ioredis client should not attempt to connect to IP addresses from cluster configuration

[bsnote]

I configured Redis cluster consisting of 2 nodes running on the same machine:

redis.conf for 1st node:

bind 0.0.0.0
port 6379
cluster-enabled yes
cluster-config-file nodes.conf
cluster-node-timeout 5000

redis.conf for 2nd node:

bind 0.0.0.0
port 6380
cluster-enabled yes
cluster-config-file nodes.conf
cluster-node-timeout 5000

redis-cli -p 6379 cluster addslots $(seq 0 8191)
redis-cli -p 6380 cluster addslots $(seq 8192 16383)
redis-cli -p 6380 cluster meet 127.0.0.1 6379

redis-cli -p 6379 cluster slots returns the following result:

1) 1) (integer) 8192
   2) (integer) 16383
   3) 1) "127.0.0.1"
      2) (integer) 6380
      3) "c8e9a312411cefdd8ee4f74c923e5b7099db5edd"
2) 1) (integer) 0
   2) (integer) 8191
   3) 1) "127.0.0.1"
      2) (integer) 6379
      3) "e44691fe2b1adb2772fd5114d577dab105482f32"

Now running DEBUG=ioredis:* node 1 on a different machine where

// 1.js
var Redis = require('ioredis');

var cluster = new Redis.Cluster([{
  port: 6379,
  host: '10.2.19.39'
}]);

shows the following:

  ioredis:cluster status: [empty] -> connecting +0ms
  ioredis:redis status[10.2.19.39:6379]: [empty] -> wait +6ms
  ioredis:cluster getting slot cache from 10.2.19.39:6379 +3ms
  ioredis:redis status[10.2.19.39:6379]: wait -> connecting +3ms
  ioredis:redis queue command[0] -> cluster(slots) +1ms
  ioredis:redis status[10.2.19.39:6379]: connecting -> connect +7ms
  ioredis:redis write command[0] -> info() +1ms
  ioredis:redis status[10.2.19.39:6379]: connect -> ready +7ms
  ioredis:connection send 1 commands in offline queue +0ms
  ioredis:redis write command[0] -> cluster(slots) +1ms
  ioredis:redis status[127.0.0.1:6380]: [empty] -> wait +4ms
  ioredis:redis status[127.0.0.1:6379]: [empty] -> wait +0ms
  ioredis:cluster status: connecting -> connect +1ms
  ioredis:redis status[127.0.0.1:6380]: wait -> connecting +1ms
  ioredis:redis queue command[0] -> cluster(info) +0ms
  ioredis:connection error: Error: connect ECONNREFUSED 127.0.0.1:6380 +2ms
  ioredis:redis status[127.0.0.1:6380]: connecting -> close +1ms
  ioredis:connection skip reconnecting because `retryStrategy` is not a function +0ms
  ioredis:redis status[127.0.0.1:6380]: close -> end +0ms
  ioredis:redis status[10.2.19.39:6379]: ready -> close +6ms
  ioredis:connection skip reconnecting since the connection is manually closed. +0ms
  ioredis:redis status[10.2.19.39:6379]: close -> end +0ms
  ioredis:redis status[127.0.0.1:6379]: wait -> connecting +0ms
  ioredis:redis status[127.0.0.1:6379]: connecting -> connect +1ms
  ioredis:redis write command[0] -> info() +0ms
  ioredis:redis status[127.0.0.1:6379]: connect -> ready +1ms
  ioredis:cluster getting slot cache from 127.0.0.1:6379 +97ms
  ioredis:redis write command[0] -> cluster(slots) +0ms
  ioredis:delayqueue send 1 commands in failover queue +2ms
  ioredis:redis queue command[0] -> cluster(info) +1ms
  ioredis:redis status[127.0.0.1:6379]: ready -> close +0ms
  ioredis:connection skip reconnecting since the connection is manually closed. +0ms
  ioredis:redis status[127.0.0.1:6379]: close -> end +0ms
  ...

