terraform/builtin/providers/aws/resource_aws_elasticache_re...

440 lines
14 KiB
Go

package aws
import (
"fmt"
"log"
"regexp"
"strings"
"time"
"github.com/aws/aws-sdk-go/aws"
"github.com/aws/aws-sdk-go/aws/awserr"
"github.com/aws/aws-sdk-go/service/elasticache"
"github.com/hashicorp/terraform/helper/resource"
"github.com/hashicorp/terraform/helper/schema"
)
func resourceAwsElasticacheReplicationGroup() *schema.Resource {
resourceSchema := resourceAwsElastiCacheCommonSchema()
resourceSchema["replication_group_id"] = &schema.Schema{
Type: schema.TypeString,
Required: true,
ForceNew: true,
ValidateFunc: validateAwsElastiCacheReplicationGroupId,
}
resourceSchema["automatic_failover_enabled"] = &schema.Schema{
Type: schema.TypeBool,
Optional: true,
Default: false,
}
resourceSchema["replication_group_description"] = &schema.Schema{
Type: schema.TypeString,
Required: true,
}
resourceSchema["number_cache_clusters"] = &schema.Schema{
Type: schema.TypeInt,
Required: true,
ForceNew: true,
}
resourceSchema["primary_endpoint_address"] = &schema.Schema{
Type: schema.TypeString,
Computed: true,
}
resourceSchema["engine"].Required = false
resourceSchema["engine"].Optional = true
resourceSchema["engine"].Default = "redis"
resourceSchema["engine"].ValidateFunc = validateAwsElastiCacheReplicationGroupEngine
return &schema.Resource{
Create: resourceAwsElasticacheReplicationGroupCreate,
Read: resourceAwsElasticacheReplicationGroupRead,
Update: resourceAwsElasticacheReplicationGroupUpdate,
Delete: resourceAwsElasticacheReplicationGroupDelete,
Schema: resourceSchema,
}
}
func resourceAwsElasticacheReplicationGroupCreate(d *schema.ResourceData, meta interface{}) error {
conn := meta.(*AWSClient).elasticacheconn
tags := tagsFromMapEC(d.Get("tags").(map[string]interface{}))
params := &elasticache.CreateReplicationGroupInput{
ReplicationGroupId: aws.String(d.Get("replication_group_id").(string)),
ReplicationGroupDescription: aws.String(d.Get("replication_group_description").(string)),
AutomaticFailoverEnabled: aws.Bool(d.Get("automatic_failover_enabled").(bool)),
CacheNodeType: aws.String(d.Get("node_type").(string)),
Engine: aws.String(d.Get("engine").(string)),
Port: aws.Int64(int64(d.Get("port").(int))),
NumCacheClusters: aws.Int64(int64(d.Get("number_cache_clusters").(int))),
Tags: tags,
}
if v, ok := d.GetOk("engine_version"); ok {
params.EngineVersion = aws.String(v.(string))
}
preferred_azs := d.Get("availability_zones").(*schema.Set).List()
if len(preferred_azs) > 0 {
azs := expandStringList(preferred_azs)
params.PreferredCacheClusterAZs = azs
}
if v, ok := d.GetOk("parameter_group_name"); ok {
params.CacheParameterGroupName = aws.String(v.(string))
}
if v, ok := d.GetOk("subnet_group_name"); ok {
params.CacheSubnetGroupName = aws.String(v.(string))
}
security_group_names := d.Get("security_group_names").(*schema.Set).List()
if len(security_group_names) > 0 {
params.CacheSecurityGroupNames = expandStringList(security_group_names)
}
security_group_ids := d.Get("security_group_ids").(*schema.Set).List()
if len(security_group_ids) > 0 {
params.SecurityGroupIds = expandStringList(security_group_ids)
}
snaps := d.Get("snapshot_arns").(*schema.Set).List()
if len(snaps) > 0 {
params.SnapshotArns = expandStringList(snaps)
}
if v, ok := d.GetOk("maintenance_window"); ok {
