fix: leader should switchover first before scaling in (#3301)

apecloud · May 18, 2023 · b5abe41 · b5abe41
1 parent 9cff6ca
commit b5abe41
Showing 1 changed file with 24 additions and 7 deletions.
diff --git a/deploy/apecloud-mysql/templates/scripts.yaml b/deploy/apecloud-mysql/templates/scripts.yaml
@@ -58,12 +58,7 @@ data:
         cluster_info="$cluster_info;";
       fi;
       tmp_host=$(eval echo \$KB_MYSQL_"$i"_HOSTNAME)
-      # setup pod weight, prefer pod 0 to be leader
-      if [ $i -eq 0 ]; then 
-        cluster_info="$cluster_info$tmp_host:13306#9N";
-      else
-        cluster_info="$cluster_info$tmp_host:13306#1N";
-      fi
+      cluster_info="$cluster_info$tmp_host:13306";
     done; 
     cluster_info="$cluster_info@$(($idx+1))"; 
     echo "cluster_info=$cluster_info"; 
@@ -156,7 +151,6 @@ data:
   pre-stop.sh: |
     #!/bin/bash
     drop_followers() {
-    leader=`cat /etc/annotations/leader`
     echo "leader=$leader" >> /data/mysql/.kb_pre_stop.log
     echo "KB_POD_NAME=$KB_POD_NAME" >> /data/mysql/.kb_pre_stop.log
     if [ -z "$leader" -o "$KB_POD_NAME" = "$leader" ]; then
@@ -178,11 +172,34 @@ data:
     echo "mysql $host_flag -uroot $password_flag -e \"call dbms_consensus.drop_learner('$host:13306');\" 2>&1 " >> /data/mysql/.kb_pre_stop.log
     mysql $host_flag -uroot $password_flag -e "call dbms_consensus.drop_learner('$host:13306');" 2>&1
     }
+    switchover() {
+      if [ ! -z $MYSQL_ROOT_PASSWORD ]; then 
+        password_flag="-p$MYSQL_ROOT_PASSWORD"
+      fi
+      new_leader_host=$KB_MYSQL_0_HOSTNAME
+      if [ "$KB_POD_NAME" = "$leader" ]; then
+        echo "self is leader, need to switchover" >> /data/mysql/.kb_pre_stop.log
+        echo "mysql -uroot $password_flag -e \"call dbms_consensus.change_leader('$new_leader_host:13306');\" 2>&1" >> /data/mysql/.kb_pre_stop.log
+        mysql -uroot $password_flag -e "call dbms_consensus.change_leader('$new_leader_host:13306');" 2>&1
+        sleep 1
+        role_info=`mysql -uroot $password_flag -e "select * from information_schema.wesql_cluster_local;" 2>&1`
+        echo "role_info=$role_info" >> /data/mysql/.kb_pre_stop.log
+        is_follower=`echo $role_info | grep "Follower"`
+        if [ ! -z "$is_follower" ]; then
+          echo "new_leader=$new_leader_host" >> /data/mysql/.kb_pre_stop.log
+          leader=`echo "$new_leader_host" | cut -d "." -f 1`
+          idx=${KB_POD_NAME##*-}
+        fi
+      fi
+    }
+    leader=`cat /etc/annotations/leader`
     idx=${KB_POD_NAME##*-}
     current_component_replicas=`cat /etc/annotations/component-replicas`
     echo "current replicas: $current_component_replicas" >> /data/mysql/.kb_pre_stop.log
     if [ ! $idx -lt $current_component_replicas ] && [ $current_component_replicas -ne 0 ]; then 
         # if idx greater than or equal to current_component_replicas means the cluster's scaling in
+        # switch leader before leader scaling in itself
+        switchover
         # only scaling in need to drop followers
         drop_followers
     else