关于RHCS的仲裁试探功能

vinciho

我测试发现，RHCS v5没有仲裁分区时，当心跳线断开时，两个节点会不断的被对方重启。

为了避免这个问题，我从网上看到，似乎启用仲裁分区的试探功能，去ping网卡，只要通的，就不会重启对方，从而避免了心跳断开的严重问题。

但是，我测试发现，根本没用，还是会互相fencing。

难道，我理解的仲裁分区的试探不是这个意思？？ 请高人讲解！！！

dooza

心跳是怎么连接的？

vinciho

我测试环境的心跳，是网线直连的。

请高人指点！！！

vinciho

没人吗？……

心跳线除了做bonding，并分别连接至两台做冗余的交换机外，有其他解决心跳线这个单一故障点的方案吗？

vinciho

这个版块的高手呢？我等了4天了……

litao941

兄台，我最近也遭遇这情况，两台 DELL R810 + CentOS 5.3 + RHCS oracle，没有使用仲裁盘，出现过 几次 primary 被 standby fence 掉，查了下 RHCS 相关参考，仲裁在前面（默认超时时间比fence 要短）， fence 是最后一道关口，觉得添加仲裁盘应该无效，所有先升级了集群软件包，以观时变。

vinciho

你有查为什么primary被standby fence??你的问题应该不在心跳线吧，心跳线断开的话，应该会互相fence而重启的。

litao941

兄台，虽然我的standby 没有被fence，但还是觉得跟心跳有关系，
primary 上日志

Jul 24 23:21:31 JCXora01 openais[6538]: [TOTEM] The token was lost in the OPERATIONAL state.
Jul 24 23:21:31 JCXora01 openais[6538]: [TOTEM] Receive multicast socket recv buffer size (320000 bytes).
Jul 24 23:21:31 JCXora01 openais[6538]: [TOTEM] Transmit multicast socket send buffer size (320000 bytes).
Jul 24 23:21:31 JCXora01 openais[6538]: [TOTEM] entering GATHER state from 2.
Jul 24 23:21:51 JCXora01 openais[6538]: [TOTEM] entering GATHER state from 0.
Jul 24 23:21:51 JCXora01 openais[6538]: [TOTEM] Creating commit token because I am the rep.
Jul 24 23:21:51 JCXora01 openais[6538]: [TOTEM] Saving state aru 30 high seq received 30
Jul 24 23:21:51 JCXora01 openais[6538]: [TOTEM] Storing new sequence id for ring 4b8
Jul 24 23:21:51 JCXora01 openais[6538]: [TOTEM] entering COMMIT state.
Jul 24 23:21:51 JCXora01 openais[6538]: [TOTEM] entering RECOVERY state.
Jul 24 23:21:51 JCXora01 openais[6538]: [TOTEM] position [0] member 219.13.240.172:
Jul 24 23:21:51 JCXora01 openais[6538]: [TOTEM] previous ring seq 1204 rep 219.13.240.74
Jul 24 23:21:51 JCXora01 openais[6538]: [TOTEM] aru 30 high delivered 30 received flag 1
Jul 24 23:21:51 JCXora01 openais[6538]: [TOTEM] Did not need to originate any messages in recovery.
Jul 24 23:21:51 JCXora01 openais[6538]: [TOTEM] Sending initial ORF token

standby 上日志

Jul 24 23:20:14 JCXora02 openais[22792]: [TOTEM] The token was lost in the OPERATIONAL state.
Jul 24 23:20:14 JCXora02 openais[22792]: [TOTEM] Receive multicast socket recv buffer size (320000 bytes).
Jul 24 23:20:14 JCXora02 openais[22792]: [TOTEM] Transmit multicast socket send buffer size (320000 bytes).
Jul 24 23:20:14 JCXora02 openais[22792]: [TOTEM] entering GATHER state from 2.
Jul 24 23:20:34 JCXora02 openais[22792]: [TOTEM] entering GATHER state from 0.
Jul 24 23:20:34 JCXora02 openais[22792]: [TOTEM] Creating commit token because I am the rep.
Jul 24 23:20:34 JCXora02 openais[22792]: [TOTEM] Saving state aru 30 high seq received 30
Jul 24 23:20:34 JCXora02 openais[22792]: [TOTEM] Storing new sequence id for ring 4b8
Jul 24 23:20:34 JCXora02 openais[22792]: [TOTEM] entering COMMIT state.
Jul 24 23:20:34 JCXora02 openais[22792]: [TOTEM] entering RECOVERY state.
Jul 24 23:20:34 JCXora02 openais[22792]: [TOTEM] position [0] member 219.13.240.74:
Jul 24 23:20:34 JCXora02 openais[22792]: [TOTEM] previous ring seq 1204 rep 219.13.240.74
Jul 24 23:20:34 JCXora02 openais[22792]: [TOTEM] aru 30 high delivered 30 received flag 1
Jul 24 23:20:34 JCXora02 openais[22792]: [TOTEM] Did not need to originate any messages in recovery.
Jul 24 23:20:34 JCXora02 openais[22792]: [TOTEM] Sending initial ORF token

兄弟以为如何？

kkubsd

增加仲裁盘可以解决这个问题，好像RHCS仲裁盘会有问题，最好查一下有没有新的补丁。还有最好有CLVM去做。这个是RHCS推荐的。

pxf520

用heartbeat 做吧。心跳做2个，一个网线直连。一个串口