JedisCluster does not translate IP address of remote host into one it can use remotely

[allanwax]

If a machine is set up with multiple instances using 127.0.0.1 as it's address, then if another machine tries to create a JedisCluster and specifies the first machine address, the lookup of the members of the cluster are incorrectly created. The first machine node is the proper address but the rest of the cluster is 127.0.0.1 which is a local reference on the cluster.

import org.apache.log4j.ConsoleAppender;
import org.apache.log4j.Level;
import org.apache.log4j.Logger;

import redis.clients.jedis.HostAndPort;
import redis.clients.jedis.JedisCluster;

public class JDCPBug {
static String redisHost = null;
static Logger log = null;

static {
Logger rootLogger = Logger.getRootLogger();
if (!rootLogger.getAllAppenders().hasMoreElements()) {
rootLogger.setLevel(Level.INFO);
rootLogger.addAppender(new ConsoleAppender(new org.apache.log4j.EnhancedPatternLayout("%d{yyyy-MM-dd HH:mm:ss.SSS} [%t; %C{1}] %-5p -- %m%n")));
}

log = Logger.getLogger(JDCPBug.class);

try {
    redisHost = InetAddress.getByName("test2").getHostAddress();
}
catch (Exception e) {
    log.fatal("Could not get address for redis host");
    System.exit(1);
}

}

public static void main(String[] args) {
Set jedisClusterNodes = new HashSet();
jedisClusterNodes.add(new HostAndPort(redisHost, 17000));
JedisCluster jc = new JedisCluster(jedisClusterNodes);
log.info("Created jedis cluster on host " + redisHost + ", id=" + System.identityHashCode(jc));
log.info("Cluster nodes: " + jc.getClusterNodes().toString());
}

}

2014-08-15 10:28:36.714 [main; JDCPBug] INFO -- Created jedis cluster on host 10.11.14.23, id=1184623795
2014-08-15 10:28:36.717 [main; JDCPBug] INFO -- Cluster nodes: {10.11.14.23:17000=redis.clients.jedis.JedisPool@1cad8848, 127.0.0.1:17002=redis.clients.jedis.JedisPool@44ce3dad, 127.0.0.1:17001=redis.clients.jedis.JedisPool@40334c25, 127.0.0.1:17000=redis.clients.jedis.JedisPool@67cc3210, 127.0.0.1:17005=redis.clients.jedis.JedisPool@eeaedad, 127.0.0.1:17003=redis.clients.jedis.JedisPool@619a0081, 127.0.0.1:17004=redis.clients.jedis.JedisPool@45974736}

If the cluster is rebuilt to use the host IP instead, the addresses are fine:
2014-08-15 12:43:09.388 [main; JDCPBug] INFO -- Created jedis cluster on host 10.11.14.23, id=887408677
2014-08-15 12:43:09.391 [main; JDCPBug] INFO -- Cluster nodes: {10.11.14.23:17000=redis.clients.jedis.JedisPool@6aa9f1d0, 10.11.14.23:17002=redis.clients.jedis.JedisPool@3b42c565, 10.11.14.23:17001=redis.clients.jedis.JedisPool@1cad8848, 10.11.14.23:17005=redis.clients.jedis.JedisPool@44ce3dad, 10.11.14.23:17003=redis.clients.jedis.JedisPool@40334c25, 10.11.14.23:17004=redis.clients.jedis.JedisPool@67cc3210}

[marcosnils]

@allanwax thanks for reporting the issue. I'll give it a look during the weekend.

[HeartSaVioR]

Duplicated by #633 and #585.
If you feel not comfortable with not translating IP, it's up to Redis Cluster because JedisCluster just build cluster node information by "cluster info" command.
When you think it's up to Jedis, please reopen issue and let me know any reasons.

[HeartSaVioR]

@xetorthio @marcosnils We may have to update FAQ wiki as same non-bug issues are resolved multiple times.

[allanwax]

I don't know how to reopen this issue but here's the reason why it should be.

The problem is for JedisCluster, not Jedis as it's dealing with a many-to-one issue.

With a simple redis connection, I request a connection to a remote server. The remote server could actually be local (127.0.0.1). Done. With JedisCluster I'm making a connection to multiple machines at the same time but it functions on the surface like a single machine. Under the hood it is many connections to potentially many machines. So if the only master node I contact to open the cluster has a public ip address, but the cluster is partially or completely composed of local machines (the test example above), the information I get from the other nodes only makes sense to that machine, but is completely legitimate, IN THAT MACHINE'S CONTEXT. it fails because the assumption is that all the nodes I want to contact are just as 'remote' as the one I started with.

There is a simple change to the cluster that will get around this. Contact the first master and request all the nodes. If any of the nodes are a local address (but on a different port) then replace the local address with the master's non-local address. Problem solved with a few lines of code.

[HeartSaVioR]

@allanwax I've reopen the issue. But, I'm confused due to antirez/redis#1848 so I'm waiting for reply.
Actually I don't like "CONTEXT" because we're creating "client library" that can be compatible to Redis 2.x and 3.x (maybe?).
We can get higher compatibility with avoid decision and just use Redis' response.
When Redis can give enough information (and I think it should) we don't need to decide whether using "CONTEXT" or not.

[HeartSaVioR]

Fixed by #668
Sorry @allanwax, you're right. Unless Redis doesn't decide its IP, we should decide based on context.

[allanwax]

(y)