params.PreferredMaintenanceWindow = aws.String(v.(string))
}
if v, ok := d.GetOk("notification_topic_arn"); ok {
params.NotificationTopicArn = aws.String(v.(string))
}
if v, ok := d.GetOk("snapshot_retention_limit"); ok {
params.SnapshotRetentionLimit = aws.Int64(int64(v.(int)))
}
if v, ok := d.GetOk("snapshot_window"); ok {
params.SnapshotWindow = aws.String(v.(string))
}
if v, ok := d.GetOk("snapshot_name"); ok {
params.SnapshotName = aws.String(v.(string))
}
resp, err := conn.CreateReplicationGroup(params)
if err != nil {
return fmt.Errorf("Error creating Elasticache Replication Group: %s", err)
}
d.SetId(*resp.ReplicationGroup.ReplicationGroupId)
pending := []string{"creating", "modifying", "restoring"}
stateConf := &resource.StateChangeConf{
Pending: pending,
Target: []string{"available"},
Refresh: cacheReplicationGroupStateRefreshFunc(conn, d.Id(), "available", pending),
Timeout: 40 * time.Minute,
MinTimeout: 10 * time.Second,
Delay: 30 * time.Second,
}
log.Printf("[DEBUG] Waiting for state to become available: %v", d.Id())
_, sterr := stateConf.WaitForState()
if sterr != nil {
return fmt.Errorf("Error waiting for elasticache replication group (%s) to be created: %s", d.Id(), sterr)
}
return resourceAwsElasticacheReplicationGroupRead(d, meta)
}
func resourceAwsElasticacheReplicationGroupRead(d *schema.ResourceData, meta interface{}) error {
conn := meta.(*AWSClient).elasticacheconn
req := &elasticache.DescribeReplicationGroupsInput{
ReplicationGroupId: aws.String(d.Id()),
}
res, err := conn.DescribeReplicationGroups(req)
if err != nil {
if eccErr, ok := err.(awserr.Error); ok && eccErr.Code() == "ReplicationGroupNotFoundFault" {
log.Printf("[WARN] Elasticache Replication Group (%s) not found", d.Id())
d.SetId("")
return nil
}
return err
}
var rgp *elasticache.ReplicationGroup
for _, r := range res.ReplicationGroups {
if *r.ReplicationGroupId == d.Id() {
rgp = r
}
}
if rgp == nil {
log.Printf("[WARN] Replication Group (%s) not found", d.Id())
d.SetId("")
return nil
}
if *rgp.Status == "deleting" {
log.Printf("[WARN] The Replication Group %q is currently in the `deleting` state", d.Id())
d.SetId("")
return nil
}
d.Set("automatic_failover_enabled", rgp.AutomaticFailover)
d.Set("replication_group_description", rgp.Description)
d.Set("number_cache_clusters", len(rgp.MemberClusters))
d.Set("replication_group_id", rgp.ReplicationGroupId)
if rgp.NodeGroups != nil {
cacheCluster := *rgp.NodeGroups[0].NodeGroupMembers[0]
res, err := conn.DescribeCacheClusters(&elasticache.DescribeCacheClustersInput{
CacheClusterId: cacheCluster.CacheClusterId,
ShowCacheNodeInfo: aws.Bool(true),
})
if err != nil {
return err
}
if len(res.CacheClusters) == 0 {
return nil
}
c := res.CacheClusters[0]
d.Set("node_type", c.CacheNodeType)
d.Set("engine", c.Engine)
d.Set("engine_version", c.EngineVersion)
d.Set("subnet_group_name", c.CacheSubnetGroupName)
d.Set("security_group_names", c.CacheSecurityGroups)
d.Set("security_group_ids", c.SecurityGroups)
d.Set("parameter_group_name", c.CacheParameterGroup)
d.Set("maintenance_window", c.PreferredMaintenanceWindow)
d.Set("snapshot_window", c.SnapshotWindow)
d.Set("snapshot_retention_limit", c.SnapshotRetentionLimit)
d.Set("primary_endpoint_address", rgp.NodeGroups[0].PrimaryEndpoint.Address)
}
return nil
}