As it can be seen it's trying to connect to 127.0.0.1 which is wrong as this address is used for inter-node communication and is not necessarily the one that clients use. I can make it work by running this command:

redis-cli -p 6380 cluster meet 10.2.19.39 6379

After that redis-cli -p 6379 cluster slots returns the following result:

1) 1) (integer) 8192
   2) (integer) 16383
   3) 1) "10.2.19.39"
      2) (integer) 6380
      3) "c8e9a312411cefdd8ee4f74c923e5b7099db5edd"
2) 1) (integer) 0
   2) (integer) 8191
   3) 1) "10.2.19.39"
      2) (integer) 6379
      3) "e44691fe2b1adb2772fd5114d577dab105482f32"

and DEBUG=ioredis:* node 1 works properly:

  ioredis:cluster status: [empty] -> connecting +0ms
  ioredis:redis status[10.2.19.39:6379]: [empty] -> wait +5ms
  ioredis:cluster getting slot cache from 10.2.19.39:6379 +4ms
  ioredis:redis status[10.2.19.39:6379]: wait -> connecting +3ms
  ioredis:redis queue command[0] -> cluster(slots) +1ms
  ioredis:redis status[10.2.19.39:6379]: connecting -> connect +6ms
  ioredis:redis write command[0] -> info() +1ms
  ioredis:redis status[10.2.19.39:6379]: connect -> ready +7ms
  ioredis:connection send 1 commands in offline queue +0ms
  ioredis:redis write command[0] -> cluster(slots) +0ms
  ioredis:redis status[10.2.19.39:6380]: [empty] -> wait +4ms
  ioredis:cluster status: connecting -> connect +1ms
  ioredis:redis write command[0] -> cluster(info) +1ms
  ioredis:cluster status: connect -> ready +1ms

I've checked the code of ioredis and indeed, it's using cluster slots command to get IP address to connect to:

  redis.cluster('slots', utils.timeout(function (err, result) {
    if (err) {
      redis.disconnect();
      return callback(err);
    }
    var nodes = [];

    for (var i = 0; i < result.length; ++i) {
      var items = result[i];
      var slotRangeStart = items[0];
      var slotRangeEnd = items[1];

      var keys = [];
      for (var j = 2; j < items.length; j++) {
        items[j] = { host: items[j][0]/* || redis.options.host*/, port: items[j][1] };
        items[j].readOnly = j !== 2;
        nodes.push(items[j]);
        keys.push(items[j].host + ':' + items[j].port);
      }

      for (var slot = slotRangeStart; slot <= slotRangeEnd; slot++) {
        _this.slots[slot] = keys;
      }
    }

    _this.connectionPool.reset(nodes);
    callback();
  }, 1000));

I think this is inherently wrong as IP addresses used for inter-node communication do not necessarily coincide with IP addresses used by clients. For example if a client for whatever reason is forced to use public IP address then the only way to make it work is to specify public IP for inter-node communication:

redis-cli -p 6380 cluster meet 52.17.215.146 6379

[luin]

You should bind the redis to the correct interface instead of 0.0.0.0. 10.2.1.1 for example.

[bsnote]

This just forces to use internal IP and doesn't allow to connect to Redis using localhost or public IP address. I think it's better to still bind to 0.0.0.0 but use cluster meet 10.2.x.x 6379 command, which makes Redis save 10.2.x.x in nodes.conf. BTW in order to use single-node cluster, the node should cluster meet itself using internal IP address. For example redis-cli -p 6379 cluster meet 10.2.19.39 6379

In both cases it's not possible to connect to Redis using public IP address or use localhost address for inter-node communication. It's not a big deal as it's probably not a good idea to expose Redis to outside world anyway. It's just that the overall concept of using the same IP address for inter-node communication and user client communication looks not perfect. Perhaps it's more of a general problem in Redis.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed after 7 days if no further activity occurs, but feel free to re-open a closed issue if needed.