Clickhouse RepicateMergeTree Insert 副本协同流程

1、选择一个远端的副本，下载数据 副本选择算法： ①、从/replicas节点找到所有副本 ②、选择其中log_pointer最大，同时/queue中任务最少的副本2、向远端副本发送下载请求Fetch part 202106_0_0_0 from replicas/chuanmei-backup1.jd.163.org3、如果下载失败，则尝试n次max_fetch_partition_retries_count=n

replicate_tb_1

返回分区数据Sending part 202106_0_0_0

1、由insert的副本发送日志2、推送操作日志：log/log-0000000001Entry对象：format version: 4create_time: 2021-06-27 20:59:11source replica: chuanmei-backup1.jd.163.orgblock_id: 20210627_10257022201049994224_23.....type:getpartition_name:20210627_281944_281944_0

Create table

Insert value

distribute log

create local data

202106_0_0_0

tmp_fetch  202106_0_0_0

response log

1、本地新建分区202106_0_0_02、写入block_id/blocks/20210627_10257022201049994224_2370951077480815403

CK_Node2

rename

1、初始化节点/clickhouse/tables/shard01/table_replicate_tb_1_local2、注册副本节点/replicas/chuanmei-backup1.jd.163.org3、监听/log节点4、选举，第一个将会是主副本/leader_election/leader_election-0000000000

download data

Zookeeper

接收到服务调用？endpoint=DataPartsExchange

CK_Node1

1、注册副本节点/replicas/chuanmei-backup2.jd.163.org2、监听/log节点3、选举/leader_election/leader_election-0000000001

1、监听到新的操作日志，log/log-00000000012、拉起日志对象Entry3、创建task任务，并放入执行队列/replicas/chuanmei-backup2.jd.163.org/queue/4、基于/queue/队列，执行任务