On Mon, Feb 6, 2017 at 5:31 PM, 克赛号0181 <d...@dtdream.com> wrote: > In K8S, Pacemaker is not used usally, we use keepalived to implement the VIP > (OVN service endpoint)，when master nodes fails, the VIP switched earlier > than we notice such event, > so, the new master node will connect itself in a short time(about 3-10 > seconds). > > Thanks for explaining. This is the first time we heard of a use case without pacemaker.
Regardless how it happens, I agree it is desirable to prevent data loss when ovsdb-server tries to replicate itself. The implementation of the patch is pretty clever, in that the reply of the monitor request is used to preserve the database content in case it tries to replicate itself. However, I see two two potential drawbacks to this approach. 1. In case the replication set up excludes tables, or columns, the excluded data will be lost. 2. In general, to guard against undesirable configurations, it is usually more desirable to reject them (and log errors) than to silently 'fix' the problem. I realize those drawbacks are not critical, and may not even be relevant in some deployments, but it would be nice avoid them. Ben suggested a method that an OVSDB server can detect self replicating. I just posted a 3 patch series that tries to implement this idea as an alternative solution. https://mail.openvswitch.org/pipermail/ovs-dev/2017-February/328587.html If interested, please take a look and provide any feedback you may have. Thanks, --andy _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev
