Skip to content

Commit 1a49301

Browse files
lyu571lyu571
authored
Merge pull request #1593 from xawei/feat/tke-gpu-new-args
feat: support setting tke gpu args
2 parents d33edb5 + da674f7 commit 1a49301

9 files changed

+521
-78
lines changed

‎.changelog/1593.txt‎

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,9 @@
1+
```release-note:enhancement
2+
resource/tencentcloud_kubernetes_scale_worker: support setting tke gpu args
3+
```
4+
```release-note:enhancement
5+
resource/tencentcloud_kubernetes_cluster_attachment: support setting tke gpu args
6+
```
7+
```release-note:enhancement
8+
resource/tencentcloud_kubernetes_node_pool: support setting tke gpu args
9+
```

‎tencentcloud/resource_tc_kubernetes_cluster_attachment.go‎

Lines changed: 227 additions & 77 deletions
Original file line numberDiff line numberDiff line change
@@ -5,95 +5,95 @@ Example Usage
55
66
```hcl
77
8-
variable "availability_zone" {
9-
default = "ap-guangzhou-3"
10-
}
8+
variable "availability_zone" {
9+
default = "ap-guangzhou-3"
10+
}
1111
12-
variable "cluster_cidr" {
13-
default = "172.16.0.0/16"
14-
}
12+
variable "cluster_cidr" {
13+
default = "172.16.0.0/16"
14+
}
1515
16-
variable "default_instance_type" {
17-
default = "S1.SMALL1"
18-
}
16+
variable "default_instance_type" {
17+
default = "S1.SMALL1"
18+
}
1919
20-
data "tencentcloud_images" "default" {
21-
image_type = ["PUBLIC_IMAGE"]
22-
os_name = "centos"
23-
}
20+
data "tencentcloud_images" "default" {
21+
image_type = ["PUBLIC_IMAGE"]
22+
os_name = "centos"
23+
}
2424
25+
data "tencentcloud_vpc_subnets" "vpc" {
26+
is_default = true
27+
availability_zone = var.availability_zone
28+
}
2529
26-
data "tencentcloud_vpc_subnets" "vpc" {
27-
is_default = true
28-
availability_zone = var.availability_zone
29-
}
30+
data "tencentcloud_instance_types" "default" {
31+
filter {
32+
name = "instance-family"
33+
values = ["SA2"]
34+
}
3035
31-
data "tencentcloud_instance_types" "default" {
32-
filter {
33-
name = "instance-family"
34-
values = ["SA2"]
35-
}
36+
cpu_core_count = 8
37+
memory_size = 16
38+
}
3639
37-
cpu_core_count = 8
38-
memory_size = 16
39-
}
40+
resource "tencentcloud_instance" "foo" {
41+
instance_name = "tf-auto-test-1-1"
42+
availability_zone = var.availability_zone
43+
image_id = data.tencentcloud_images.default.images.0.image_id
44+
instance_type = var.default_instance_type
45+
system_disk_type = "CLOUD_PREMIUM"
46+
system_disk_size = 50
47+
}
4048
41-
resource "tencentcloud_instance" "foo" {
42-
instance_name = "tf-auto-test-1-1"
43-
availability_zone = var.availability_zone
44-
image_id = data.tencentcloud_images.default.images.0.image_id
45-
instance_type = var.default_instance_type
46-
system_disk_type = "CLOUD_PREMIUM"
47-
system_disk_size = 50
48-
}
49+
resource "tencentcloud_kubernetes_cluster" "managed_cluster" {
50+
vpc_id = data.tencentcloud_vpc_subnets.vpc.instance_list.0.vpc_id
51+
cluster_cidr = "10.1.0.0/16"
52+
cluster_max_pod_num = 32
53+
cluster_name = "keep"
54+
cluster_desc = "test cluster desc"
55+
cluster_max_service_num = 32
4956
50-
resource "tencentcloud_kubernetes_cluster" "managed_cluster" {
51-
vpc_id = data.tencentcloud_vpc_subnets.vpc.instance_list.0.vpc_id
52-
cluster_cidr = "10.1.0.0/16"
53-
cluster_max_pod_num = 32
54-
cluster_name = "keep"
55-
cluster_desc = "test cluster desc"
56-
cluster_max_service_num = 32
57-
58-
worker_config {
59-
count = 1
60-
availability_zone = var.availability_zone
61-
instance_type = var.default_instance_type
62-
system_disk_type = "CLOUD_SSD"
63-
system_disk_size = 60
64-
internet_charge_type = "TRAFFIC_POSTPAID_BY_HOUR"
65-
internet_max_bandwidth_out = 100
66-
public_ip_assigned = true
67-
subnet_id = data.tencentcloud_vpc_subnets.vpc.instance_list.0.subnet_id
68-
69-
data_disk {
70-
disk_type = "CLOUD_PREMIUM"
71-
disk_size = 50
72-
}
73-
74-
enhanced_security_service = false
75-
enhanced_monitor_service = false
76-
user_data = "dGVzdA=="
77-
password = "ZZXXccvv1212"
78-
}
79-
80-
cluster_deploy_type = "MANAGED_CLUSTER"
81-
}
57+
worker_config {
58+
count = 1
59+
availability_zone = var.availability_zone
60+
instance_type = var.default_instance_type
61+
system_disk_type = "CLOUD_SSD"
62+
system_disk_size = 60
63+
internet_charge_type = "TRAFFIC_POSTPAID_BY_HOUR"
64+
internet_max_bandwidth_out = 100
65+
public_ip_assigned = true
66+
subnet_id = data.tencentcloud_vpc_subnets.vpc.instance_list.0.subnet_id
8267
83-
resource "tencentcloud_kubernetes_cluster_attachment" "test_attach" {
84-
cluster_id = tencentcloud_kubernetes_cluster.managed_cluster.id
85-
instance_id = tencentcloud_instance.foo.id
86-
password = "Lo4wbdit"
68+
data_disk {
69+
disk_type = "CLOUD_PREMIUM"
70+
disk_size = 50
71+
}
8772
88-
labels = {
89-
"test1" = "test1",
90-
"test2" = "test2",
91-
}
73+
enhanced_security_service = false
74+
enhanced_monitor_service = false
75+
user_data = "dGVzdA=="
76+
password = "ZZXXccvv1212"
77+
}
78+
79+
cluster_deploy_type = "MANAGED_CLUSTER"
80+
}
81+
82+
resource "tencentcloud_kubernetes_cluster_attachment" "test_attach" {
83+
cluster_id = tencentcloud_kubernetes_cluster.managed_cluster.id
84+
instance_id = tencentcloud_instance.foo.id
85+
password = "Lo4wbdit"
86+
87+
labels = {
88+
"test1" = "test1",
89+
"test2" = "test2",
90+
}
91+
92+
worker_config_overrides {
93+
desired_pod_num = 8
94+
}
95+
}
9296
93-
worker_config_overrides {
94-
desired_pod_num = 8
95-
}
96-
}
9797
```
9898
*/
9999
package tencentcloud
@@ -112,6 +112,101 @@ import (
112112
"github.com/tencentcloudstack/terraform-provider-tencentcloud/tencentcloud/ratelimit"
113113
)
114114

