terraform/builtin/providers/aws/resource_aws_spot_instance_...

320 lines
9.2 KiB
Go
Raw Normal View History

provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
package aws
import (
"fmt"
"log"
"time"
"github.com/aws/aws-sdk-go/aws"
"github.com/aws/aws-sdk-go/aws/awserr"
"github.com/aws/aws-sdk-go/service/ec2"
"github.com/hashicorp/terraform/helper/resource"
"github.com/hashicorp/terraform/helper/schema"
)
func resourceAwsSpotInstanceRequest() *schema.Resource {
return &schema.Resource{
Create: resourceAwsSpotInstanceRequestCreate,
Read: resourceAwsSpotInstanceRequestRead,
Delete: resourceAwsSpotInstanceRequestDelete,
Update: resourceAwsSpotInstanceRequestUpdate,
Schema: func() map[string]*schema.Schema {
// The Spot Instance Request Schema is based on the AWS Instance schema.
s := resourceAwsInstance().Schema
// Everything on a spot instance is ForceNew except tags
for k, v := range s {
if k == "tags" {
continue
}
v.ForceNew = true
}
s["spot_price"] = &schema.Schema{
Type: schema.TypeString,
Required: true,
ForceNew: true,
}
2015-09-23 20:30:44 +02:00
s["spot_type"] = &schema.Schema{
Type: schema.TypeString,
Optional: true,
Default: "persistent",
}
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
s["wait_for_fulfillment"] = &schema.Schema{
Type: schema.TypeBool,
Optional: true,
Default: false,
}
s["spot_bid_status"] = &schema.Schema{
Type: schema.TypeString,
Computed: true,
}
s["spot_request_state"] = &schema.Schema{
Type: schema.TypeString,
Computed: true,
}
s["spot_instance_id"] = &schema.Schema{
Type: schema.TypeString,
Computed: true,
}
s["block_duration_minutes"] = &schema.Schema{
Type: schema.TypeInt,
Optional: true,
ForceNew: true,
}
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
return s
}(),
}
}
func resourceAwsSpotInstanceRequestCreate(d *schema.ResourceData, meta interface{}) error {
conn := meta.(*AWSClient).ec2conn
instanceOpts, err := buildAwsInstanceOpts(d, meta)
if err != nil {
return err
}
spotOpts := &ec2.RequestSpotInstancesInput{
SpotPrice: aws.String(d.Get("spot_price").(string)),
2015-09-23 20:30:44 +02:00
Type: aws.String(d.Get("spot_type").(string)),
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
// Though the AWS API supports creating spot instance requests for multiple
// instances, for TF purposes we fix this to one instance per request.
// Users can get equivalent behavior out of TF's "count" meta-parameter.
InstanceCount: aws.Int64(1),
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
LaunchSpecification: &ec2.RequestSpotLaunchSpecification{
BlockDeviceMappings: instanceOpts.BlockDeviceMappings,
EbsOptimized: instanceOpts.EBSOptimized,
Monitoring: instanceOpts.Monitoring,
IamInstanceProfile: instanceOpts.IAMInstanceProfile,
ImageId: instanceOpts.ImageID,
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
InstanceType: instanceOpts.InstanceType,
KeyName: instanceOpts.KeyName,
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
Placement: instanceOpts.SpotPlacement,
SecurityGroupIds: instanceOpts.SecurityGroupIDs,
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
SecurityGroups: instanceOpts.SecurityGroups,
SubnetId: instanceOpts.SubnetID,
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
UserData: instanceOpts.UserData64,
},
}
if v, ok := d.GetOk("block_duration_minutes"); ok {
spotOpts.BlockDurationMinutes = aws.Int64(int64(v.(int)))
}
// If the instance is configured with a Network Interface (a subnet, has
// public IP, etc), then the instanceOpts.SecurityGroupIds and SubnetId will
// be nil
if len(instanceOpts.NetworkInterfaces) > 0 {
spotOpts.LaunchSpecification.SecurityGroupIds = instanceOpts.NetworkInterfaces[0].Groups
spotOpts.LaunchSpecification.SubnetId = instanceOpts.NetworkInterfaces[0].SubnetId
}
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
// Make the spot instance request
log.Printf("[DEBUG] Requesting spot bid opts: %s", spotOpts)
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
resp, err := conn.RequestSpotInstances(spotOpts)
if err != nil {
return fmt.Errorf("Error requesting spot instances: %s", err)
}
if len(resp.SpotInstanceRequests) != 1 {
return fmt.Errorf(
"Expected response with length 1, got: %s", resp)
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
}
sir := *resp.SpotInstanceRequests[0]
d.SetId(*sir.SpotInstanceRequestId)
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
if d.Get("wait_for_fulfillment").(bool) {
spotStateConf := &resource.StateChangeConf{
// http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/spot-bid-status.html
Pending: []string{"start", "pending-evaluation", "pending-fulfillment"},
Target: []string{"fulfilled"},
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
Refresh: SpotInstanceStateRefreshFunc(conn, sir),
Timeout: 10 * time.Minute,
Delay: 10 * time.Second,
MinTimeout: 3 * time.Second,
}
log.Printf("[DEBUG] waiting for spot bid to resolve... this may take several minutes.")
