610上用集成的两块网卡实现AFT（Adapter Fault Tolerance）?

老农a

[quote]原帖由 "炸鸡"]如果用EVENT的话，还是要装HA吧，那倒不如直接定义SVC和STDBY，那就不用设置EVENT了，对吗？[/quote 发表：

写个小脚本监控errlog就可以了啊。

炸鸡

我先试HACMP，只定义CLUSTER、NODE、ADAPTER，修改了/etc/hosts和.rhosts之后，做了同步，正常。netstat、clstat的输出都正常。
我拔掉SVC网线，一直在监察hacmp.out，情况如下：
35'22" swap_adapter
41'38" swap_adapter_complete
41'43" join_standby
41'47" network_up
41'47" network_up_complete
41'52" network_down
41'52" network_down_complete
41'56" fail_standby
最后的情况是这样：从拔网线开始有约5、6分钟PING不通，后来通了。clstat报找不到任何CLUSTER，netstat输出正常，两张网卡的地址换了过来。虽然现在可以PING通，但中间的DOWNTIME太长，没意义。我看过以前的帖子，好象有人试过这样做，但结果是不通，我估计是他还没有等到这个6分钟。

larryh

DOWNTIME和很多因素有关，包括交换机的端口速度，比如CISCO 6509不打开FASTPORT，就会发生做HACMP halt -q测试时网卡状态不正常的问题，但是，做clstop TAKEOVER测试正常，因为这种切换是软件操作的，交换机可以及时得到通知，而halt -q就确实是模拟实际系统崩溃的情况。

如果你换成HUB，保证DOWNTIME很短。

我给客户安装HA，最好都要做halt -q测试，除非用户不让。

炸鸡

假设我忽略了DOWNTIME，但最后的结果还是有点不正常，因为clstat找不到任何的CLUSTER。

若大家没有其他的意见，我会测试ETHERCHANNEL。

larryh

clstat经常不能正常工作，即使在正常的HA环境下。我也不知怎么回事，还是看hacmp.out最稳妥。

炸鸡

但hacmp.out也不正常啊，在swap_adapter_complete后面的那一大堆都不应该出现啊。

larryh

只要没有EVENT ERROR，clstat看到的HA状态就是正常的，所以说clstat不看也罢。
至于swap_adapter_complete后面的，除非反复出现，否则也是正常的。

大猫

我觉得老农的脚本最有吸引力,谁编好了share一下

老农a

Overview of RSCT Resource Monitoring and Control
The RSCT Resource Monitoring and Control (RMC) application is part of Reliable Scalable Cluster Technology (RSCT). It provides consistent comprehensive monitoring of system resources. By monitoring conditions of interest and providing automated responses when these conditions occur, RSCT Resource Monitoring and Control helps maintain system availability. RSCT Resource Monitoring and Control is installed as part of the base operating system and is administered by means of the easy-to-use Web-based System Manager graphical user interface or the command line.

You can set up monitoring easily through the Web-based System Manager user interface or you can use the command line. See Using the Monitoring Application for more details.

The Monitoring application offers a comprehensive set of monitoring and response capabilities that lets you detect, and in many cases correct, system resource problems such as a critical filesystem becoming full. You can monitor virtually all aspects of your system resources and specify a wide range of actions to be taken when a problem occurs, from simple notification by e-mail to recovery that runs a user-written script. You can specify an unlimited number of actions to be taken in response to an event.

As system administrator, you have a great deal of flexibility in responding to events. You can respond to an event in different ways based on the day of the week and time of day. The following are some examples of how you can use monitoring:

You can be alerted by e-mail if /tmp is unmounted during working hours, but you can have the problem logged if /tmp is unmounted during nonworking hours.
You can be notified by e-mail when /var is 80% full.
You can configure your paging software to notify you if a critical file system goes offline.
You can have a user-written script run automatically to delete the oldest unnecessary files when /tmp is 90% full.

炸鸡

刚才看IBM的2002年总第18期蓝色航标，里面介绍虚拟IP，我看可以当做解决这个问题的第四个方法。不过我的测试机器虽然设好了虚拟IP，但是PING不通。我觉得如果这个方法可行的话，都挺好，效果跟ETHERCHANNEL差不多。