membase的replication有BUG？

Question

最近在做membase的性能测试，测试集群的复制功能时出了很诡异的问题，membase用的是community_x86_64_1.7.1.1 ；enable replication =2

1. 两台server构成的集群（IP分别是193、194）

向membase依次插入10条数据，可以看到数据平均分布到了2台server，194上保存有193对象的所有副本，但193上却没有194对象的副本，实际通过failover-get测试也证实了这一点

 

2.三台server构成的集群（IP分别是191、193、194），一条一条的插入

插入数据，数据分配在了193上，此数据在191和194上各有一份副本

继续插入，数据分配在了194上，但数据在其他服务器上没有副本

继续插入，数据分配在了191上，数据在194上有一个副本

继续插入，数据分配在了194上，数据在191上有一个副本

--------------------------------

结论：membase集群的副本拷贝功能完全不靠谱？官方文档上关于replication的资料都没有说明他拷贝副本的时候是依据什么选择节点的。

Answer 1

Yes, even if the replia is set as 2. 

Answer 2

2台服务器，replication 最大只能是1吧？

Answer 3

Your RAM usage on 193 is abnormally high if you just populated 5 items.

Answer 4

BTW, i am working at couchbase.  And i know it is working. Otherwise, we won't have those customers.

Answer 5

I download the same version on windows 64 bit version. And populated 150 items. They are equally distributed to two nodes and right replicated items on each node too.

I suggest you generate the diagnosis file and ask the question in www.couchbase.org/forums/.  They will answer your questions. I am sure there are some errors in your log file during replication. 

Answer 6

Answer 7

Answer 8

机制问题？ 比如我就如问题里所说的2台server构成的集群，客户端pools连接方式，按这种说不清道不明的所谓“机制”，193上没有194的对象副本，这时候194挂了，客户端根本从193上就取不到对象。

这种不靠谱的集群，我要它何用？还不如用memcached呢

Answer 9

It is such a basic feature that it is so hard to believe that it doesn't work. 

You should monitor curr_items for active items, and curr_items_total for active items + replica items. Eventually, each node will have same number of curr_items_total, which means each node will have both data and replicated data.

If all nodes are in memory, the above process will happen very quickly. And it will take longer to replicate data from disk.

Answer 10

这是机制问题吧，不是bug