12.5 使用故障转移协议应对代理或网络故障
12.5 Surviving network or broker failure with the failover protocol
12.5 使用故障转移协议应对代理或网络故障
?
We introduced the failover protocol in chapter 4, where we explained the basics
behind allowing a client to fail over to another ActiveMQ broker in the case of failure.
The failover protocol is the default protocol used by the Java client, so it’s worth looking
in more detail at some of its optional features and capabilities.
?
在第4章中.我们介绍恶劣失效转移协议,解释了在失效的情况下允许客户端转移到另外一个
ActiveMQ代理的背后基础知识.失效转移协议是ActiveMQ代理的Java客户端使用的默认
协议,因此我们有必要了解更多该协议相关的特性和功能.
?
By default, you specify in the client URI one or more ActiveMQ brokers that could
be used for the connection:
?
默认情况下,你给客户端的URI指定了一个或者多个ActiveMQ代理以便用来创建连接:
failover:(tcp://host1:61616,tcp://host2:61616,ssl://host2:61616)
?
By the way, specifying the failover transport URI like this is okay, too, although it can
get a bit messy if there are any embedded query parameters.
?
这种方式指定URL是没有问题的,但是假如有一些嵌入的查询参数的话这种方式指定的URI看起
来有些混乱.
?
ActiveMQ will connect to one of the brokers defined in the list, selecting one at
random. With the failover protocol, the ActiveMQ client will instantiate a periodic
keepalive protocol, so that it can detect whether the broker is no longer reachable
(connection or broker lost). When it detects that the broker is no longer available, it
will randomly select another broker from the list provided at startup. If only one broker
URI is provided, the client will periodically check to see if the broker is available
again. It’s possible to listen for transport interrupts by setting a TransportListener
on the ActiveMQ connection:
?
ActiveMQ将在指定的URI列表中随机选择一个然后创建到该代理的连接.通过使用失效转移协议,
ActiveMQ客户端会定期启动一个保持连接协议,以便侦测代理是否不可达(连接丢失或代理丢失).
如果侦测到代理不再可用,失效转移连接器会从提供的URI列表中随机的选择另一个代理的URI.
如果仅提供了一个代理URI,客户端会隔一段时间后再检测这个唯一代理是否可用了.
可以通过给ActiveMQ连接设置TransportListener来监听传输的是否中断了,如下代码所示:
?
import org.apache.activemq.ActiveMQConnection;
import org.apache.activemq.ActiveMQConnectionFactory;
import org.apache.activemq.transport.DefaultTransportListener;
...
ActiveMQConnection connection = (ActiveMQConnection)
connectionFactory.createConnection();
?
connection.addTransportListener(new DefaultTransportListener()?
{
public void onException(IOException arg0)?
{
System.err.println("This is bad");
}
?
public void transportInterrupted()?
{
System.out.println("Transport interrupted");
}
?
public void transportResumed()?
{
System.out.println("Transport resumed");
}
});
connection.start();
?
When you supply the failover protocol with more than one transport URI to use, by
default it will select one at random to use, to ensure load balancing of clients across
brokers. When you want to have a guaranteed order of brokers that the client will
connect to, you need to disable the random selection of brokers by disabling the
randomize failover property:
?
如果你为失效转移连接器提供了多个URI,那么默认情况下,连接器会随机使用其中的一个URI,以便
保证客户端访问代理的负载均衡.当你需要确保客户端按照设定的顺序来连接指定的代理是,你
要设置失效转移连接器的randomize属性为false以便关闭随机选择代理URI来连接的功能.
?
failover:(tcp://master:61616,tcp://slave:61616)?randomize=false
?
If none of the ActiveMQ brokers specified by the failover URI are available, then by
default the failover transport will wait before trying again. The failover protocol will
wait an increasing amount of time between each successive failure to connect to an
ActiveMQ broker—this is called an exponential back-off. The failover protocol by default
has useExponentialBackOff enabled. The wait time between successive attempts to
connect is called the initialReconnectDelay (initial value is 10ms) and the multiplier
to increase the wait time is called the backOffMultiplier (default value is 2.0).
