请教关于HADR的几个疑问

wangzhonnew

1) no, since alternate server info stored on server, if client unable to connect to primary at beginning, it will not get any info about alternate server then...
2) when client connect with server, in each communication package between server and client, it will include the alternate server information. if primary down at somepoint, client got disconnected then it will try alternate server information.
3) i don't believe it will try primary again... that is deadloop based on what you describe....
4) you cannot have both primary/secondary started as HADR primary.... you need to issue TAKEOVER command on standby to make it serve requests, when primary server is up, you need to activate the database using start HADR as standby, then manually do TAKEOVER again on primary...
5) no, it will not try alternate server when hitting those messages, if they hit maxapp errors, some SQL CODE will be returned to client, so it's not TCPIP error like connection has been refused or connection timeout...


above are my understanding, never tested myself, you are more than welcome to test if you want (and let us know the result)

[ 本帖最后由 wangzhonnew 于 2008-6-20 01:23 编辑 ]

unixnewbie

QUOTE:
--------------------------------------------------------------------------------
原帖由 askgyliu 于 20/6/2008 00:46 发表


我们整个的APPLICATION都是IBM来的，所以也不知道是不是能支持HADR。

1) 若是一个普通的DB2 CLIENT，比如说我自己的PC，若是在DB2 HADR FAILOVER后，我的PC也重起了，那我的DB2 CLIENT有办法知道REROUTE到SECONDARY SERVER吗？

2) ALTERNATE SERVER INFORMATION 不是只存在于SERVER吗？还是在客户端在第一次连接成功前也能存有PRIMARY 跟 SECONDARY CATALOG INFORMATION？红皮书/MANUAL是说在SERVER中设置ALTERNATE SERVER？

3) 若是客户端在第一次连接成功前就有PRIMARY 跟 SECONDARY  CATALOG INFORMATION的话，那客户端是不是会不停地尝试连接到数据库？就是说，如果不能连接到PRIMARY，那就自动尝试SECONDARY；如果还不能，那就尝试回PRIMARY？直到连接成功为止？

4) 如果3是肯定的回答的话，那如果在FAILOVER TO SECONDARY后，因为某些原因，PRIMARY也被启动了，那会不会造成一些客户端连接到SECONDARY，而另一些(重起的)客户端连接到PRIMARY？

5) 如果3是肯定的回答的话，那如果在FAILOVER TO SECONDARY后，因为某些原因，PRIMARY也被启动了，会不会造成某些客户端即使没重起也可能连接到PRIMARY？比如说，尝试连接到SECONDARY时，遇到MAX APPLICAION，那就去尝试(REROUTE) ALTERNATE SERVER(PRIMARY SERVER IN THIS CASE)？
--------------------------------------------------------------------------------


1. Yes. Only after your client (type 2) has been able to connect to the server prior to HADR FAILOVER. 'Update alternate server' command is run on SERVER. The first time when client connects to the server it grabs those information and stores them in the client's local database directory. It doesn't matter how many times you bounce your client & the server since then. The client has already known where the alternate server is.

2. Already explain in 1.

3. No. If the client is failed to connect to PRIMARY or SESONDARY, it stops trying and will give you some message.

4. The symptoms that having two PRIMARY ("dual heads") sometimes could be happening when PRIMARY is disconnected from clients and SECONDARY, then DBA issued TAKEOVER BY FORCE on SECONDARY. All existing clients will be connecting to this new 'PRIMARY' at this point in time. All of a suddent the link between the old PRIMARY and clients is active again. Then all those new clients comming through will definately connect to the old PRIMARY. In this case, there will be dual heads. Hence DBA should be extremly cautious when issuing TAKEOVER BY FORCE command.

5. Client rerouting only happens when TCP/IP layer has problems.

[ 本帖最后由 wangzhonnew 于 2008-6-20 09:20 编辑 ]

wangzhonnew

sorry to the above, clicked wrong button

are you sure about 1)?

on infocenter i see the following:
Regardless of whether you have authority to update the database directory at the client machine, the alternate server information is always kept in memory. In other words, if you did not have authority to update the database directory (or because it is a read-only database directory), other applications will not be able to determine and use the alternate server, because the memory is not shared among applications.

http://publib.boulder.ibm.com/in ... 5%72%76%65%72%22%20

[ 本帖最后由 wangzhonnew 于 2008-6-20 09:37 编辑 ]

wangzhonnew

hmm... i think maybe you are right anyway it's good to have some tests... or maybe you already tested it before?

wxx_126

type2和type是否时通过在WAS上的设置后，不管中间主/备服务器角色又切换了几次，都是终身受用了？？？
db2 client和catalog没有直接关系吧？只要catalog 到起始主服务器，经过一次重路由设置后，而不管中间主/备服务器角色又切换了几次，都不用再管了？？？

[ 本帖最后由 wxx_126 于 2008-6-20 11:04 编辑 ]

unixnewbie

Simply just do a test by yourself.
Suppose you've got a DB2 instance on AIX as PRIMARY instance which has a HADR connection set up with another instance as SECONDARY.

Install a DB2 administation client on your XP.
1. UPDATE alternate server on PRIMARY & SECONDARY.
2. catalog PRIMARY DB in your XP client.
3. db2 list db directory <--- alternate server hostname would be blank.
4. db2 connect to PRIMARY DB
5. db2 list db directory <--- you will see alternate server infor should has been fetched from server.

Since then, you'll never have to recatalog your PRIMARY/SECONDARY DB again. Those alternate server infor has been stored in your local db directory.

unixnewbie

原帖由 wxx_126 于 20/6/2008 12:57 发表
type2和type是否时通过在WAS上的设置后，不管中间主/备服务器角色又切换了几次，都是终身受用了？？？
db2 client和catalog没有直接关系吧？只要catalog 到起始主服务器，经过一次重路由设置后，而不管中间主/备服务器角色又切换了几次，都不用再管了？？？

When you configured WAS to access DB2 by using type 2 (you have a local administration client installed), once alternate server has been configured on servers and WAS connects to one of the servers. That's it. No further recatalog actions are required after TAKEOVER or WAS server being bounced (restarted). In this case, WEB clients talk to WAS instead of DB2 server directly.

wangzhonnew

原帖由 unixnewbie 于 2008-6-20 13:47 发表
Simply just do a test by yourself.
Suppose you've got a DB2 instance on AIX as PRIMARY instance which has a HADR connection set up with another instance as SECONDARY.

Install a DB2 administation client on your XP.
1. UPDATE alternate server on PRIMARY & SECONDARY.
2. catalog PRIMARY DB in your XP client.
3. db2 list db directory  

good point thanks

twisters-pb

在websphere上的db2 hadr-jdbc-failover配置 http://twistman.buyberry.net/was-6-config-db2-hadr-jdbc-failover/
在weblogic上的db2 hadr jdbc failover 配置：http://twistman.buyberry.net/web ... hadr-jdbc-failover/