesystem.java
Token<DelegationTokenIdentifier> getDelegationToken(Text renewer) throws IOException {
...
DelegationTokenIdentifier dtId = new DelegationTokenIdentifier(owner, renewer, realUser);
token = new Token<DelegationTokenIdentifier>(dtId, dtSecretManager);
...
return token;
}
// Token.java
public Token(T id, SecretManager<T> mgr) {
password = mgr.createPassword(id);
identifier = id.getBytes();
kind = id.getKind();
service = new Text();
}
// AbstractDelegationTokenSecretManager
protected synchronized byte[] createPassword(TokenIdent identifier) {
long now = Time.now();
identifier.setMaxDate(now tokenMaxLifetime);
...
}2）RM接收到任务提交请求后，先进行一次更新得到token的下次超时时间，然后再根据超时时间设置定时器时间触发进行更新。
代码语言：javascript复制public void addApplicationSync(
ApplicationId applicationId,
Credentials ts,
boolean shouldCancelAtEnd,
String user)
throws IOException, InterruptedException {
handleAppSubmitEvent(
new DelegationTokenRenewerAppSubmitEvent(
applicationId, ts, shouldCancelAtEnd, user, new Configuration()));
}
private void handleAppSubmitEvent(AbstractDelegationTokenRenewerAppEvent evt)
throws IOException, InterruptedException {
...
Credentials ts = evt.getCredentials();
Collection<Token<?>> tokens = ts.getAllTokens();
for (Token<?> token : tokens) {
DelegationTokenToRenew dttr = allTokens.get(token);
if (dttr == null) {
dttr = new DelegationTokenToRenew(
Arrays.asList(applicationId),
token, tokenConf, now, shouldCancelAtEnd,
evt.getUser());
try {
// 先进行一次更新
renewToken(dttr)
} catch (IOException ioe) {
...
}
}
tokenList.add(dttr);
}
if (!tokenList.isEmpty()) {
for (DelegationTokenToRenew dtr : tokenList) {
DelegationTokenToRenew currentDtr = allTokens.putIfAbsent(dtr.token, dtr);
if (currentDtr != null) {
// another job beat us
currentDtr.referringAppIds.add(applicationId);
appTokens.get(applicationId).add(currentDtr);
} else {
appTokens.get(applicationId).add(dtr);
setTimerForTokenRenewal(dtr);
}
}
}
}
protected void renewToken(final DelegationTokenToRenew dttr)
throws IOException {
// need to use doAs so that http can find the kerberos tgt
// NOTE: token renewers should be responsible for the correct UGI!
try {
// 更新delegation token 并得到下次超时时间
dttr.expirationDate =
UserGroupInformation.getLoginUser().doAs(
new PrivilegedExceptionAction<Long>() {
@Override
public Long run() throws Exception {
return dttr.token.renew(dttr.conf);
}
});
} catch (InterruptedException e) {
throw new IOException(e);
}
LOG.info("Renewed delegation-token= [" dttr "]");
}
protected void setTimerForTokenRenewal(DelegationTokenToRenew token)
throws IOException {
// calculate timer time
long expiresIn = token.expirationDate - System.currentTimeMillis();
if (expiresIn <= 0) {
LOG.info("Will not renew token " token);
return;
}
long renewIn = token.expirationDate - expiresIn / 10; // little bit before the expiration
// need to create new task every time
RenewalTimerTask tTask = new RenewalTimerTask(token);
token.setTimerTask(tTask); // keep reference to the timer
renewalTimer.schedule(token.timerTask, new Date(renewIn));
LOG.info(
"Renew " token " in " expiresIn " ms, appId = "
token.referringAppIds);
}再来看更新token的请求与处理细节：
代码语言：javascript复制// 客户端发送更新请求
public long renew(Token<?> token, Configuration conf) throws IOException {
Token<DelegationTokenIdentifier> delToken = (Token<DelegationTokenIdentifier>) token;
ClientProtocol nn = getNNProxy(delToken, conf);
try {
return nn.renewDelegationToken(delToken);
} catch (RemoteException re) {
throw re.unwrapRemoteException(InvalidToken.class,
AccessControlException.class);
}
}
// 服务端的响应处理
long renewDelegationToken(Token<DelegationTokenIdentifier> token)
throws InvalidToken, IOException {
try {
...
expiryTime = dtSecretManager.renewToken(token, renewer);
} catch (AccessControlException ace) {
...
