update for #68: add node architecture

QingCloudAppcenter · Feb 11, 2020 · be5244e · be5244e
1 parent 37e216b
commit be5244e
Showing 8 changed files with 37 additions and 8 deletions.
diff --git a/ansible/roles/redis-standalone/files/etc/confd/templates/redis-server.sh/01.common.func.tmpl b/ansible/roles/redis-standalone/files/etc/confd/templates/redis-server.sh/01.common.func.tmpl
@@ -0,0 +1,11 @@
+getSlavePriority(){
+  local mySid="{{ getv "/host/sid" }}"
+  local ourSid="$(echo "{{ range gets "/hosts/*/sid" }}
+  {{- printf .Value }} 
+  {{ end }}" |sort -n)"
+  local sid counter=0; for sid in $ourSid;do
+    counter=$(($counter+1))
+    [[ "$sid" == "$mySid" ]] && break
+  done
+  if [[ $counter -le 3 ]];then echo 100; else echo 0;fi
+}
diff --git a/...emplates/redis-server.sh/01.node.env.tmpl → ...emplates/redis-server.sh/02.node.env.tmpl b/...emplates/redis-server.sh/01.node.env.tmpl → ...emplates/redis-server.sh/02.node.env.tmpl
@@ -5,7 +5,7 @@
 flush /opt/app/bin/envs/node-redis-standalone.env << REDIS_ENV_EOF
 SERVICES="\$SERVICES $(echo "
 redis-server/true/tcp:{{ getv "/env/port" "6379" }}
-redis-sentinel/{{ eq (len $allHosts) 3 }}/tcp:26379
+redis-sentinel/$(if {{ ge (len $allHosts) 3 }} && [[ $(getSlavePriority) -eq 100 ]];then echo "true";else echo "false";fi)/tcp:26379
 " | xargs)"
 NODE_CTL="redis-standalone"
 MY_NODE_ID="{{ getv "/host/node_id" }}"

diff --git a/...edis-server.sh/02.redis.changed.conf.tmpl → ...edis-server.sh/03.redis.changed.conf.tmpl b/...edis-server.sh/02.redis.changed.conf.tmpl → ...edis-server.sh/03.redis.changed.conf.tmpl
@@ -25,7 +25,9 @@ renameCommands() {
   done
 }
 
+
 flush /opt/app/conf/redis-standalone/redis.changed.conf << REDIS_CHANGED_CONF_EOF
+slave-priority $(getSlavePriority)
 databases {{ getv "/env/databases" "16" }}
 activerehashing {{ getv "/env/activerehashing" "yes" }}
 appendfsync {{ getv "/env/appendfsync" "everysec" }}

diff --git a/ansible/roles/redis-standalone/files/opt/app/bin/node/redis-standalone.sh b/ansible/roles/redis-standalone/files/opt/app/bin/node/redis-standalone.sh
@@ -2,6 +2,7 @@ FIND_MASTER_IP_ERR=210
 LEAVING_REDIS_NODES_IS_NONE_ERR=211
 LEAVING_REDIS_NODES_INCLUDE_MASTER_ERR=212
 MASTER_SALVE_SWITCH_WHEN_DEL_NODE_ERR=213
+LEAVING_REDIS_NODES_INCLUDE_SENTINEL_NODE=214
 REDIS_COMMAND_EXECUTE_FAIL_ERR=220
 RDB_FILE_EXIST_ERR=221
 RDB_FILE_CHECK_ERR=222
@@ -93,6 +94,16 @@ checkMasterNotLeaving() {
 preScaleIn() {
   [ -n "$LEAVING_REDIS_NODES" ] || return $LEAVING_REDIS_NODES_IS_NONE_ERR
   checkMasterNotLeaving
+
+  # 防止最终计划剩余3个节点的时候，删除了 sentinel 节点
+  local leavingNodesCount; leavingNodesCount=$(echo "$LEAVING_REDIS_NODES" |wc -w)
+  local allNodesCount; allNodesCount=$(echo "$REDIS_NODES" |wc -w)
+  if [[ $(($allNodesCount-$leavingNodesCount)) -ge 3 ]];then
+    local firstToThirdNode="$(eval echo "$allNodesCount" |xargs -n1| sort -n -t'/' -k1| xargs|cut -d" " -f1-3)"
+    local node;for node in $firstToThirdNode;do
+      echo "$LEAVING_REDIS_NODES" |grep -vq "$node" || return $LEAVING_REDIS_NODES_INCLUDE_SENTINEL_NODE
+    done
+  fi
 }
 
