We are using XDCR for Disaster recovery, but we found it to be really usefull to split “memcache” operations and “views” operations.
Our “memcache Ops” are typically front and respond to synchronous needs;
They are running on the XDCR source cluster
The “views Ops” are much more for backoffice and asynchronous purpose, and are running on the destination cluster.
Most of our views a running on large set of data, where each item can be updated 1000+/Minute
Is there any way to make XDCR buffer the items to be replicated n seconds, meaning that only the last update for each item will be sent to destination, and that we will save a lot of CPU and IO on views indexing and compaction ?
We understand that is doesn’t fit well with real time disaster recovery, but it could make a lot of sense for splitting “memcache” and “views” Ops while significantly reducing the CPU/IO needs, and making the indexing much faster