在本地启动海豚调度的服务,出现zookeeper connect timeout异常,但是检查zookeeper节点都是正常的。经过一轮分析,发现个大坑!!!
海豚调度的zookeeper配置信息:
#
# Licensed to the Apache Software Foundation (ASF) under one or more
# contributor license agreements. See the NOTICE file distributed with
# this work for additional information regarding copyright ownership.
# The ASF licenses this file to You under the Apache License, Version 2.0
# (the "License"); you may not use this file except in compliance with
# the License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
registry.plugin.name=zookeeper
registry.servers=x.x.x.x:2181
registry.namespace=dolphinscheduler
registry.base.sleep.time.ms=60
registry.max.sleep.ms=300
registry.max.retries=5
registry.session.timeout.ms=30000
registry.connection.timeout.ms=7500
registry.block.until.connected.wait=600
registry.digest=
异常启动日志如下:
Caused by: org.apache.dolphinscheduler.registry.api.RegistryException: zookeeper connect timeout
at org.apache.dolphinscheduler.plugin.registry.zookeeper.ZookeeperRegistry.start(ZookeeperRegistry.java:111) ~[classes/:na]
at org.apache.dolphinscheduler.service.registry.RegistryClient.start(RegistryClient.java:291) ~[classes/:na]
at org.apache.dolphinscheduler.service.registry.RegistryClient.afterConstruct(RegistryClient.java:81) ~[classes/:na]
at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method) ~[na:1.8.0_332]
at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62) ~[na:1.8.0_332]
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43) ~[na:1.8.0_332]
at java.lang.reflect.Method.invoke(Method.java:498) ~[na:1.8.0_332]
at org.springframework.beans.factory.annotation.InitDestroyAnnotationBeanPostProcessor$LifecycleElement.invoke(InitDestroyAnnotationBeanPostProcessor.java:389) ~[spring-beans-5.3.12.jar:5.3.12]
at org.springframework.beans.factory.annotation.InitDestroyAnnotationBeanPostProcessor$LifecycleMetadata.invokeInitMethods(InitDestroyAnnotationBeanPostProcessor.java:333) ~[spring-beans-5.3.12.jar:5.3.12]
at org.springframework.beans.factory.annotation.InitDestroyAnnotationBeanPostProcessor.postProcessBeforeInitialization(InitDestroyAnnotationBeanPostProcessor.java:157) ~[spring-beans-5.3.12.jar:5.3.12]
... 45 common frames omitted
在配置文件中其中一个关键的配置:registry.block.until.connected.wait,意思是curator等待zookeeper连接超时,超过这个时间的话,异常退出。
海豚调度连接zookeeper关键代码
public void start(Map<String, String> config) {
CuratorFrameworkFactory.Builder builder =
CuratorFrameworkFactory.builder()
.connectString(SERVERS.getParameterValue(config.get(SERVERS.getName())))
.retryPolicy(buildRetryPolicy(config))
.namespace(NAME_SPACE.getParameterValue(config.get(NAME_SPACE.getName())))
.sessionTimeoutMs(SESSION_TIMEOUT_MS.getParameterValue(config.get(SESSION_TIMEOUT_MS.getName())))
.connectionTimeoutMs(CONNECTION_TIMEOUT_MS.getParameterValue(config.get(CONNECTION_TIMEOUT_MS.getName())));
String digest = DIGEST.getParameterValue(config.get(DIGEST.getName()));
if (!Strings.isNullOrEmpty(digest)) {
buildDigest(builder, digest);
}
client = builder.build();
client.start();
try {
// 这行!!!等待连接,超出配置时间异常退出
if (!client.blockUntilConnected(BLOCK_UNTIL_CONNECTED_WAIT_MS.getParameterValue(config.get(BLOCK_UNTIL_CONNECTED_WAIT_MS.getName())), MILLISECONDS)) {
client.close();
throw new RegistryException("zookeeper connect timeout");
}
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
throw new RegistryException("zookeeper connect error", e);
}
}
那么,既然是因为等待超时导致的异常,尝试调整registry.block.until.connected.wait,从600ms修改到60000ms,也就60s,启动!!!等待大概20秒左右,正常启动了。真见鬼!!!这种情况下,我开始怀疑是不是网络有问题,开始抓包。
发现程序初始化完成,开始连接zookeeper,但迟迟没有发送tcp握手包,在这情况下,我开始怀疑代码的问题了。
通过分析zookeeper的源码,找出了慢的关键代码
private void startConnect(InetSocketAddress addr) throws IOException {
// initializing it for new connection
saslLoginFailed = false;
state = States.CONNECTING;
// 在这!!!