 destroy() {
@@ -410,9 +421,10 @@ check(){
 }
 
 getRedisRoles(){ 
-  local node nodeIp myRole; for node in $REDIS_NODES; do
+  local node nodeIp myRole counter=0; for node in $(echo "$REDIS_NODES" |sort -n -t"/" -k1); do
+    counter=$(($counter+1))
     nodeIp="$(echo "$node" |cut -d"/" -f3)"
-    myRole="$(runRedisCmd --ip "$nodeIp" role | head -n1)"
-    echo "$nodeIp $myRole"
-  done | jq -Rc 'split(" ") | [ . ]' | jq -s add | jq -c '{"labels":["ip","role"],"data":.}'
+    myRole="$(runRedisCmd --ip "$nodeIp" role | head -n1 || echo "unknown")"
+    echo "$nodeIp $myRole $(if [[ $counter -le 3 ]];then echo "true"; else echo "false";fi)"
+  done | jq -Rc 'split(" ") | [ . ]' | jq -s add | jq -c '{"labels":["ip","role","disabled_delete"],"data":.}'
 }
diff --git a/ansible/roles/redis-standalone/files/opt/app/conf/redis-standalone/redis.default.conf b/ansible/roles/redis-standalone/files/opt/app/conf/redis-standalone/redis.default.conf
@@ -17,7 +17,6 @@ rdbchecksum yes
 rdbcompression yes
 repl-disable-tcp-nodelay no  
 save "" 
-slave-priority 100
 slave-read-only yes
 slave-serve-stale-data yes 
 slowlog-max-len 128 

diff --git a/app-standalone/cluster.json.mustache b/app-standalone/cluster.json.mustache
@@ -17,7 +17,7 @@
     "container": {
       "type": "lxc",
       "zone": "pek3a",
-      "image": "img-tb8o5ojo"
+      "image": "img-x6cv1kyp"
     },
     "instance_class": {{cluster.node.instance_class}},
     "count": {{cluster.node.count}},

diff --git a/app-standalone/config.json b/app-standalone/config.json
@@ -98,7 +98,10 @@
               "default": 1,
               "range": [
                 1,
-                3
+                3,
+                5,
+                7,
+                9
               ],
               "required": "yes"
             }

diff --git a/app-standalone/locale/zh-cn.json b/app-standalone/locale/zh-cn.json
@@ -122,6 +122,7 @@
   "err_code211": "待删除节点为空，禁止删除操作，请通过工单与我们沟通",
   "err_code212": "为了避免数据丢失，请选择从节点进行删除，主从关系可以通过监控项获取",
   "err_code213": "刚才可能发生了主从切换，导致要删除的节点中包含主节点；为了避免数据丢失，请稍后选择从节点进行删除",
+  "err_code214": "剩余的节点数大于或等于三个节点时，我们有限制其中三个节点禁止删除，禁止删除的节点可以从「节点实时角色」获取",
   "err_code220": "redis 命令执行失败，请通过工单与我们沟通",
   "err_code221": "RDB 文件不存在，请您先上传 RDB 文件",
   "err_code222": "RDB 文件格式有误，请您上传正确的 RDB 文件",
@@ -149,6 +150,7 @@
   "roles": "节点实时角色",
   "ip": "节点 IP",
   "role": "节点角色",
+  "disabled_delete": "删除后节点数大于等于3，禁止删除",
   "confirm": "确认操作？",
   "All existing data will be erased, and please ensure there is only one redis node; choose 'yes' to continue": "该操作仅支持单节点集群，且会清空现有集群的所有数据，数据需要自行备份，yes 表示确认并继续执行，no 表示不执行",
   "params": "参数",