115+
func TKEGpuArgsSetting() map[string]*schema.Schema {
116+
return map[string]*schema.Schema{
117+
"mig_enable": {
118+
Type: schema.TypeBool,
119+
Optional: true,
120+
Default: false,
121+
Description: "Whether to enable MIG.",
122+
},
123+
"driver": {
124+
Type: schema.TypeMap,
125+
Optional: true,
126+
ValidateFunc: validateTkeGpuDriverVersion,
127+
Elem: &schema.Resource{
128+
Schema: map[string]*schema.Schema{
129+
"version": {
130+
Type: schema.TypeString,
131+
Required: true,
132+
Description: "Version of GPU driver or CUDA.",
133+
},
134+
"name": {
135+
Type: schema.TypeString,
136+
Required: true,
137+
Description: "Name of GPU driver or CUDA.",
138+
},
139+
},
140+
},
141+
Description: "GPU driver version.",
142+
},
143+
"cuda": {
144+
Type: schema.TypeMap,
145+
Optional: true,
146+
ValidateFunc: validateTkeGpuDriverVersion,
147+
Elem: &schema.Resource{
148+
Schema: map[string]*schema.Schema{
149+
"version": {
150+
Type: schema.TypeString,
151+
Required: true,
152+
Description: "Version of GPU driver or CUDA.",
153+
},
154+
"name": {
155+
Type: schema.TypeString,
156+
Required: true,
157+
Description: "Name of GPU driver or CUDA.",
158+
},
159+
},
160+
},
161+
Description: "CUDA version.",
162+
},
163+
"cudnn": {
164+
Type: schema.TypeMap,
165+
Optional: true,
166+
ValidateFunc: validateTkeGpuDriverVersion,
167+
Elem: &schema.Resource{
168+
Schema: map[string]*schema.Schema{
169+
"version": {
170+
Type: schema.TypeString,
171+
Required: true,
172+
Description: "cuDNN version.",
173+
},
174+
"name": {
175+
Type: schema.TypeString,
176+
Required: true,
177+
Description: "cuDNN name.",
178+
},
179+
"doc_name": {
180+
Type: schema.TypeString,
181+
Optional: true,
182+
Description: "Doc name of cuDNN.",
183+
},
184+
"dev_name": {
185+
Type: schema.TypeString,
186+
Optional: true,
187+
Description: "Dev name of cuDNN.",
188+
},
189+
},
190+
},
191+
Description: "cuDNN version.",
192+
},
193+
"custom_driver": {
194+
Type: schema.TypeMap,
195+
Optional: true,
196+
Elem: &schema.Resource{
197+
Schema: map[string]*schema.Schema{
198+
"address": {
199+
Type: schema.TypeString,
200+
Optional: true,
201+
Description: "URL of custom GPU driver address.",
202+
},
203+
},
204+
},
205+
Description: "Custom GPU driver.",
206+
},
207+
}
208+
}
209+
115210
func TkeInstanceAdvancedSetting() map[string]*schema.Schema {
116211
return map[string]*schema.Schema{
117212
"mount_target": {
@@ -206,6 +301,16 @@ func TkeInstanceAdvancedSetting() map[string]*schema.Schema {
206301
Optional: true,
207302
Description: "Indicate to set desired pod number in node. valid when the cluster is podCIDR.",
208303
},
304+
"gpu_args": {
305+
Type: schema.TypeList,
306+
Optional: true,
307+
ForceNew: true,
308+
MaxItems: 1,
309+
Elem: &schema.Resource{
310+
Schema: TKEGpuArgsSetting(),
311+
},
312+
Description: "GPU driver parameters.",
313+
},
209314
}
210315
}
211316