?
如果失效转移连接器中设定的URI都不可用,那么默认情况下,失效转移连接器会等待一段时间
然后重新尝试连接.相邻两次失败后尝试重新连接之前的等待时间是递增的--这被称为指数级
back-off.失效转移连接器默认会开启useExponentialBackOff.在重新尝试连接之前的等待时间
称为initialReconnectDelay(初始值为10毫秒)下一次失败后尝试重新连接之前的等待时间与前一次
等待时间的倍数称为backOffMultiplier(默认值为2.0).
?
You can also set the maximum time period for the failover protocol by using max-
ReconnectDelay (default is 30000ms). An example configuration is shown next:
?
你也可以使用maxReconnectDelay(默认值为30000毫秒)设置失效后重新连接之前等待
时长的最大值.下面是一个示例配置代码:
?
failover:(tcp://master:61616,tcp://slave:61616)?backOffMultiplier=1.5,initialReconnectDelay=1000
?
One potential problem that you may run into using any transport protocol based on
TCP is the ability to know when a peer (for ActiveMQ, this will be the broker) has died.
This can happen for several reasons, like the failure of the ActiveMQ broker or loss of
network.?
?
使用基于TCP的传输连接器可能导致一个潜在的问题,即,连接的对端(对于ActiveMQ来说,这个对端是指代理)
是否已经失效了.导致这个问题的原因有几个,比如ActiveMQ代理失效或者网络连接丢失.
?
Also, if there’s a firewall between the ActiveMQ client and broker, it may
drop the connection if it’s inactive for some time. It’s possible to configure keepalive
on the TCP connection, but this is operating system–specific and can require changes
to kernel parameters—and doesn’t work well in heterogeneous environments. For
this reason, ActiveMQ uses a keepalive protocol on top of its transports, to keep firewalls
open and also detect whether the broker is no longer reachable. The keepalive
protocol periodically sends a lightweight command message to the broker, and
expects a response.?
?
If it doesn’t receive one within a given time period, ActiveMQ will
assume that the transport is no longer valid. The failover transport listens for failed
transports and will select another transport to use on such a failure. The parameter
used by the keepalive protocol in ActiveMQ is maxInactivityDuration, which is an
OpenWire property; the default is 30000 (milliseconds). You can specify a different
timeout to be used with the failover transport, as shown:
?
另外,如果在ActiveMQ代理和客户端之间存在防火墙,如果连接长时间处于非活动状态,防火墙可能会
切断连接.可以配置TCP连接的keepalive属性,但这个属性是操作系统属性,可能会导致修改系统内容
参数并且在复杂环境中不能很好的工作.鉴于此,ActiveMQ在传输连接器的上层使用一个保持连接协议
以保持防火墙始终打开同时监测代理是否不可达.keepalive协议每隔一段时间会发送一个轻量级的
命令消息给代理然后等待代理的响应.
?
如果没有在规定时间内收到响应,ActiveMQ会认为当前的传输连接器不再可用.失效转移连接器监听到
失效的连接器后会选择另外一个连接器来应对前面连接器的失效.ActiveMQ在keepalive协议中使用
maxInactivityDuration参数是一个OpenWire属性,默认值为30000(毫秒).你可以为失效转移协议
的这个参数设置新值,如下代码所示:
?
failover:(tcp://host1:61616?wireFormat.maxInactivityDuration=1000,tcp://host2:61616?wireFormat.maxInactivityDuration=1000)
?
Note that you have to set this parameter (and any other OpenWire properties) on the
transports used by the failover protocol, not the failover protocol itself.
?
注意,你必须为使用失效转移协议的连接器设置这个参数(以及其他所有的OpenWire属性),而不是为失效转移
本身设置这个属性.
?