_, err = spotStateConf.WaitForState()
if err != nil {
return fmt.Errorf("Error while waiting for spot request (%s) to resolve: %s", sir, err)
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
}
}
return resourceAwsSpotInstanceRequestUpdate(d, meta)
}
// Update spot state, etc
func resourceAwsSpotInstanceRequestRead(d *schema.ResourceData, meta interface{}) error {
conn := meta.(*AWSClient).ec2conn
req := &ec2.DescribeSpotInstanceRequestsInput{
SpotInstanceRequestIds: []*string{aws.String(d.Id())},
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
}
resp, err := conn.DescribeSpotInstanceRequests(req)
if err != nil {
// If the spot request was not found, return nil so that we can show
// that it is gone.
if ec2err, ok := err.(awserr.Error); ok && ec2err.Code() == "InvalidSpotInstanceRequestID.NotFound" {
d.SetId("")
return nil
}
// Some other error, report it
return err
}
// If nothing was found, then return no state
if len(resp.SpotInstanceRequests) == 0 {
d.SetId("")
return nil
}
request := resp.SpotInstanceRequests[0]
// if the request is cancelled, then it is gone
if *request.State == "cancelled" {
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
d.SetId("")
return nil
}
d.Set("spot_bid_status", *request.Status.Code)
// Instance ID is not set if the request is still pending
if request.InstanceId != nil {
d.Set("spot_instance_id", *request.InstanceId)
2015-11-16 22:16:41 +01:00
// Read the instance data, setting up connection information
if err := readInstance(d, meta); err != nil {
return fmt.Errorf("[ERR] Error reading Spot Instance Data: %s", err)
}
}
d.Set("spot_request_state", request.State)
d.Set("block_duration_minutes", request.BlockDurationMinutes)
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
d.Set("tags", tagsToMap(request.Tags))
2015-11-16 22:16:41 +01:00
return nil
}
func readInstance(d *schema.ResourceData, meta interface{}) error {
conn := meta.(*AWSClient).ec2conn
resp, err := conn.DescribeInstances(&ec2.DescribeInstancesInput{
InstanceIds: []*string{aws.String(d.Get("spot_instance_id").(string))},
})
if err != nil {
// If the instance was not found, return nil so that we can show
// that the instance is gone.