setName(getName().replaceAll("\\(.*\\)",
"(" + addr.getHostName() + ":" + addr.getPort() + ")"));
if (ZooKeeperSaslClient.isEnabled()) {
try {
zooKeeperSaslClient = new ZooKeeperSaslClient(SaslServerPrincipal.getServerPrincipal(addr));
} catch (LoginException e) {
// An authentication error occurred when the SASL client tried to initialize:
// for Kerberos this means that the client failed to authenticate with the KDC.
// This is different from an authentication error that occurs during communication
// with the Zookeeper server, which is handled below.
LOG.warn("SASL configuration failed: " + e + " Will continue connection to Zookeeper server without "
+ "SASL authentication, if Zookeeper server allows it.");
eventThread.queueEvent(new WatchedEvent(
Watcher.Event.EventType.None,
Watcher.Event.KeeperState.AuthFailed, null));
saslLoginFailed = true;
}
}
logStartConnect(addr);
clientCnxnSocket.connect(addr);
}
从上面的代码看到setName这个方法,作用是修改线程名称的,其中调用了addr.getHostName(),罪魁祸首就是这个!!!
java.net.InetSocketAddress#getHostName,作用是获取主机名称,如果在初始化这个对象的时候,传入的hostname是一个域名,那么可以直接返回,但是!!!!如果是一个ip地址的话,那问题就来了。
如果是一个ip地址,getHostName可以通过ip地址去反查域名,这个操作不常见,原理是通过DNS查询PTR记录,但一般来说DNS服务器是没有这个记录,所以查询不到,查询不到就会从本地的hosts文件找,在都找不到的情况下,这过程大概会阻塞个20秒,海豚设置的创建连接超时是600ms,所以必定失败。
解决方法:
例子:
x.x.x.x my-zookeeper
然后~~~就正常了。问题到这里就解决了,所以在用zookeeper的时候,推荐使用域名的方式,不然就有可能出现连接zookeeper非常慢!!!
getHostName这个方法到底做了什么操作?
getHostName的源码中会调用到java.net.InetAddress#getHostFromNameService获取主机名称
private static String getHostFromNameService(InetAddress addr, boolean check) {
String host = null;
for (NameService nameService : nameServices) {
try {
// first lookup the hostname
// 关键代码在这
host = nameService.getHostByAddr(addr.getAddress());
/* check to see if calling code is allowed to know
* the hostname for this IP address, ie, connect to the host
*/
if (check) {
SecurityManager sec = System.getSecurityManager();
if (sec != null) {
sec.checkConnect(host, -1);
}
}
/* now get all the IP addresses for this hostname,
* and make sure one of them matches the original IP
* address. We do this to try and prevent spoofing.
*/
InetAddress[] arr = InetAddress.getAllByName0(host, check);
boolean ok = false;
if(arr != null) {
for(int i = 0; !ok && i < arr.length; i++) {
ok = addr.equals(arr[i]);
}
}
//XXX: if it looks a spoof just return the address?
