I happen to be working on this issue with Mr. Pyle. Is there anybody out there successfully working with Artemis colocated backups and scale-down that are able to recover a failed node successfully, and have the recovered node reestablish a backup? if so we would love to compare configuration notes. is this a type of configuration that we should expect to work or are we possibly mixing together many different and incompatible concepts? would it be appropriate for us to create a jira issue with the issue we are seeing?
On Fri, Jan 25, 2019 at 11:54 AM Jason Pyle <seth.pyl...@gmail.com> wrote: > I sent this last week via Nabble but I don't think it got mailed out to > anybody, if it did I apologize for the spam. > > We're developing a strategy for backups and HA. Ideally we'd like to use > colocated backups to ensure data integrity and availability with scale-down > configured from the slave to the master host. > > We ran into an issue when bringing a server back up, consider this > situation. > > Servers 1 and 2 are brought up, and make colocated backups 1b and 2b. 1b > existing on server 2 and 2b existing on server 1. If I bring server 2 > offline, 2b comes online then scales down into server 1 as intended. When I > bring server 2 back up, 2b does not failback. This leads to server 2 > starting an infinite vote loop to find another server to create a backup > for > it. Since server 1 already possesses backup 2b and is only configured for 1 > backup it will infinitely reply that it does not have space for another > backup. > > In this state, if more messages are sent to server 2 and server 2 > experiences a crash those messages are lost. > > I've created an example of this problem based on one of the examples in the > artemis source here > https://github.com/SethPyle376/colocated-scaledown-problem > > I've tested this situation with both replication and shared-store and the > problem persists. Any help would be great, we need colocated scaledown > failback working correctly. > -- Dan Langford 801.683.0213 http://google.com/profiles/danlangford