}
return expiryTime;
}
public synchronized long renewToken(
Token<TokenIdent> token,
String renewer)
throws InvalidToken, IOException {
ByteArrayInputStream buf = new ByteArrayInputStream(token.getIdentifier());
DataInputStream in = new DataInputStream(buf);
TokenIdent id = createIdentifier();
id.readFields(in);
LOG.info(
"Token renewal for identifier: " formatTokenId(id)
"; total currentTokens " currentTokens.size());
long now = Time.now();
if (id.getMaxDate() < now) {
throw new InvalidToken(
renewer " tried to renew an expired token "
formatTokenId(id) " max expiration date: "
Time.formatTime(id.getMaxDate())
" currentTime: " Time.formatTime(now));
}
if ((id.getRenewer() == null) || (id.getRenewer().toString().isEmpty())) {
throw new AccessControlException(
renewer
" tried to renew a token " formatTokenId(id)
" without a renewer");
}
if (!id.getRenewer().toString().equals(renewer)) {
throw new AccessControlException(
renewer
" tries to renew a token " formatTokenId(id)
" with non-matching renewer " id.getRenewer());
}
DelegationKey key = getDelegationKey(id.getMasterKeyId());
if (key == null) {
throw new InvalidToken(
"Unable to find master key for keyId="
id.getMasterKeyId()
" from cache. Failed to renew an unexpired token "
formatTokenId(id) " with sequenceNumber="
id.getSequenceNumber());
}
byte[] password = createPassword(token.getIdentifier(), key.getKey());
if (!MessageDigest.isEqual(password, token.getPassword())) {
throw new AccessControlException(
renewer
" is trying to renew a token "
formatTokenId(id) " with wrong password");
}
long renewTime = Math.min(id.getMaxDate(), now tokenRenewInterval);
String trackingId = getTrackingIdIfEnabled(id);
DelegationTokenInformation info =
new DelegationTokenInformation(renewTime, password, trackingId);
if (getTokenInfo(id) == null) {
throw new InvalidToken(
"Renewal request for unknown token " formatTokenId(id));
}
updateToken(id, info);
return renewTime;
}3）token达到最大生命周期的处理
在定时器中，会捕获更新抛出的异常，并直接移除失效的token。
但是注意：在每次更新之前，会按需重新申请新的delegation token（后面再展开讲解）
代码语言：javascript复制public void run() {
if (cancelled.get()) {
return;
}
Token<?> token = dttr.token;
try {
// 先判断是否需要申请新的token
requestNewHdfsDelegationTokenIfNeeded(dttr);
// if the token is not replaced by a new token, renew the token
if (!dttr.isTimerCancelled()) {
renewToken(dttr);
setTimerForTokenRenewal(dttr);// set the next one
} else {
LOG.info("The token was removed already. Token = [" dttr "]");
}
} catch (Exception e) {
LOG.error("Exception renewing token" token ". Not rescheduled", e);
removeFailedDelegationToken(dttr);
}
}【问题分析】
来看看前面问题失败的相关日志，复盘分析下。
首先从NM的日志中发现任务在重试时，因为无法下载资源（到本地）导致无法启动任务，而下载资源失败的原因则是因为无效的token。
代码语言：javascript复制2022-07-18 13:44:18,665 WARN org.apache.hadoop.ipc.Client: Exception encountered while connecting to the server : org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.security.token.SecretManager$InvalidToken): token (HDFS_DELEGATION_TOKEN token 4361 for hncscwc) can't be found in cache
2022-07-18 13:44:18,669 WARN org.apache.hadoop.yarn.server.nodemanager.containermanager.localizer.ResourceLocalizationService: { hdfs://hdfsHACluster/user/hncscwc/.flink/application_1637733238080_3800/application_1637733238080_38002636034628721129021.tmp, 1656925873322, FILE, null } failed: token (HDFS_DELEGATION_TOKEN token 4361 for hncscwc) can't be found in cache
org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.security.token.SecretManager$InvalidToken): token (HDFS_DELEGATION_TOKEN token 4361 for hncscwc) can't be found in cache
at org.apache.hadoop.ipc.Client.getRpcResponse(Client.java:1486)
at org.apache.hadoop.ipc.Client.call(Client.java:1432)
at org.apache.hadoop.ipc.Client.call(Client.java:1342)
at org.apache.hadoop.ipc.ProtobufRpcEngine$Invoker.invoke(ProtobufRpcEngine.java:227)
at org.apache.hadoop.ipc.ProtobufRpcEngine$Invoker.invoke(ProtobufRpcEngine.java:116)
at com.sun.proxy.$Proxy15.getFileInfo(Unknown Source)
at org.apache.hadoop.hdfs.protocolPB.ClientNamenodeProtocolTranslatorPB.getFileInfo(ClientNamenodeProtocolTranslatorPB.java:796)
