Now session handler all time tries to connect and get all nodes from first node that defined in session.save_path

This is also necessary to make PHPRedis available during a network partition where the first node in the list is still available to the webserver, but does not have a connection to the majority of the cluster. This will take down the whole web application even though the cluster is still running.

Hey,

Thanks so much for the pull request, and sorry for the late response. I have pulled your code down and run it but I ran into a couple of issues.

shuffle operates on the array itself (a.k.a. takes a reference), which means that the way the code works now, it will modify the userland seed array in place (and the user will get it back with the reordered version).

More importantly, when I run it through the unit tests, it actually causes corruption if you call RedisCluster::__construct multiple times (again, this is to do with the weird reference thing in PHP).

The solution is to make a copy of the array, shuffle the copy, and then use the shuffled copy for construction. Something like this:

/* Helper method to shuffle a retrieved array */
int cluster_shuffle_array(INTERNAL_FUNCTION_PARAMETERS, zval **array) {
    zval *func, *copy, **args[1], *retval = NULL;

    /* Init function name */
    MAKE_STD_ZVAL(func);
    ZVAL_STRINGL(func, "shuffle", sizeof("shuffle")-1, 0);

    /* shuffle operates on the provided value, so copy the array */
    MAKE_STD_ZVAL(copy);
    *copy = **array;
    zval_copy_ctor(copy);
    args[0] = ©

    /* Call shuffle on our copy */
    if (call_user_function_ex(CG(function_table), NULL, func, &retval, 1, args, 0, NULL
                              TSRMLS_CC) == FAILURE)
    {
        efree(func);
        zval_dtor(copy);
        efree(copy);
        return 0;
    }

    /* Update pointer to copied and randomized array */
    *array = copy;

    /* cleanup */
    if (retval) efree(retval);
    efree(func);

    return 1;
}

I pulled the logic into a separate function because it became enough code that it probably shouldn't live in multiple places..

But here is where the Zend API makes it fun for us. You can't use INTERNAL_FUNCTION_PARAMETERS from within a session method.

What do you think about using fyshuffle inside of cluster_init_seeds?

Sorry about the wall of text!

Hi, thx for comment. It's really good point to put it into cluster_init_seeds and just shuffle HashTable there

I will try to do that

Awesome, thanks. There is already an fyshuffle in cluster_library.c

Let me know if you have any questions :)

Created new pull request #777