JBOSS/Keycloak 集群等待 1 分钟再投票给 coordinator

Posted 2023-02-19

【中文标题】JBOSS/Keycloak 集群等待 1 分钟再投票给 coordinator

【英文标题】：JBOSS/Keycloak cluster wait 1 minute before voting for coordinator

【发布时间】：2021-08-13 12:53:13

【问题描述】：

我有 3 个节点的 keycloak 集群。如果有一个节点宕机，JBOSS 会在 1 分钟内启动协调器选择。由于停机时间，是否可以减少此超时？如何配置失败节点检测超时？

[root@keycloak-01 ~]# date; systemctl stop keycloak
Tue May 25 11:35:46 MSK 2021

11:36:58,629 INFO  [org.infinispan.CLUSTER] (VERIFY_SUSPECT.TimerThread-43,ejb,keycloak-02) ISPN000094: Received new cluster view for channel ejb: [keycloak-02|24] (2) [keycloak-02, keycloak-03]
11:36:58,630 INFO  [org.infinispan.CLUSTER] (VERIFY_SUSPECT.TimerThread-43,ejb,keycloak-02) ISPN100001: Node keycloak-01 left the cluster
11:36:58,772 INFO  [org.infinispan.CLUSTER] (non-blocking-thread--p7-t1) [Context=quartz] ISPN100007: After merge (or coordinator change), recovered members [keycloak-01, keycloak-02, keycloak-03] with topology id 104
11:36:58,774 INFO  [org.infinispan.CLUSTER] (non-blocking-thread--p7-t1) [Context=quartz] ISPN100008: Updating cache members list [keycloak-02, keycloak-03], topology id 105
11:36:58,808 INFO  [org.infinispan.CLUSTER] (non-blocking-thread--p15-t2) [Context=offlineClientSessions] ISPN100002: Starting rebalance with members [keycloak-02, keycloak-03], phase READ_OLD_WRITE_ALL, topology id 106

【参考方案1】：

我提议更加关注“故障检测”（FD 和 FD_ALL）in docs。我在以下人员的帮助下解决了我的任务：

<protocol type="FD_ALL">
  <property name="timeout">5000</property>
  <property name="interval">3000</property>
  <property name="timeout_check_interval">2000</property>
</protocol>