浅看rabbitmq的mnesia部署

伸缩性：根据系统负载，可以在运行中过程中添加或者删除服务节点，改变系统处理规模。
mnesia是一个分布式数据库模块，由多个节点构成的数据库cluster，数据表的位置对应用是透明的。透过该特性，很容易构建出一个具有高伸缩性的系统。
rabbitmq是一个分布式的消息中间件，在mnesia-cluster的机制上可由多个节点共同构建。

rabbitmq在一个节点上初始化mnesia的过程概况如下：
1.启动mnesia，尝试连接到系统的其他节点上；
2.若无法连接到任何节点，表示该节点是系统中第一个启动的节点；
    2.1.在这种情况下可以检查数据库的老表，然后等待其他节点的连接；
3.若连接到一些节点，则这些节点会同步并合并schema表；
    3.1.在当前节点上创建schema表和其他数据表的副本后，该节点就和所连接的节点数据将保持同步和一致；

 

%%rabbit_mnesia.erl
init_db(ClusterNodes) ->
    case mnesia:change_config(extra_db_nodes, ClusterNodes -- [node()]) of
    end

 启动mnesia后连接到其他节点。

 

%%rabbit_mnesia.erl
    case mnesia:change_config(extra_db_nodes, ClusterNodes -- [node()]) of
        {ok, []} -> 

 无法连接到任何节点，则检查当前数据库目录下的老表是否正确。（数据库目录为空则建立新表）

 

%%rabbit_mnesia.erl
    case mnesia:change_config(extra_db_nodes, ClusterNodes -- [node()]) of
        {ok, [_|_]} ->
            IsDiskNode = ClusterNodes == [] orelse %%ClusterNodes==[]主节点
                lists:member(node(), ClusterNodes),
            ok = wait_for_replicated_tables(),
            ok = create_local_table_copy(schema, disc_copies),
            ok = create_local_table_copies(case IsDiskNode of

成功连接到一些节点后，mnesia之间交换数据库元信息，并等待在当前节点上有磁盘副本(disc_copies)的表和cluster完成同步。
            ok = wait_for_replicated_tables(),
如果当前节点是数据存储节点，还要在该节点上建立一些表格的磁盘副本
            ok = create_local_table_copies(case IsDiskNode of

 

我做了些简单的实验来观察mnesia相互连接时的特性。

(a@localhost)1> mnesia:create_schema([node()]).
ok
(a@localhost)2> mnesia:start().
ok
(a@localhost)3> mnesia:create_table(user, [{disc_copies, [node()]}]).
{atomic,ok}
(a@localhost)4> mnesia:info().
---> Processes holding locks <--- 
---> Processes waiting for locks <--- 
---> Participant transactions <--- 
---> Coordinator transactions <---
---> Uncertain transactions <--- 
---> Active tables <--- 
user           : with 0        records occupying 304      words of mem
schema         : with 2        records occupying 524      words of mem
===> System info in version "4.4.10", debug level = none <===
opt_disc. Directory "/home/hwh/a" is used.
use fallback at restart = false
running db nodes   = [a@localhost]
stopped db nodes   = [] 
master node tables = []
remote             = []
ram_copies         = []
disc_copies        = [schema,user]
disc_only_copies   = []
[{a@localhost,disc_copies}] = [schema,user]
3 transactions committed, 0 aborted, 0 restarted, 1 logged to disc
0 held locks, 0 in queue; 0 local transactions, 0 remote
0 transactions waits for other nodes: []
ok

 a节点上有两个磁盘表，分别是schema和user。

 

(b@localhost)1> mnesia:start().
ok
(b@localhost)2> mnesia:change_config(extra_db_nodes, ['a@localhost', 'b@localhost', 'c@localhost']--[node()]).
{ok,[a@localhost]}

 b节点尝试连接a,c节点，最终连接上了正在运行的a节点。连接之后b节点是什么状态呢？

 

(b@localhost)3> mnesia:info().
---> Processes holding locks <--- 
---> Processes waiting for locks <--- 
---> Participant transactions <--- 
---> Coordinator transactions <---
---> Uncertain transactions <--- 
---> Active tables <--- 
schema         : with 2        records occupying 533      words of mem
===> System info in version "4.4.10", debug level = none <===
opt_disc. Directory "/home/hwh/b" is NOT used.
use fallback at restart = false
running db nodes   = [a@localhost,b@localhost]
stopped db nodes   = [] 
master node tables = []
remote             = [user]
ram_copies         = [schema]
disc_copies        = []
disc_only_copies   = []
[{a@localhost,disc_copies}] = [user]
[{a@localhost,disc_copies},{b@localhost,ram_copies}] = [schema]
4 transactions committed, 0 aborted, 0 restarted, 0 logged to disc
0 held locks, 0 in queue; 0 local transactions, 0 remote
0 transactions waits for other nodes: []
ok

