Elasticache Memcached az_mode requires ForceNew

[s-maj]

Fixes: #726

Due to some AWS API quirkiness, it's best to set up memcached like this:

resource "aws_elasticache_cluster" "memcached" {
  cluster_id             = "${format("%.16s-%.3s", lower(var.name), random_id.token.hex)}"
  (...)
  az_mode                = "${var.nodes_number == 1 ? "single-az" : "cross-az"}"
  engine                 = "memcached"
}

otherwise it might be hard to scale down cluster. Or just stick with single-az if needed.

=== RUN   TestAccAWSElasticacheCluster_basic
--- PASS: TestAccAWSElasticacheCluster_basic (499.00s)
=== RUN   TestAccAWSElasticacheCluster_snapshotsWithUpdates
--- PASS: TestAccAWSElasticacheCluster_snapshotsWithUpdates (1059.79s)
=== RUN   TestAccAWSElasticacheCluster_decreasingCacheNodes
--- PASS: TestAccAWSElasticacheCluster_decreasingCacheNodes (1137.61s)
=== RUN   TestAccAWSElasticacheCluster_multiAZInVpc
--- PASS: TestAccAWSElasticacheCluster_multiAZInVpc (676.54s)s
=== RUN   TestAccAWSElasticacheCluster_vpc
--- PASS: TestAccAWSElasticacheCluster_vpc (600.21s)

[radeksimko]

I wish we had ForceNewWhen so we could express this appropriately in the schema, but we don't so relying on the server/API side validation seems like a sensible way to do this. 👍

I can confirm that the API does the validation and returns error when trying to scale down:

* aws_elasticache_cluster.bar: [WARN] Error updating ElastiCache cluster (cluster-example), error: InvalidParameterCombination: Can only specify new availability zones or AZ mode when adding cache nodes.
	status code: 400, request id: 67ecf258-ae83-11e7-b63a-c3f6fd37683f

Thanks for the PR.

[ghost]

I'm going to lock this issue because it has been closed for 30 days ⏳. This helps our maintainers find and focus on the active issues.

If you feel this issue should be reopened, we encourage creating a new issue linking back to this one for added context. Thanks!