[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: status of multi-master replication?

Nir Soffer wrote: 
That's why I said that you can use the multimaster in capability in
conjunction with a heartbeat mechanism. This has the added value that
when the failed node resumes service, you can use it's replication log
to restore your lost data (Even automatically).

I think :)

Right! I agree with you! What I'm saying is that multimaster functionality shouldn't and wouldn't rely on a lower layer implementation to provide multimaster capability.

Multimasterness is a feature of the slapd application, and wouldn't rely on an HA implementation to resolve the differences of object states that Howard was speaking of. slapd the application would be soley responsible for resolving these differences outside of any HA implementation.

I think what you're proposing is different than multimaster. The ldap servers in your HA cluster would resolve differences by using the replog, and would promoted to masters or demoted to slaves by using features of the HA cluster (a workable solution depending on your requirements), whereas multimaster slapd servers would resolve differences on their own within the application, and both would be masters at the same time, sharing updates between each other.

This might enable you to put a layer 4 load balancing devices in front of your ldap servers and seamlessly add and remove slapd masters from your network (never done that myself). But in this senario the layer 4 device is not tied to multimaster functionality at all, it is just an HA/loadbalancing implementation.

Regards,
Elliot