可以看到user表是remote的，在a节点上有磁盘副本，在b节点上没有任何类型的副本。schema表已经合并，分别存储在a,b节点上。
此时在b节点上就可操作user表，表的位置是透明的。

 

(b@localhost)4> mnesia:change_table_copy_type(schema, node(), disc_copies).
{atomic,ok}
(b@localhost)5> mnesia:add_table_copy(user, node(), disc_copies).
{atomic,ok}
(b@localhost)6> mnesia:info().
---> Processes holding locks <--- 
---> Processes waiting for locks <--- 
---> Participant transactions <--- 
---> Coordinator transactions <---
---> Uncertain transactions <--- 
---> Active tables <--- 
user           : with 0        records occupying 304      words of mem
schema         : with 2        records occupying 542      words of mem
===> System info in version "4.4.10", debug level = none <===
opt_disc. Directory "/home/hwh/b" is used.
use fallback at restart = false
running db nodes   = [a@localhost,b@localhost]
stopped db nodes   = [] 
master node tables = []
remote             = []
ram_copies         = []
disc_copies        = [schema,user]
disc_only_copies   = []
[{a@localhost,disc_copies},{b@localhost,disc_copies}] = [schema,user]
6 transactions committed, 0 aborted, 0 restarted, 2 logged to disc
0 held locks, 0 in queue; 0 local transactions, 0 remote
0 transactions waits for other nodes: []
ok

在b节点上建立schema表和user表的磁盘副本后，发现user表不再是remote属性了，可从本地直接读取。

 

先退出b节点，再退出a节点，然后只重启b节点。

(b@localhost)1> mnesia:start().
ok
(b@localhost)2> mnesia:info().
---> Processes holding locks <--- 
---> Processes waiting for locks <--- 
---> Participant transactions <--- 
---> Coordinator transactions <---
---> Uncertain transactions <--- 
---> Active tables <--- 
schema         : with 2        records occupying 542      words of mem
===> System info in version "4.4.10", debug level = none <===
opt_disc. Directory "/home/hwh/b" is used.
use fallback at restart = false
running db nodes   = [b@localhost]
stopped db nodes   = [a@localhost] 
master node tables = []
remote             = []
ram_copies         = []
disc_copies        = [schema,user]
disc_only_copies   = []
[] = [user]
[{b@localhost,disc_copies}] = [schema]
2 transactions committed, 0 aborted, 0 restarted, 0 logged to disc
0 held locks, 0 in queue; 0 local transactions, 0 remote
0 transactions waits for other nodes: []
ok
(b@localhost)3> mnesia:dirty_read(user, key).
** exception exit: {aborted,{no_exists,[user,key]}}
     in function  mnesia:abort/1

 这时看到user表此时是不可用的，因为a,b节点共同维护user表的一致性，但是b节点先退出，所以user表的最终状态由a节点决定。
启动a节点之后，user表就变成可用状态了。

(a@localhost)1> mnesia:start().

(b@localhost)4> mnesia:info().               
---> Processes holding locks <--- 
---> Processes waiting for locks <--- 
---> Participant transactions <--- 
---> Coordinator transactions <---
---> Uncertain transactions <--- 
---> Active tables <--- 
user           : with 0        records occupying 304      words of mem
schema         : with 2        records occupying 542      words of mem
===> System info in version "4.4.10", debug level = none <===
opt_disc. Directory "/home/hwh/b" is used.
use fallback at restart = false
running db nodes   = [a@localhost,b@localhost]
stopped db nodes   = [] 
master node tables = []
remote             = []
ram_copies         = []
disc_copies        = [schema,user]
disc_only_copies   = []
[{a@localhost,disc_copies},{b@localhost,disc_copies}] = [schema,user]
3 transactions committed, 0 aborted, 0 restarted, 0 logged to disc
0 held locks, 0 in queue; 0 local transactions, 0 remote
0 transactions waits for other nodes: []
ok

 

评论

2 楼 zdx3578 2012-02-02  

good  不错

1 楼 whhbest 2010-08-15  

讲的很透彻，目前正在学习rabbitmq源码