By default, the delivery mode for sending messages from an ActiveMQ client is persistent
(this is so ActiveMQ is compliant with the JMS specification). A message sent
with a persistent delivery mode will be synchronous—send() will block until it gets a
receipt from the broker that it has successfully received and stored the message. For
applications where performance is important, using nonpersistent delivery can dramatically
improve results (see chapter 13). When nonpersistent delivery is used, messages
are sent asynchronously, which has the downside that you can potentially lose
messages in flight if a transport fails. You can configure the failover transport to prevent
this by enabling message caching with the trackMessages failover transport
property. You can also control the maximum size of this message cache by use of the
maxCacheSize failover property—the default is 128 KB (the memory allocation size
allowed for the message cache).?
?
默认情况下,ActiveMQ客户端在发送消息时使用的分发模式是持久化的(因此ActiveMQ是遵循JMS规范的).
使用持久化分发模式的消息将使用同步方式发送,这种方式发送消息后,发送消息的线程将会阻塞直到收到代理的
回执以确定消息成功发送以及成功存储了.对于关注性能的程序来说,使用非持久化分发方式能显著提高性能(参
见13章).使用非你持久化分发时,消息都是异步发送,这样就产生了隐患--如果传输连接器失效你可能会丢失消息.
你可以通过配置失效转移连接器的trackMessages属性来启用消息缓存以避免消息丢失.这种缓存消息的最大值
也可以使用失效转移连接器的maxCacheSize配置--该配置的默认值是128KB(允许分配给消息缓存的内存大小).
?
Here’s an example configuration for enabling cachingand setting the maximum cache size:
下面是示例配置代码,配置中通过设置最大缓存大小启用了消息缓存:
failover:(tcp://host1:61616,tcp://host2:61616)?trackMessages=true,maxCacheSize=256000
?
For high-performance applications, fast failover is important too. It takes a considerable
amount of time to build up a new transport connection (in the order of tens to
hundreds of milliseconds), so to enable fast failover, ActiveMQ can optionally allow
the failover protocol to build a backup connection ready to go if the primary transport
fails. The failover property to set to allow a backup connection is unsurprisingly
called backup. You can have more than one backup enabled (the default is 1) by setting
the failover property backupPoolSize.?
?
对于高性能程序来说,失效转移协议也是非常重要的.新建一个传输连接需要的时间是很可观的(需要几十到
几百毫秒),因此启用了快速失效转移后,ActiveMQ能有选择性的运行失效转移协议创建一个备份连接,以便
能在主连接失效时迅速接替主连接继续工作.设置运行失效转移连接器建立备份连接的属性毫无悬念的被
称为backup.你可以通过backupPoolSize属性设置多于一个的备份(默认值是1个).
?
An example failover URI using backup is shown next:
启用备份连接的失效转移URI示例代码如下:
?
failover:(tcp://host1:61616,tcp://host2:61616,tcp://host3:61616)?backup=true,backupPoolSize=2
?
So far we’ve looked at configuring the failover transport with a static list of
URIs ?to ?the ?broker, but ?an ?ActiveMQ ?broker does ?know ?what ?brokers it’s
connected to, so ?it can optionally ?dynamically update clients ?with changes to
the cluster as brokers come and go. ?To enable dynamic updates of brokers to ?an
ActiveMQ client, ?we need ?to enable ?the property ?updateClusterClients on ?the
TransportConnector used in the ActiveMQ broker configuration. Properties on ?the
TransportConnector are used to control the updates; these are as shown in ?table
12.1.
?
到目前为止,我们都是使用一个静态的URI列表来配置失效转移连接器的,但是ActiveMQ是可以
自行侦测需要连接的代理,因此ActiveMQ客户端可以在代理新增到集群或者从集群中移除时
动态的更新代理集群.我们需要启用传输连接器的updateClusterClients属性来为ActiveMQ
客户端启用动态更新代理集群.TransportConnector中用于控制代理集群自动更新的属性如
表12.1所示.
?
Table 12.1 TransportConnector properties for updating clients of cluster changes
表12.1 TransportConnector中用于控制代理集群自动更新的属性
?
