Check Failing() before serving random node

[AnatolyRugalev]

We encounter errors during Redis failover with RouteRandomly set to true.

After some debugging we found that Failing is never checked when serving random node for the slot.

This should be pretty easy to reproduce:

1. Start cluster using this docker-compose service:

  redis-cluster:
    image: grokzen/redis-cluster:6.0.9
    container_name: redis-cluster.go-redis-test
    environment:
      - "IP=0.0.0.0"
    ports:
      - 7000-7005:7000-7005

2. Connect to cluster, perform read/write operations
3. docker-compose exec redis-cluster supervisorctl stop redis-2

[vmihailenco]

Nice catch - thanks 👍

Does it make sense to try random node here? It will result in MOVED error and will re-try the command again on the same failing nodes.

I think we should:

• remove c.nodes.Random() and just use first/random failing node as we did previously
• add an optimization for len(nodes) == 1 since rand.Perm is somewhat expensive

if len(nodes) == 1 {
    return nodes[0]
}

[vmihailenco]

Slightly better:

if node := nodes[idx]; !node.Failing() {
    return node, nil
}

[monkey92t]

Thanks!

[monkey92t]

There are still some other issues that need to be resolved, we will release a new version later.