func resourceAwsElasticacheReplicationGroupUpdate(d *schema.ResourceData, meta interface{}) error {
conn := meta.(*AWSClient).elasticacheconn
requestUpdate := false
params := &elasticache.ModifyReplicationGroupInput{
ApplyImmediately: aws.Bool(d.Get("apply_immediately").(bool)),
ReplicationGroupId: aws.String(d.Id()),
}
if d.HasChange("replication_group_description") {
params.ReplicationGroupDescription = aws.String(d.Get("replication_group_description").(string))
requestUpdate = true
}
if d.HasChange("automatic_failover_enabled") {
params.AutomaticFailoverEnabled = aws.Bool(d.Get("automatic_failover_enabled").(bool))
requestUpdate = true
}
if d.HasChange("security_group_ids") {
if attr := d.Get("security_group_ids").(*schema.Set); attr.Len() > 0 {
params.SecurityGroupIds = expandStringList(attr.List())
requestUpdate = true
}
}
if d.HasChange("security_group_names") {
if attr := d.Get("security_group_names").(*schema.Set); attr.Len() > 0 {
params.CacheSecurityGroupNames = expandStringList(attr.List())
requestUpdate = true
}
}
if d.HasChange("preferred_maintenance_window") {
params.PreferredMaintenanceWindow = aws.String(d.Get("preferred_maintenance_window").(string))
requestUpdate = true
}
if d.HasChange("notification_topic_arn") {
params.NotificationTopicArn = aws.String(d.Get("notification_topic_arn").(string))
requestUpdate = true
}
if d.HasChange("parameter_group_name") {
params.CacheParameterGroupName = aws.String(d.Get("parameter_group_name").(string))
requestUpdate = true
}
if d.HasChange("engine_version") {
params.EngineVersion = aws.String(d.Get("engine_version").(string))
requestUpdate = true
}
if d.HasChange("snapshot_retention_limit") {
params.SnapshotRetentionLimit = aws.Int64(int64(d.Get("snapshot_retention_limit").(int)))
requestUpdate = true
}
if d.HasChange("snapshot_window") {
params.SnapshotWindow = aws.String(d.Get("snapshot_window").(string))
requestUpdate = true
}
if d.HasChange("node_type") {
params.CacheNodeType = aws.String(d.Get("node_type").(string))
requestUpdate = true
}
if requestUpdate {
_, err := conn.ModifyReplicationGroup(params)
if err != nil {
return fmt.Errorf("Error updating Elasticache replication group: %s", err)
}
pending := []string{"creating", "modifying", "snapshotting"}
stateConf := &resource.StateChangeConf{
Pending: pending,
Target: []string{"available"},
Refresh: cacheReplicationGroupStateRefreshFunc(conn, d.Id(), "available", pending),
Timeout: 40 * time.Minute,
MinTimeout: 10 * time.Second,
Delay: 30 * time.Second,
}
log.Printf("[DEBUG] Waiting for state to become available: %v", d.Id())
_, sterr := stateConf.WaitForState()
if sterr != nil {
return fmt.Errorf("Error waiting for elasticache replication group (%s) to be created: %s", d.Id(), sterr)
}
}
return resourceAwsElasticacheReplicationGroupRead(d, meta)
}
func resourceAwsElasticacheReplicationGroupDelete(d *schema.ResourceData, meta interface{}) error {
conn := meta.(*AWSClient).elasticacheconn
req := &elasticache.DeleteReplicationGroupInput{
ReplicationGroupId: aws.String(d.Id()),
}
_, err := conn.DeleteReplicationGroup(req)
if err != nil {
if ec2err, ok := err.(awserr.Error); ok && ec2err.Code() == "ReplicationGroupNotFoundFault" {
d.SetId("")
return nil
}
return fmt.Errorf("Error deleting Elasticache replication group: %s", err)
}