at sun.reflect.GeneratedMethodAccessor172.invoke(Unknown Source)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
at java.lang.reflect.Method.invoke(Method.java:498)
at org.apache.hadoop.io.retry.RetryInvocationHandler.invokeMethod(RetryInvocationHandler.java:411)
at org.apache.hadoop.io.retry.RetryInvocationHandler$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:348)
at com.sun.proxy.$Proxy16.getFileInfo(Unknown Source)
at org.apache.hadoop.hdfs.DFSClient.getFileInfo(DFSClient.java:1649)
at org.apache.hadoop.hdfs.DistributedFileSystem$27.doCall(DistributedFileSystem.java:1440)
at org.apache.hadoop.hdfs.DistributedFileSystem$27.doCall(DistributedFileSystem.java:1437)
at org.apache.hadoop.fs.FileSystemLinkResolver.resolve(FileSystemLinkResolver.java:81)
at org.apache.hadoop.hdfs.DistributedFileSystem.getFileStatus(DistributedFileSystem.java:1452)
at org.apache.hadoop.yarn.util.FSDownload.copy(FSDownload.java:253)
at org.apache.hadoop.yarn.util.FSDownload.access$000(FSDownload.java:63)
at org.apache.hadoop.yarn.util.FSDownload$2.run(FSDownload.java:361)
at org.apache.hadoop.yarn.util.FSDownload$2.run(FSDownload.java:359)
at java.security.AccessController.doPrivileged(Native Method)
at javax.security.auth.Subject.doAs(Subject.java:422)
at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1922)
at org.apache.hadoop.yarn.util.FSDownload.call(FSDownload.java:359)
at org.apache.hadoop.yarn.util.FSDownload.call(FSDownload.java:62)
at java.util.concurrent.FutureTask.run(FutureTask.java:266)
at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
at java.util.concurrent.FutureTask.run(FutureTask.java:266)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
at java.lang.Thread.run(Thread.java:748)为什么会出现无效的token，接着再看RM的日志。
代码语言：javascript复制2022-07-04 17:11:13,400 INFO org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler: Application 'application_1637733238080_3800' is submitted without priority hence considering default queue/cluster priority: 0
2022-07-04 17:11:13,424 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renewed delegation-token= [Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657012273422; apps=[application_1637733238080_3800]]
2022-07-05 14:47:13,462 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renewed delegation-token= [Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657090033446; apps=[application_1637733238080_3800]]
2022-07-06 12:23:13,467 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renewed delegation-token= [Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657167793465; apps=[application_1637733238080_3800]]
2022-07-07 09:59:13,487 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renewed delegation-token= [Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657245553484; apps=[application_1637733238080_3800]]
2022-07-08 07:35:13,532 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renewed delegation-token= [Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657323313511; apps=[application_1637733238080_3800]]
2022-07-09 05:11:13,551 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renewed delegation-token= [Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657401073532; apps=[application_1637733238080_3800]]
2022-07-10 02:47:13,564 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renewed delegation-token= [Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657478833547; apps=[application_1637733238080_3800]]
2022-07-11 00:23:13,591 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renewed delegation-token= [Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657530673393; apps=[application_1637733238080_3800]]
2022-07-11 17:11:07,361 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renewed delegation-token= [Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657530673393; apps=[application_1637733238080_3800]]
2022-07-11 17:11:07,361 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renew Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657530673393; apps=[application_1637733238080_3800] in 6032 ms, appId = [application_1637733238080_3800]
