Master and backup role election strategy in kept

stay Keepalived colony in , In fact, there is no strict sense of the Lord 、 For the node , Although you can Keepalived Set... In the configuration file “state” The options are “MASTER” state , However, this does not mean that this node has always been Master role . The role of the control node is Keepalived In the configuration file “priority” value , But it does not control the role of all nodes , Another way to change the role of nodes is in vrrp_script Module “weight” value , Both options correspond to an integer value , among “weight” The value can be a negative integer , A node is at colony The role in is determined by the size of these two values .

One active and multiple standby Keepalived colony in ,“priority” The one with the largest value will be the one in the cluster Master node , And everything else is Backup node . stay Master After the node fails ,Backup Between nodes “ Democratic elections ”, By setting the node priority value “priority” and ““weight” The calculation of , Select new Master The node takes over the cluster service .

stay vrrp_script Module , If not set “weight” Option value , Then the cluster priority will be selected by Keepalived In the configuration file “priority” Values determine , When it is necessary to flexibly control the priority in the cluster , It can be done by vrrp_script Set in the module “weight” It's worth it . Here is an example to illustrate .

Suppose there is A and B Two nodes Keepalived colony , stay A node keepalived.conf In file , Set up “priority” The value is 100, And in the B node keepalived.conf In file , Set up “priority” The value is 80, also A、B Both nodes use “vrrp_script” Module to monitor MySQL service , At the same time “weight” The value is 10, Then the following will happen .

Start on both nodes Keepalived After service , The normal situation is A Nodes will become Master node , and B Automatically become Backup node , At this time will be A Node mysql Service closing , Discover by looking at the log , It didn't show up B Node takeover A Log of node ,B The node is still in Backup state , and A The node is still Master state , In this case, the whole HA Cluster will lose its meaning .

Now let's analyze the reasons for this situation , This is the same. Keepalived Master in cluster 、 Prepare for the problem of role election strategy . The following is a summary of Keepalived Use in vrrp_script Module is the election algorithm of the whole cluster role , because “weight” The value can be positive or negative , therefore , There are two situations to explain .

1.  “weight” When the value is positive

stay vrrp_script If the script specified in is detected successfully , that Master The weight of the node will be “weight Value and ”priority“ Sum of values , If script detection fails , that Master The weight of the node remains “priority” value , Therefore, the switching strategy is ：

Master node “vrrp_script” When script detection fails , If Master node “priority” Less than Backup node “weight Value and ”priority“ Sum of values , The main 、 Standby switching .

Master node “vrrp_script” When the script detection is successful , If Master node “weight” Value and “priority” The sum of values is greater than Backup node “weight” Value and “priority” Sum of values , The master node is still the master node , No switching .

2.  “weight” When the value is negative

stay “vrrp_script” If the script specified in is detected successfully , that Master The weight of the node is still “priority” value , When script detection fails ,Master The weight of the node will be “priority“ Value and “weight” The difference in value , Therefore, the switching strategy is ：

Master node “vrrp_script” When script detection fails , If Master node “priority” Value and “weight” The difference between values is less than Backup node “priority” value , The main 、 Standby switching .

Master node “vrrp_script” When the script detection is successful , If Master node “priority” Greater than Backup node “priority” When the value of , The master node is still the master node , No switching .

In familiar with Keepalived Lord 、 Prepare for the role of the election strategy , Let's analyze the example just now , because A、B Two nodes are set “weight” Values are 10, Therefore, the first kind of election strategy , stay A Nodes stop Mysql After service ,A Script detection of the node will fail , here A The weight of the node will remain A Set on the node “priority” value , That is to say 100, and B The weight of the node will change to “weight” Value and “priority” Sum of values , That is to say 90（10+80）, That's what happened A The node weight is still greater than B Node weights , Therefore, there will be no master 、 Standby switching .

about “weight” Value settings , There is a simple criterion , namely “weight” The absolute value of value should be greater than Master and Backup node “priority” The difference in value . For the top A、B Examples of two nodes , Just set “weight” Greater than 20 It can ensure the normal operation and switching of the cluster . thus it can be seen , about “weight Value settings , Be very careful , If the settings are not good , Will cause the cluster role election to fail , Paralyze the cluster .