)]}'
{"doc/ha-guide/source/controller-ha-rabbitmq.rst":[{"author":{"_account_id":841,"name":"Akihiro Motoki","email":"amotoki@gmail.com","username":"amotoki"},"change_message_id":"3c91130465d745888d6c2bb59d4804be24d570f6","unresolved":false,"context_lines":[{"line_number":177,"context_line":"         \u0027pause_minority\u0027. Both \u0027ignore\u0027 and \u0027autoheal\u0027 will cause"},{"line_number":178,"context_line":"         troubles like message duplication, inconsistent and lost when"},{"line_number":179,"context_line":"         the cluster nodes is in network partitions or rejoin, and"},{"line_number":180,"context_line":"         some nova and neutron services will fail their processing."},{"line_number":181,"context_line":""},{"line_number":182,"context_line":".. _rabbitmq-services:"},{"line_number":183,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"3a29b11f_2975a430","line":180,"updated":"2015-10-23 10:24:02.000000000","message":"Selecting cluster_partition_handling highly depends on requirements of individual deployments, and it is based on CAP theorem: availability, data integrity vs partition tolerance.\n\nAssume you have only three nodes for RabbitMQ cluster. Does a RabbitMQ cluster continues to work when two nodes fail? In my understanding, in pause_minority mode, minor node(s) will stop transmitting messages. If two of three nodes in the cluster fail, the remaining one node cannot get the majority (because the cluster size must be \u003e\u003d 2 to get the majority). As a result, a whole RabbitMQ cluster will stop even though one node is alive.\nI think it is a big and important notice when using pause_minority mode.\n\nIn the autoheal mode, we cannot guarantee the exact data integrity but it continues to work even though two of three nodes fail. It works in most cases in OpenStack deployments.\n\nIt seems better to describe more pros and cons.","commit_id":"47491a089d0c3a2d9405cdc290044bbd952d572c"},{"author":{"_account_id":6926,"name":"Bogdan Dobrelya","email":"bdobreli@redhat.com","username":"bogdando"},"change_message_id":"46d25943ff992885eee8848cf898e7d8bd21546d","unresolved":false,"context_lines":[{"line_number":177,"context_line":"         \u0027pause_minority\u0027. Both \u0027ignore\u0027 and \u0027autoheal\u0027 will cause"},{"line_number":178,"context_line":"         troubles like message duplication, inconsistent and lost when"},{"line_number":179,"context_line":"         the cluster nodes is in network partitions or rejoin, and"},{"line_number":180,"context_line":"         some nova and neutron services will fail their processing."},{"line_number":181,"context_line":""},{"line_number":182,"context_line":".. _rabbitmq-services:"},{"line_number":183,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a7b0d38_67ade4e8","line":180,"in_reply_to":"3a29b11f_2975a430","updated":"2015-12-23 14:43:14.000000000","message":"+1 for neutral describing of use cases with cons/pros. No recommendations should be done for the cluster_partition_handling, perhaps unless there will be the merge policy introduced.","commit_id":"47491a089d0c3a2d9405cdc290044bbd952d572c"}]}