2022-07-11 17:11:12,793 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renewed delegation-token= [Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657530673393; apps=[application_1637733238080_3800]]
2022-07-11 17:11:12,793 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renew Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657530673393; apps=[application_1637733238080_3800] in 600 ms, appId = [application_1637733238080_3800]
2022-07-11 17:11:13,337 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renewed delegation-token= [Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657530673393; apps=[application_1637733238080_3800]]
2022-07-11 17:11:13,337 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renew Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657530673393; apps=[application_1637733238080_3800] in 56 ms, appId = [application_1637733238080_3800]
2022-07-11 17:11:13,391 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renewed delegation-token= [Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657530673393; apps=[application_1637733238080_3800]]
2022-07-11 17:11:13,391 INFO org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Renew Kind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc);exp=1657530673393; apps=[application_1637733238080_3800] in 2 ms, appId = [application_1637733238080_3800]
2022-07-11 17:11:13,398 ERROR org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: Exception renewing tokenKind: HDFS_DELEGATION_TOKEN, Service: ha-hdfs:hdfsHACluster, Ident: (HDFS_DELEGATION_TOKEN token 4361 for hncscwc). Not rescheduled
org.apache.hadoop.security.token.SecretManager$InvalidToken: hadoop tried to renew an expired token (HDFS_DELEGATION_TOKEN token 4361 for hncscwc) max expiration date: 2022-07-11 17:11:13,393 0800 currentTime: 2022-07-11 17:11:13,394 0800
at org.apache.hadoop.security.token.delegation.AbstractDelegationTokenSecretManager.renewToken(AbstractDelegationTokenSecretManager.java:499)
at org.apache.hadoop.hdfs.server.namenode.FSNamesystem.renewDelegationToken(FSNamesystem.java:5952)
at org.apache.hadoop.hdfs.server.namenode.NameNodeRpcServer.renewDelegationToken(NameNodeRpcServer.java:675)
at org.apache.hadoop.hdfs.protocolPB.ClientNamenodeProtocolServerSideTranslatorPB.renewDelegationToken(ClientNamenodeProtocolServerSideTranslatorPB.java:1035)
at org.apache.hadoop.hdfs.protocol.proto.ClientNamenodeProtocolProtos$ClientNamenodeProtocol$2.callBlockingMethod(ClientNamenodeProtocolProtos.java)
at org.apache.hadoop.ipc.ProtobufRpcEngine$Server$ProtoBufRpcInvoker.call(ProtobufRpcEngine.java:447)
at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:989)
at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:850)
at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:793)
at java.security.AccessController.doPrivileged(Native Method)
at javax.security.auth.Subject.doAs(Subject.java:422)
at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1922)
at org.apache.hadoop.ipc.Server$Handler.run(Server.java:2489)
at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method)
at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:62)
at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45)
at java.lang.reflect.Constructor.newInstance(Constructor.java:423)
at org.apache.hadoop.ipc.RemoteException.instantiateException(RemoteException.java:121)
at org.apache.hadoop.ipc.RemoteException.unwrapRemoteException(RemoteException.java:88)
at org.apache.hadoop.hdfs.DFSClient$Renewer.renew(DFSClient.java:761)
at org.apache.hadoop.security.token.Token.renew(Token.java:458)
at org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer$1.run(DelegationTokenRenewer.java:601)
at org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer$1.run(DelegationTokenRenewer.java:598)
at java.security.AccessController.doPrivileged(Native Method)
at javax.security.auth.Subject.doAs(Subject.java:422)
