<div style="line-height:1.7;color:#000000;font-size:14px;font-family:Arial"><div>Hi Joel,</div><div><br /></div><div>For the first question, we have another component in our system implements paxos and provide distribute lock. <br /></div><div>When the network connection between two nodes is disconnected, the first node that can get the distributed lock</div><div>will continue to provide services, so we can handle this case.</div><div><br /></div><div>For the second question, we have to use iSCSI in my case.<br /></div><div><br /></div><div>For the last question, we have implemented our own cluster manager, the first version for failover is same with what you suggested. <br /></div><div>But in my use-case, it introduce much IO-non-response time in failover. So we want to provide two active iscsi pathes for client then <br /></div><div>the client can do failover by his multipath software.
<br /></div><div><br /></div><div>Best regards,</div><div>Xu<br /></div><br /><div style="position:relative;zoom:1"></div><br /><pre><br />From: Joel Colledge <joel.colledge@linbit.com>
Date: 2022-05-10 20:55:15
To: "rui.xu" <rui.xu@easystack.cn>
Cc: Philipp Reisner <philipp.reisner@linbit.com>,drbd-dev@lists.linbit.com,dongsheng.yang@easystack.cn
Subject: Re: Re: drbd: a proposal of two-primaries mode>Hi Xu,
>
>Dual primary mode is not recommended for this scenario due to data
>divergence considerations. How do you handle a loss of network
>connectivity between the storage nodes?
>
>Is it possible to use diskless DRBD clients instead of iSCSI
>initiators for your purposes?
>
>Assuming you have to use iSCSI, the normal solution is to only have
>one active target. Use a cluster manager to support failover to the
>other storage node. For instance, add a third node with minimal
>resources as a diskless tiebreaker, enable DRBD quorum, and use the
>drbd-reactor promoter as a simple cluster manager. Have you considered
>this option?
>
>Best regards,
>Joel
</pre></div><br>