/
resource_pytorch_job.go
194 lines (161 loc) · 5.51 KB
/
resource_pytorch_job.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
package kubeflowtraining
import (
"fmt"
"log"
"time"
"github.com/hashicorp/terraform-plugin-sdk/helper/resource"
"github.com/hashicorp/terraform-plugin-sdk/helper/schema"
kubeflowv1 "github.com/kubeflow/training-operator/pkg/apis/kubeflow.org/v1"
"github.com/rh01/terraform-provider-kubeflow-training/kubeflowtraining/client"
"github.com/rh01/terraform-provider-kubeflow-training/kubeflowtraining/schema/pytorch_job"
"github.com/rh01/terraform-provider-kubeflow-training/kubeflowtraining/utils"
"github.com/rh01/terraform-provider-kubeflow-training/kubeflowtraining/utils/patch"
"k8s.io/apimachinery/pkg/api/errors"
)
func resourceKubeFlowPyTorchJob() *schema.Resource {
return &schema.Resource{
Create: resourceKubeFlowPyTorchJobCreate,
Read: resourceKubeFlowPyTorchJobRead,
Update: resourceKubeFlowPyTorchJobUpdate,
Delete: resourceKubeFlowPyTorchJobDelete,
Exists: resourceKubeFlowPyTorchJobExists,
Importer: &schema.ResourceImporter{
State: schema.ImportStatePassthrough,
},
Timeouts: &schema.ResourceTimeout{
Create: schema.DefaultTimeout(40 * time.Minute),
Delete: schema.DefaultTimeout(5 * time.Minute),
},
Schema: pytorch_job.PyTorchJobFields(),
}
}
func resourceKubeFlowPyTorchJobCreate(resourceData *schema.ResourceData, meta interface{}) error {
cli := (meta).(client.Client)
dv, err := pytorch_job.FromResourceData(resourceData)
if err != nil {
return err
}
log.Printf("[INFO] Creating new data volume: %#v", dv)
if err := cli.CreatePyTorchJob(dv); err != nil {
return err
}
log.Printf("[INFO] Submitted new data volume: %#v", dv)
if err := pytorch_job.ToResourceData(*dv, resourceData); err != nil {
return err
}
resourceData.SetId(utils.BuildId(dv.ObjectMeta))
// Wait for data volume instance's status phase to be succeeded:
name := dv.ObjectMeta.Name
namespace := dv.ObjectMeta.Namespace
stateConf := &resource.StateChangeConf{
Pending: []string{"Creating"},
Target: []string{"Succeeded"},
Timeout: resourceData.Timeout(schema.TimeoutCreate),
Refresh: func() (interface{}, string, error) {
var err error
dv, err = cli.GetPyTorchJob(namespace, name)
if err != nil {
if errors.IsNotFound(err) {
log.Printf("[DEBUG] data volume %s is not created yet", name)
return dv, "Creating", nil
}
return dv, "", err
}
// switch dv.Status.Phase {
// case cdiv1.Succeeded:
// return dv, "Succeeded", nil
// case cdiv1.Failed:
// return dv, "", fmt.Errorf("data volume failed to be created, finished with phase=\"failed\"")
// }
log.Printf("[DEBUG] data volume %s is being created", name)
return dv, "Creating", nil
},
}
if _, err := stateConf.WaitForState(); err != nil {
return fmt.Errorf("%s", err)
}
return pytorch_job.ToResourceData(*dv, resourceData)
}
func resourceKubeFlowPyTorchJobRead(resourceData *schema.ResourceData, meta interface{}) error {
cli := (meta).(client.Client)
namespace, name, err := utils.IdParts(resourceData.Id())
if err != nil {
return err
}
log.Printf("[INFO] Reading data volume %s", name)
dv, err := cli.GetPyTorchJob(namespace, name)
if err != nil {
log.Printf("[DEBUG] Received error: %#v", err)
return err
}
log.Printf("[INFO] Received data volume: %#v", dv)
return pytorch_job.ToResourceData(*dv, resourceData)
}
func resourceKubeFlowPyTorchJobUpdate(resourceData *schema.ResourceData, meta interface{}) error {
cli := (meta).(client.Client)
namespace, name, err := utils.IdParts(resourceData.Id())
if err != nil {
return err
}
ops := pytorch_job.AppendPatchOps("", "", resourceData, []patch.PatchOperation{})
data, err := ops.MarshalJSON()
if err != nil {
return fmt.Errorf("Failed to marshal update operations: %s", err)
}
log.Printf("[INFO] Updating data volume: %s", ops)
out := &kubeflowv1.PyTorchJob{}
if err := cli.UpdatePyTorchJob(namespace, name, out, data); err != nil {
return err
}
log.Printf("[INFO] Submitted updated data volume: %#v", out)
return resourceKubeFlowPyTorchJobRead(resourceData, meta)
}
func resourceKubeFlowPyTorchJobDelete(resourceData *schema.ResourceData, meta interface{}) error {
cli := (meta).(client.Client)
namespace, name, err := utils.IdParts(resourceData.Id())
if err != nil {
return err
}
log.Printf("[INFO] Deleting data volume: %#v", name)
if err := cli.DeletePyTorchJob(namespace, name); err != nil {
return err
}
// Wait for data volume instance to be removed:
stateConf := &resource.StateChangeConf{
Pending: []string{"Deleting"},
Timeout: resourceData.Timeout(schema.TimeoutDelete),
Refresh: func() (interface{}, string, error) {
dv, err := cli.GetPyTorchJob(namespace, name)
if err != nil {
if errors.IsNotFound(err) {
return nil, "", nil
}
return dv, "", err
}
log.Printf("[DEBUG] data volume %s is being deleted", dv.GetName())
return dv, "Deleting", nil
},
}
if _, err := stateConf.WaitForState(); err != nil {
return fmt.Errorf("%s", err)
}
log.Printf("[INFO] data volume %s deleted", name)
resourceData.SetId("")
return nil
}
func resourceKubeFlowPyTorchJobExists(resourceData *schema.ResourceData, meta interface{}) (bool, error) {
cli := (meta).(client.Client)
namespace, name, err := utils.IdParts(resourceData.Id())
if err != nil {
return false, err
}
log.Printf("[INFO] Checking data volume %s", name)
if _, err := cli.GetPyTorchJob(namespace, name); err != nil {
if statusErr, ok := err.(*errors.StatusError); ok && statusErr.ErrStatus.Code == 404 {
return false, nil
}
log.Printf("[DEBUG] Received error: %#v", err)
return true, err
}
return true, nil
}