if ec2err, ok := err.(awserr.Error); ok && ec2err.Code() == "InvalidInstanceID.NotFound" {
return fmt.Errorf("no instance found")
}
// Some other error, report it
return err
}
// If nothing was found, then return no state
if len(resp.Reservations) == 0 {
return fmt.Errorf("no instances found")
}
instance := resp.Reservations[0].Instances[0]
// Set these fields for connection information
if instance != nil {
d.Set("public_dns", instance.PublicDnsName)
d.Set("public_ip", instance.PublicIpAddress)
d.Set("private_dns", instance.PrivateDnsName)
d.Set("private_ip", instance.PrivateIpAddress)
2015-11-16 21:51:14 +01:00
// set connection information
if instance.PublicIpAddress != nil {
d.SetConnInfo(map[string]string{
"type": "ssh",
"host": *instance.PublicIpAddress,
})
} else if instance.PrivateIpAddress != nil {
d.SetConnInfo(map[string]string{
"type": "ssh",
"host": *instance.PrivateIpAddress,
})
}
}
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
return nil
}
func resourceAwsSpotInstanceRequestUpdate(d *schema.ResourceData, meta interface{}) error {
conn := meta.(*AWSClient).ec2conn
d.Partial(true)
if err := setTags(conn, d); err != nil {
return err
} else {
d.SetPartial("tags")
}
d.Partial(false)
return resourceAwsSpotInstanceRequestRead(d, meta)
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
}
func resourceAwsSpotInstanceRequestDelete(d *schema.ResourceData, meta interface{}) error {
conn := meta.(*AWSClient).ec2conn
log.Printf("[INFO] Cancelling spot request: %s", d.Id())
_, err := conn.CancelSpotInstanceRequests(&ec2.CancelSpotInstanceRequestsInput{
SpotInstanceRequestIds: []*string{aws.String(d.Id())},
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
})
if err != nil {
return fmt.Errorf("Error cancelling spot request (%s): %s", d.Id(), err)
}
if instanceId := d.Get("spot_instance_id").(string); instanceId != "" {
log.Printf("[INFO] Terminating instance: %s", instanceId)
if err := awsTerminateInstance(conn, instanceId); err != nil {
return fmt.Errorf("Error terminating spot instance: %s", err)
}
}
return nil
}
// SpotInstanceStateRefreshFunc returns a resource.StateRefreshFunc that is used to watch
// an EC2 spot instance request
func SpotInstanceStateRefreshFunc(
conn *ec2.EC2, sir ec2.SpotInstanceRequest) resource.StateRefreshFunc {
return func() (interface{}, string, error) {
resp, err := conn.DescribeSpotInstanceRequests(&ec2.DescribeSpotInstanceRequestsInput{
SpotInstanceRequestIds: []*string{sir.SpotInstanceRequestId},
provider/aws: spot_instance_request This is an iteration on the great work done by @dalehamel in PRs #2095 and #2109. The core team went back and forth on how to best model Spot Instance Requests, requesting and then rejecting a separate-resource implementation in #2109. After more internal discussion, we landed once again on a separate resource to model Spot Instance Requests. Out of respect for @dalehamel's already-significant donated time, with this I'm attempting to pick up the work to take this across the finish line. Important architectural decisions represented here: * Spot Instance Requests are always of type "persistent", to properly match Terraform's declarative model. * The spot_instance_request resource exports several attributes that are expected to be constantly changing as the spot market changes: spot_bid_status, spot_request_state, and instance_id. Creating additional resource dependencies based on these attributes is not recommended, as Terraform diffs will be continually generated to keep up with the live changes. * When a Spot Instance Request is deleted/canceled, an attempt is made to terminate the last-known attached spot instance. Race conditions dictate that this attempt cannot guarantee that the associated spot instance is terminated immediately. Implementation notes: * This version of aws_spot_instance_request borrows a lot of common code from aws_instance. * In order to facilitate borrowing, we introduce `awsInstanceOpts`, an internal representation of instance details that's meant to be shared between resources. The goal here would be to refactor ASG Launch Configurations to use the same struct. * The new aws_spot_instance_request acc. test is passing. * All aws_instance acc. tests remain passing.
2015-06-05 17:12:09 +02:00
})
if err != nil {
if ec2err, ok := err.(awserr.Error); ok && ec2err.Code() == "InvalidSpotInstanceRequestID.NotFound" {
// Set this to nil as if we didn't find anything.
resp = nil
} else {
log.Printf("Error on StateRefresh: %s", err)
return nil, "", err
}
}
if resp == nil || len(resp.SpotInstanceRequests) == 0 {
// Sometimes AWS just has consistency issues and doesn't see
// our request yet. Return an empty state.
return nil, "", nil
}
req := resp.SpotInstanceRequests[0]
return req, *req.Status.Code, nil
}
}