wohali commented on issue #1385: 3 Node Properly Configured Cluster, 1 Node get 'stolen', There seems to be no way to heal URL: https://github.com/apache/couchdb/issues/1385#issuecomment-397160730 Hi @seigel , we've got a [PR coming with improved documentation on this](https://github.com/apache/couchdb-documentation/pull/268) from @garbados. TL;DR: If your new node comes back *as Node 3*, with the same hostname, it will automatically get all of those shards re-replicated to it. This is the recommended approach, especially in AWS where transparently replacing a machine is easy. You just need to have your own DNS set up for these machines - don't use the AWS internal host names, because those are predicated on internal IP addresses you can't control. You really want host names like `node#.clustername.yourdomain.com`. For `q=9, n=3` you have 27 shards total, 9 per machine. Databases in CouchDB are "presharded," meaning that the shard map is fixed. Adding a fourth machine (without removing any machines on the network) will not result in any of those shards moving to the new machine to balance the load. There is a way to bring your new node back as node 4, add it to the cluster membership, and edit the shard map to ensure that a given database is supposed to have shards on that node. You can pre-seed those shards by copying them from other machines, or you can let Couch do all the work. Both approaches are discussed in the PR that has yet to land. Hope this help!
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