PropertyDefault valueDescription
updateClusterClients false If true, pass information to connected clients
about changes in the topology of the broker cluster.
updateClusterClients false 如果设置为true,当代理集群发送变化时会传递消息给以连接到集群的客户端
?
rebalanceClusterClientsfalse If true, connected clients will be asked to rebalance
across a cluster of brokers when a new broker
joins the network of brokers.
rebalanceClusterClientsfalse 如果设置为true,当新代理加入集群时,客户端会被要求重新为进行进行负载均衡
?
updateClusterClientsOnRemove false If true, will update clients when a cluster is
removed from the network. Having this as separate
option enables clients to be updated when
new brokers join, but not when brokers leave.
updateClusterClientsOnRemove false 如果设置为true,当代理集群从代理网络中移除时更新客户端.
将这个属性独立出来是为了能够在新代理假加入代理网络时通知客户端,
但是移除代理时可以不通知.
?
updateClusterFilter null Comma-separated list of regular expression filters
used to match broker names of brokers to
designate as being part of the failover cluster for
the clients.
updateClusterFilter null 逗号分隔的正则表达式列表,用于匹配代理名称,
? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? 只有匹配上的代理加入到客户端失效转移代理集群中时,才会更新客户端.
.
?
An interesting property is rebalanceClusterClients which, if enabled, ensures that
the ActiveMQ clients will evenly distribute themselves across the cluster when a new
broker joins.
rebalanceClusterClients是一个有趣的属性,启用该属性时,当一个新代理假如集群,ActiveMQ客户端将平均
分配它们的负载到集群中所有的代理.
?
An example configuration for an ActiveMQ broker on a machine named tokyo
using these properties is shown next:
使用上述属性,在一台机器名为tokyo机器上配置ActiveMQ的示例代码如下:
<broker>
...
<transportConnectors>
? <transportConnector?
? ? ? name="clustered"
? ? ? uri="tcp://0.0.0.0:61616"
? ? ? updateClusterClients="true"
? ? ? updateClusterFilter="*newyork*,*london*" />
<transportConnectors>
...
</broker>
?
This configuration will update any clients that are using the failover transport protocol
with the locations of any brokers joining that have newyork or london in their broker
names. With updateClusterClients enabled, you only need to configure the failover
protocol with one broker in the cluster, for example:
?
上述配置将在代理名称中含有newyork或london的代理加入代理集群中时更新所有使用失效转移
传输连接器的客户端.开启updateClusterClients后,你仅需要为集群中的一个代理配置失效转移协议,
例如:
failover:(tcp://tokyo:61616)
?
As the client will be updated automatically as new brokers join and leave the cluster, if
the machine tokyo should fail, the client would automatically fail over to either newyork
or london.
?
因为客户端在新代理假如集群,或从集群中移除代理时会自动更新,所以如果名称为tokyo机器宕机了,
客户端会自动的进行失效转移从而连接到机器名为newyork或london的代理.
?
You may wish for your clients to automatically be distributed around all the
machines in a cluster, so all the machines share the load of your messaging application.
By enabling the property rebalanceClusterClients on the TransportConnector, as
ActiveMQ brokers join and leave the cluster, this will automatically happen.
?
你可能希望你的客户端能够自动的针对代理集群中所有的集群进行分布式部署,这样所有的机器
可以为消息程序均衡负载.通过启用传输连接器的rebalanceClusterClients属性,在ActiveMQ
代理加入集群或从集群中移除代理时,上述客户端针对代理集群进行分布式部署将自动完成.
?
In this section we’ve taken a deeper look at some the functionality that can be
used with the failover transport protocol. You should now have a better understanding
of how to configure an ActiveMQ client to detect and survive a network outage or
broker failure.
?
本节中,我们深入的了解了失效转移连接器的相关功能.现在你应该能够很好的理解如何配置ActiveMQ
客户端以便侦测网络问题或代理失效并且在网络发生问题和代理失效是继续工作.
?
In the next section we’re going to look at scheduling a message to be delivered by
the ActiveMQ broker at some time in the future.
?
下一节中,我们将看到如何配置ActiveMQ代理的消息调度进行消息的延迟发送.