at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1922)
at org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer.renewToken(DelegationTokenRenewer.java:597)
at org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer$RenewalTimerTask.run(DelegationTokenRenewer.java:531)
at java.util.TimerThread.mainLoop(Timer.java:555)
at java.util.TimerThread.run(Timer.java:505)
Caused by: org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.security.token.SecretManager$InvalidToken): hadoop tried to renew an expired token (HDFS_DELEGATION_TOKEN token 4361 for hncscwc) max expiration date: 2022-07-11 17:11:13,393 0800 currentTime: 2022-07-11 17:11:13,394 0800
at org.apache.hadoop.security.token.delegation.AbstractDelegationTokenSecretManager.renewToken(AbstractDelegationTokenSecretManager.java:499)
at org.apache.hadoop.hdfs.server.namenode.FSNamesystem.renewDelegationToken(FSNamesystem.java:5952)
at org.apache.hadoop.hdfs.server.namenode.NameNodeRpcServer.renewDelegationToken(NameNodeRpcServer.java:675)
at org.apache.hadoop.hdfs.protocolPB.ClientNamenodeProtocolServerSideTranslatorPB.renewDelegationToken(ClientNamenodeProtocolServerSideTranslatorPB.java:1035)
at org.apache.hadoop.hdfs.protocol.proto.ClientNamenodeProtocolProtos$ClientNamenodeProtocol$2.callBlockingMethod(ClientNamenodeProtocolProtos.java)
at org.apache.hadoop.ipc.ProtobufRpcEngine$Server$ProtoBufRpcInvoker.call(ProtobufRpcEngine.java:447)
at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:989)
at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:850)
at org.apache.hadoop.ipc.Server$RpcCall.run(Server.java:793)
at java.security.AccessController.doPrivileged(Native Method)
at javax.security.auth.Subject.doAs(Subject.java:422)
at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1922)
at org.apache.hadoop.ipc.Server$Handler.run(Server.java:2489)
at org.apache.hadoop.ipc.Client.getRpcResponse(Client.java:1486)
at org.apache.hadoop.ipc.Client.call(Client.java:1432)
at org.apache.hadoop.ipc.Client.call(Client.java:1342)
at org.apache.hadoop.ipc.ProtobufRpcEngine$Invoker.invoke(ProtobufRpcEngine.java:227)
at org.apache.hadoop.ipc.ProtobufRpcEngine$Invoker.invoke(ProtobufRpcEngine.java:116)
at com.sun.proxy.$Proxy94.renewDelegationToken(Unknown Source)
at
org.apache.hadoop.hdfs.protocolPB.ClientNamenodeProtocolTranslatorPB.renewDelegationToken(ClientNamenodeProtocolTranslatorPB.java:964)
at sun.reflect.GeneratedMethodAccessor277.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:411)
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$Call.invokeOnce(RetryInvocationHandler.java:95)
at org.apache.hadoop.io.retry.RetryInvocationHandler.invoke(RetryInvocationHandler.java:348)
at com.sun.proxy.$Proxy95.renewDelegationToken(Unknown Source)
at org.apache.hadoop.hdfs.DFSClient$Renewer.renew(DFSClient.java:759)
... 10 more
2022-07-11 17:11:13,399 ERROR org.apache.hadoop.yarn.server.resourcemanager.security.DelegationTokenRenewer: removing failed delegation token for appid=[application_1637733238080_3800];t=ha-hdfs:hdfsHACluster从上面的日志可以看到，任务从提交后，delegation token每天都有在更新，然而运行到第7天后，更新失败而失效。失效后，NN内部会删除无效的token，此时如果任务失败需要重试，或者任务结束需要进行日志聚合，都会继续使用该无效的token来操作hdfs，最终结果就是在NN中找不到对应的token而抛异常导致失败。
【问题解决】
要解决该问题，一种最简单直接的办法就是加大delegation token的最大生命周期时间。
但一开始觉得该办法略有些low，尤其对于flink长周期运行的实时任务的场景，是无法确定任务的运行时长的，因此也就无法确定设置token的最大生命周期。
因此，再次分析了源码，发现RM中对于将要过期（超过最大生命周期）的delegation token，会按需重新申请一个新的token，也就是定时器线程中token更新之前的requestNewHdfsDelegationTokenIfNeeded方法。
来看看具体的实现逻辑：
代码语言：javascript复制private void requestNewHdfsDelegationTokenIfNeeded(
final DelegationTokenToRenew dttr)
throws IOException, InterruptedException {
// 拥有特权 并且 token类型为委派token 并且 快到最大生命周期
if (hasProxyUserPrivileges &&
dttr.maxDate - dttr.expirationDate < credentialsValidTimeRemaining &&
dttr.token.getKind().equals(HDFS_DELEGATION_KIND)) {
final Collection<ApplicationId> applicationIds;
synchronized (dttr.referringAppIds) {
applicationIds = new HashSet<>(dttr.referringAppIds);
dttr.referringAppIds.clear();
}
// remove all old expiring hdfs tokens for this application.
