On 3/13/2022 5:27 PM, dmitri maziuk wrote:
If you do the former, you could probably put the index on an iscsi volume and have it remounted on the standby host/container as part of the failover. If you do the latter, might as well do what Shawn said (but then of course "now you have two problems" because the proxy host can fail too).
My preferred solution for a minimalist SolrCloud install would be to run ucarp or pacemaker on all 3 hosts, handling a virtual IP address. If it's pacemaker, have it also manage which machine is running haproxy for load balancing. If it's ucarp, haproxy would be always running on all 3 hosts, because ucarp is a very simple program. Two of the hosts would run ZK and Solr in addition to all that, the third machine would run ZK but not Solr. The third machine would be the preferred host for the IP address and haproxy, but those would also be able to transfer to one of the other two if that host goes down.
Each Solr instance would have a full copy of the data in TLOG replicas. That way there would be two completely identical copies of all index data. Trying to shuffle an iSCSI volume between hosts sounds like a very brittle setup to me.
In my experience, pacemaker and haproxy are very stable when correctly configured.
Thanks, Shawn