if (!ok) {
host = addr.getHostAddress();
return host;
}
break;
} catch (SecurityException e) {
host = addr.getHostAddress();
break;
} catch (UnknownHostException e) {
host = addr.getHostAddress();
// let next provider resolve the hostname
}
}
return host;
}
最终会调用到java.net.Inet4AddressImpl#getHostByAddr方法
public native String getHostByAddr(byte[] addr) throws UnknownHostException;
native方法,翻jdk源码,定位到一下代码
/*
* Class: java_net_Inet6AddressImpl
* Method: getHostByAddr
* Signature: (I)Ljava/lang/String;
*/
JNIEXPORT jstring JNICALL
Java_java_net_Inet6AddressImpl_getHostByAddr(JNIEnv *env, jobject this,
jbyteArray addrArray) {
jstring ret = NULL;
#ifdef AF_INET6
char host[NI_MAXHOST+1];
int error = 0;
int len = 0;
jbyte caddr[16];
struct sockaddr_in him4;
struct sockaddr_in6 him6;
struct sockaddr *sa;
/*
* For IPv4 addresses construct a sockaddr_in structure.
*/
if ((*env)->GetArrayLength(env, addrArray) == 4) {
jint addr;
(*env)->GetByteArrayRegion(env, addrArray, 0, 4, caddr);
addr = ((caddr[0]<<24) & 0xff000000);
addr |= ((caddr[1] <<16) & 0xff0000);
addr |= ((caddr[2] <<8) & 0xff00);
addr |= (caddr[3] & 0xff);
memset((void *) &him4, 0, sizeof(him4));
him4.sin_addr.s_addr = (uint32_t) htonl(addr);
him4.sin_family = AF_INET;
sa = (struct sockaddr *) &him4;
len = sizeof(him4);
} else {
/*
* For IPv6 address construct a sockaddr_in6 structure.
*/
(*env)->GetByteArrayRegion(env, addrArray, 0, 16, caddr);
memset((void *) &him6, 0, sizeof(him6));
memcpy((void *)&(him6.sin6_addr), caddr, sizeof(struct in6_addr) );
him6.sin6_family = AF_INET6;
sa = (struct sockaddr *) &him6 ;
len = sizeof(him6) ;
}
// 关键代码在这。。。。
error = getnameinfo(sa, len, host, NI_MAXHOST, NULL, 0,
NI_NAMEREQD);
if (!error) {
ret = (*env)->NewStringUTF(env, host);
}
#endif /* AF_INET6 */
if (ret == NULL) {
JNU_ThrowByName(env, JNU_JAVANETPKG "UnknownHostException", NULL);
}
return ret;
}
在Linux中,getnameinfo是glibc的一个函数,作用就是获取主机信息
那么在glibc的代码中,关键的代码如下:
https://github.com/lattera/glibc/blob/master/inet/getnameinfo.copen in new window
/* Convert host name, AF_INET/AF_INET6 case, name only. */
static int
gni_host_inet_name (struct scratch_buffer *tmpbuf,
const struct sockaddr *sa, socklen_t addrlen,
char *host, socklen_t hostlen, int flags)
{
int herrno;
struct hostent th;
struct hostent *h = NULL;
if (sa->sa_family == AF_INET6)
{
const struct sockaddr_in6 *sin6p = (const struct sockaddr_in6 *) sa;
while (__gethostbyaddr_r (&sin6p->sin6_addr, sizeof(struct in6_addr),
AF_INET6, &th, tmpbuf->data, tmpbuf->length,
&h, &herrno))
if (herrno == NETDB_INTERNAL && errno == ERANGE)
{
if (!scratch_buffer_grow (tmpbuf))
{
__set_h_errno (herrno);
return EAI_MEMORY;
}
}
else
break;
}
else
{
const struct sockaddr_in *sinp = (const struct sockaddr_in *) sa;
// 关键代码在这
while (__gethostbyaddr_r (&sinp->sin_addr, sizeof(struct in_addr),
AF_INET, &th, tmpbuf->data, tmpbuf->length,
&h, &herrno))
if (herrno == NETDB_INTERNAL && errno == ERANGE)
{
if (!scratch_buffer_grow (tmpbuf))
{
__set_h_errno (herrno);
return EAI_MEMORY;
}
}
else
break;
}
好了,不追下去了,看到while语句块,大概是在这里阻塞了10几20秒。。。