@@ -363,6 +468,51 @@ func tkeGetInstanceAdvancedPara(dMap map[string]interface{}, meta interface{}) (
363468
setting.ExtraArgs = &clusterExtraArgs
364469
}
365470

471+
// get gpu_args
472+
if v, ok := dMap["gpu_args"]; ok && len(v.([]interface{})) > 0 {
473+
gpuArgs := v.([]interface{})[0].(map[string]interface{})
474+
475+
var (
476+
migEnable = gpuArgs["mig_enable"].(bool)
477+
driver = gpuArgs["driver"].(map[string]interface{})
478+
cuda = gpuArgs["cuda"].(map[string]interface{})
479+
cudnn = gpuArgs["cudnn"].(map[string]interface{})
480+
customDriver = gpuArgs["custom_driver"].(map[string]interface{})
481+
)
482+
tkeGpuArgs := tke.GPUArgs{}
483+
tkeGpuArgs.MIGEnable = &migEnable
484+
if len(driver) > 0 {
485+
tkeGpuArgs.Driver = &tke.DriverVersion{
486+
Version: helper.String(driver["version"].(string)),
487+
Name: helper.String(driver["name"].(string)),
488+
}
489+
}
490+
if len(cuda) > 0 {
491+
tkeGpuArgs.CUDA = &tke.DriverVersion{
492+
Version: helper.String(cuda["version"].(string)),
493+
Name: helper.String(cuda["name"].(string)),
494+
}
495+
}
496+
if len(cudnn) > 0 {
497+
tkeGpuArgs.CUDNN = &tke.CUDNN{
498+
Version: helper.String(cudnn["version"].(string)),
499+
Name: helper.String(cudnn["name"].(string)),
500+
}
501+
if cudnn["doc_name"] != nil {
502+
tkeGpuArgs.CUDNN.DocName = helper.String(cudnn["doc_name"].(string))
503+
}
504+
if cudnn["dev_name"] != nil {
505+
tkeGpuArgs.CUDNN.DevName = helper.String(cudnn["dev_name"].(string))
506+
}
507+
}
508+
if len(customDriver) > 0 {
509+
tkeGpuArgs.CustomDriver = &tke.CustomDriver{
510+
Address: helper.String(customDriver["address"].(string)),
511+
}
512+
}
513+
setting.GPUArgs = &tkeGpuArgs
514+
}
515+
366516
return setting
367517
}
368518
func resourceTencentCloudTkeClusterAttachmentRead(d *schema.ResourceData, meta interface{}) error {

0 commit comments

Comments
 (0)