log.Printf("[DEBUG] Waiting for deletion: %v", d.Id())
stateConf := &resource.StateChangeConf{
Pending: []string{"creating", "available", "deleting"},
Target: []string{},
Refresh: cacheReplicationGroupStateRefreshFunc(conn, d.Id(), "", []string{}),
Timeout: 40 * time.Minute,
MinTimeout: 10 * time.Second,
Delay: 30 * time.Second,
}
_, sterr := stateConf.WaitForState()
if sterr != nil {
return fmt.Errorf("Error waiting for replication group (%s) to delete: %s", d.Id(), sterr)
}
return nil
}
func cacheReplicationGroupStateRefreshFunc(conn *elasticache.ElastiCache, replicationGroupId, givenState string, pending []string) resource.StateRefreshFunc {
return func() (interface{}, string, error) {
resp, err := conn.DescribeReplicationGroups(&elasticache.DescribeReplicationGroupsInput{
ReplicationGroupId: aws.String(replicationGroupId),
})
if err != nil {
if eccErr, ok := err.(awserr.Error); ok && eccErr.Code() == "ReplicationGroupNotFoundFault" {
log.Printf("[DEBUG] Replication Group Not Found")
return nil, "", nil
}
log.Printf("[ERROR] cacheClusterReplicationGroupStateRefreshFunc: %s", err)
return nil, "", err
}
if len(resp.ReplicationGroups) == 0 {
return nil, "", fmt.Errorf("[WARN] Error: no Cache Replication Groups found for id (%s)", replicationGroupId)
}
var rg *elasticache.ReplicationGroup
for _, replicationGroup := range resp.ReplicationGroups {
if *replicationGroup.ReplicationGroupId == replicationGroupId {
log.Printf("[DEBUG] Found matching ElastiCache Replication Group: %s", *replicationGroup.ReplicationGroupId)
rg = replicationGroup
}
}
if rg == nil {
return nil, "", fmt.Errorf("[WARN] Error: no matching ElastiCache Replication Group for id (%s)", replicationGroupId)
}
log.Printf("[DEBUG] ElastiCache Replication Group (%s) status: %v", replicationGroupId, *rg.Status)
// return the current state if it's in the pending array
for _, p := range pending {
log.Printf("[DEBUG] ElastiCache: checking pending state (%s) for Replication Group (%s), Replication Group status: %s", pending, replicationGroupId, *rg.Status)
s := *rg.Status
if p == s {
log.Printf("[DEBUG] Return with status: %v", *rg.Status)
return s, p, nil
}
}
return rg, *rg.Status, nil
}
}
func validateAwsElastiCacheReplicationGroupEngine(v interface{}, k string) (ws []string, errors []error) {
if strings.ToLower(v.(string)) != "redis" {
errors = append(errors, fmt.Errorf("The only acceptable Engine type when using Replication Groups is Redis"))
}
return
}
func validateAwsElastiCacheReplicationGroupId(v interface{}, k string) (ws []string, errors []error) {
value := v.(string)
if (len(value) < 1) || (len(value) > 16) {
errors = append(errors, fmt.Errorf(
"%q must contain from 1 to 16 alphanumeric characters or hyphens", k))
}
if !regexp.MustCompile(`^[0-9a-zA-Z-]+$`).MatchString(value) {
errors = append(errors, fmt.Errorf(
"only alphanumeric characters and hyphens allowed in %q", k))
}
if !regexp.MustCompile(`^[a-z]`).MatchString(value) {
errors = append(errors, fmt.Errorf(
"first character of %q must be a letter", k))
}
if regexp.MustCompile(`--`).MatchString(value) {
errors = append(errors, fmt.Errorf(
"%q cannot contain two consecutive hyphens", k))
}
if regexp.MustCompile(`-$`).MatchString(value) {
errors = append(errors, fmt.Errorf(
"%q cannot end with a hyphen", k))
}
return
}