Scan policy issue (fail_on_cluster_change) about aerospike-client-c

[cstivers78]

We have found the bug, and will be working on a fix shortly.

The bug is that even though the policy.fail_on_cluster_change sent with the operation is false, it is not being picked up b/c the client configuration defines the global fail_on_cluster_change as true.

We will provide a fix shortly. For the near term, you can define fail_on_cluster_change=false in the configuration used to initialize the client. This will mean if you want fail_on_cluster_change=true for an operation, then it needs to be explicit, as it will not longer be the default.

[maxvgi]

Thanks for the quick reply!
But it seems that even the default policy is not working:

int main() {
    aerospike as;
    as_config cfg;
    as_config_init(&cfg);

    cfg.hosts[0].addr = "127.0.0.1";
    cfg.hosts[0].port = 3000;
    cfg.policies.scan.fail_on_cluster_change=false;
    as_error err;

    if (aerospike_connect(aerospike_init(&as, &cfg), &err) != AEROSPIKE_OK) {
        LOG("aerospike_connect() returned %d - %s", err.code, err.message);
        aerospike_destroy(&as);
        return -1;
    }

    as_scan scan;
    as_scan_init(&scan, "database", "userdata");

    LOG("starting scan all ...");
    if (aerospike_scan_foreach(&as, &err, NULL, &scan, scan_cb, NULL) != AEROSPIKE_OK) {
        LOG("aerospike_scan_foreach() returned %d - %s", err.code, err.message);
        as_scan_destroy(&scan);
        return -2;
    }
    LOG("scan finished");

    return 0;
}

The result is similar to the previous one.

[cstivers78]

Apologies, looking back at the code, I realize that it would not have worked.

https://github.com/aerospike/aerospike-client-c/blob/master/src/main/aerospike/_policy.c#L86

p->fail_on_cluster_change   = as_policy_resolve(fail_on_cluster_change, global->scan, local, true);

We will provide a fix.

You are welcome to short circuit that line of code, and we will have a solution out in the near future.

The change I would make to that line of code for your immediate needs would be to have:

p->fail_on_cluster_change   = as_policy_resolve(fail_on_cluster_change, global->scan, local, false);

The fix we have in mind would be to introduce an "undefined" value for boolean fields, which is what they will initialize to. This will allow the fallback mechanism will function properly, such that it will allow you to define a "false" for the policy value, and ensure it is false.

[cstivers78]

For boolean policy values, we are using -1 to indicate an undefined value. So, the boolean policy values are initialized to -1, so it leaves you free to set a client global value or operation specific value. If -1 is specified for both client and operation, then the fallback will be true.

[cstivers78]

Apologies, I rushed in the change and did not make a crucial change to the code, by changing fail_on_cluster_change to an int or enum type. That change will be coming soon.

[cstivers78]

Just pushed the change. It is now functioning. Please verify it meets your needs.