es出现replica shard unassign

基于es做日志的存储，对于历史index（没有新数据写入）会做replica数量调整，副本数从0调整为1。但是经常会发现部分index，replica shard不能assign。
调用 /_cluster/allocation/explain 接口，发现unassign状态的shard都是可以被分配的。
例如：

{

  "index": "billions-bplus-alarm-dynamic-biz-sketch-dao-@2018.02.24",

  "shard": 8,

  "primary": false,

  "current_state": "unassigned",

  "unassigned_info": {

    "reason": "REPLICA_ADDED",

    "at": "2018-02-26T17:30:04.428Z",

    "last_allocation_status": "no_attempt"

  },

  "can_allocate": "yes",

  "allocate_explanation": "can allocate the shard",

  "target_node": {

    "id": "Bp7Dya1_TDG6e79NkR99fQ",

    "name": "hzxs-billions-es-03-datanode_stale",

    "transport_address": "172.19.20.221:9301",

    "attributes": {

      "box_type": "stale"

    }

  },

  "node_allocation_decisions": [

    {

      "node_id": "DGiC5DK1SMK4cqew3iWvpA",

      "node_name": "hzxs-billions-es-18-datanode_stale",

      "transport_address": "172.19.20.236:9301",

      "node_attributes": {

        "box_type": "stale"

      },

      "node_decision": "yes",

      "weight_ranking": 23

    },

    {

      "node_id": "LLVHrovaRiek8A4cYqtjfQ",

      "node_name": "hzxs-billions-es-19-datanode_stale",

      "transport_address": "172.19.20.237:9301",

      "node_attributes": {

        "box_type": "stale"

      },

      "node_decision": "yes",

      "weight_ranking": 24

    },

    {

      "node_id": "Bp7Dya1_TDG6e79NkR99fQ",

      "node_name": "hzxs-billions-es-03-datanode_stale",

      "transport_address": "172.19.20.221:9301",

      "node_attributes": {

        "box_type": "stale"

      },

      "node_decision": "yes",

      "weight_ranking": 25

    },

这个unassign状态一直持续，不能恢复。 es的版本是5.4.3， 大家有遇到过类似的问题吗？