meta level

[acelyc111]

enum meta_function_level
{
    // there are 4 ways to modify the meta-server's status:
    // 0. DDL operation: create/drop/recall table
    // 1. downgrade primary when dectect it is not alive
    // 2. accept primary's update-request to kickoff some secondaries
    // 3. make balancer proposal, which further trigger 2
    // according to these ways, we give meta several active level.

    fl_stopped = 100, //we don't take any action to modify the meta's status, even the DDL operations are not responsed
    fl_blind = 200, //only DDL operations are responsed, 1 2 3 are just ignored
    fl_freezed = 300, //0 1 are responsed, 2 3 ignored
    fl_steady = 400, //0 1 2 are responsed, don't do any balancer
    fl_lively = 500, //full functional
    fl_invalid = 10000
}

update_configuration_on_remote触发条件：

1. drop_partition时
   • 删表时
   • 初始化时，从remote同步表状态
2. downgrade_primary_to_inactive时
   • 节点挂掉时，将其上的primary副本置为inactive
3. downgrade_stateless_nodes时
   • 节点挂掉时，将其上无状态表（是啥？）的副本置为inactive
4. on_update_configuration时
   • 在meta server收到RPC_CM_UPDATE_PARTITION_CONFIGURATION请求，且level>fl_freezed时，会转调server_state::on_update_configuration
   • 它由replica server的update_configuration_on_meta_server发出，会在以下时机触发：
     • assign_primary
     • CT_ASSIGN_PRIMARY: 新建表时，DDD时，shell
     • CT_UPGRADE_TO_PRIMARY: rebalance时，move_primary、copy_primary
     • upgrade_to_secondary_on_primary: CT_UPGRADE_TO_SECONDARY
     • potential secondary -> secondary
     • downgrade_to_secondary_on_primary: CT_DOWNGRADE_TO_SECONDARY
     • move_primary、copy_primary、downgrade_to_secondary
     • downgrade_to_inactive_on_primary: CT_DOWNGRADE_TO_INACTIVE
     • downgrade_primary_to_inactive、downgrade_secondary_to_inactive
     • remove: CT_REMOVE
     • copy_primary、copy_secondary、on_redundant_secondary、downgrade_stateless_nodes、shell
     • on_config_sync中initializing: CT_PRIMARY_FORCE_UPDATE_BALLOT

小于等于fl_stopped

没有直接使用

小于等于fl_blind

当前level低于fl_blind时，将会一直pending，直到level满足条件时会执行_meta_svc->get_remote_storage()->set_data 超时时也会执行重试

• update_configuration_on_remote 更新partition的配置信息
• update_partition_max_replica_count_on_remote 设置表的最大副本数

小于等于fl_freezed

忽略on_update_configuration请求 忽略create app请求 忽略check_all_partitions

• 由定时任务meta_service::balancer_run发起，受参数lb_interval_ms控制，默认10秒 忽略set_max_replica_count

小于等于fl_steady

设置为这些时，将清理所有表的proposal队列 on_config_sync时

• 对于已删除表，将忽略过期表分片的GC
• 对于健康表，将忽略“无用”（由collect_replica判断）分片的GC check_all_partitions时
• （小于fl_steady直接不执行）等于fl_steady时只做report

等于fl_lively

做rebalance

[acelyc111]

在freezed状态只要放过CT_UPGRADE_TO_PRIMARY消息就能避免这个问题