for (ApplicationId appId : applicationIds) {
Set<DelegationTokenToRenew> tokenSet = appTokens.get(appId);
if (tokenSet == null || tokenSet.isEmpty()) {
continue;
}
Iterator<DelegationTokenToRenew> iter = tokenSet.iterator();
synchronized (tokenSet) {
while (iter.hasNext()) {
DelegationTokenToRenew t = iter.next();
if (t.token.getKind().equals(HDFS_DELEGATION_KIND)) {
iter.remove();
allTokens.remove(t.token);
t.cancelTimer();
LOG.info("Removed expiring token " t);
}
}
}
}
LOG.info("Token= (" dttr ") is expiring, request new token.");
requestNewHdfsDelegationTokenAsProxyUser(
applicationIds, dttr.user,
dttr.shouldCancelAtEnd);
}
}申请到新的token之后，会在RM内部进行更新，然后通过NM的心跳响应同步给NM。
代码语言：javascript复制private void requestNewHdfsDelegationTokenAsProxyUser(
...
// Get new hdfs tokens for this user
Credentials credentials = new Credentials();
Token<?>[] newTokens = obtainSystemTokensForUser(user, credentials);
DataOutputBuffer dob = new DataOutputBuffer();
credentials.writeTokenStorageToStream(dob);
ByteBuffer byteBuffer = ByteBuffer.wrap(dob.getData(), 0, dob.getLength());
for (ApplicationId applicationId : referringAppIds) {
// 更新app的delegation token
// 在NM心跳时进行同步
rmContext.getSystemCredentialsForApps().put(applicationId, byteBuffer);
}
}
public NodeHeartbeatResponse nodeHeartbeat(NodeHeartbeatRequest request)
throws YarnException, IOException {
...
ConcurrentMap<ApplicationId, ByteBuffer> systemCredentials =
rmContext.getSystemCredentialsForApps();
if (!systemCredentials.isEmpty()) {
nodeHeartBeatResponse.setSystemCredentialsForApps(systemCredentials);
}
...
}NM在心跳响应中解析出token并在内存中更新保存，后续任务重试启动资源本地化和任务结束触发日志聚合时会使用到。
注意：这里只提到了资源本地化和日志聚合时会使用到更新后的token，那么正在运行的任务会用到更新后的token吗？
答案是不会（至少是2.X版本不会）。主要是因为：token已经写入到持久化文件中，任务启动时读取该文件获取token并使用；delegation token在更新后没有写入到持久化文件中，即使可以写入（更新）到该文件，也需要有机制通知任务进程更新读取该文件才行。因此正在运行中的任务在token过期后继续操作hdfs仍旧会抛出异常。
另外，在3.X的最新版本中，注意到有相关代码的改动，应该是通知正在运行的container，但具体细节还未深入研究，后面有时间再调研。
【相关配置】
与delegation token相关的配置包括：
配置项名称 | 默认值 | 说明 |
|---|---|---|
dfs.namenode.delegation.key.update-interval | 1天 | token更新密钥的时间间隔 |
dfs.namenode.delegation.token.renew-interval | 1天 | token更新的时间间隔 |
dfs.namenode.delegation.token.max-lifetime | 7天 | token的最大生命周期 |
yarn.resourcemanager.delegation-token.alwys-cancel | false | RM结束时是否需要移除token |
yarn.resourcemanager.proxy-user-privileges.enabled | false | 是否开启特权在delegation token快过期时重新申请新的token |
yarn.resourcemanager.system-credentials.valid-time-remaining | 10800000 | 距离最大生命周期之前多长时间进行重新申请token的操作，单位毫秒 |
yarn.resourcemanager.delegation-token-renewer.thread-count | 50 | RM中delegation token更新线程的线程数 |
【总结】
本文通过一个实际的问题，并结合源码讲解了hadoop的delegation token的相关原理。
文中如有不对的地方，欢迎拍砖指正。
