@datarobot/pulumi-datarobot 0.10.43 → 0.10.45
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/package.json +1 -1
- package/types/input.d.ts +73 -19
- package/types/output.d.ts +101 -23
- package/workload.d.ts +7 -11
- package/workload.js +1 -5
- package/workload.js.map +1 -1
package/package.json
CHANGED
package/types/input.d.ts
CHANGED
|
@@ -98,9 +98,13 @@ export interface ArtifactSpecContainerGroupContainer {
|
|
|
98
98
|
*/
|
|
99
99
|
environmentVars?: pulumi.Input<pulumi.Input<inputs.ArtifactSpecContainerGroupContainerEnvironmentVar>[]>;
|
|
100
100
|
/**
|
|
101
|
-
*
|
|
101
|
+
* Configuration for server-side image builds from source code.
|
|
102
102
|
*/
|
|
103
|
-
|
|
103
|
+
imageBuildConfig?: pulumi.Input<inputs.ArtifactSpecContainerGroupContainerImageBuildConfig>;
|
|
104
|
+
/**
|
|
105
|
+
* Docker image URI. Omit when using `imageBuildConfig` on draft artifacts; required when status is `locked` and `imageBuildConfig` is set.
|
|
106
|
+
*/
|
|
107
|
+
imageUri?: pulumi.Input<string>;
|
|
104
108
|
/**
|
|
105
109
|
* Container liveness check configuration.
|
|
106
110
|
*/
|
|
@@ -136,11 +140,11 @@ export interface ArtifactSpecContainerGroupContainerEnvironmentVar {
|
|
|
136
140
|
*/
|
|
137
141
|
key?: pulumi.Input<string>;
|
|
138
142
|
/**
|
|
139
|
-
* Name of the environment variable.
|
|
143
|
+
* Name of the environment variable. Required when source is "string" or "dr-credential". Optional for "api-key": when omitted, the platform injects the token as DATAROBOT*API*TOKEN.
|
|
140
144
|
*/
|
|
141
|
-
name
|
|
145
|
+
name?: pulumi.Input<string>;
|
|
142
146
|
/**
|
|
143
|
-
* Source type: "string" for plain text values, "dr-credential" for DataRobot credentials. Defaults to "string".
|
|
147
|
+
* Source type: "string" for plain text values, "dr-credential" for DataRobot credentials, "api-key" for a platform-managed DataRobot API token. Defaults to "string".
|
|
144
148
|
*/
|
|
145
149
|
source?: pulumi.Input<string>;
|
|
146
150
|
/**
|
|
@@ -148,6 +152,48 @@ export interface ArtifactSpecContainerGroupContainerEnvironmentVar {
|
|
|
148
152
|
*/
|
|
149
153
|
value?: pulumi.Input<string>;
|
|
150
154
|
}
|
|
155
|
+
export interface ArtifactSpecContainerGroupContainerImageBuildConfig {
|
|
156
|
+
/**
|
|
157
|
+
* Reference to source code in the DataRobot catalog. Optional at create; required before image build or lock.
|
|
158
|
+
*/
|
|
159
|
+
codeRef?: pulumi.Input<inputs.ArtifactSpecContainerGroupContainerImageBuildConfigCodeRef>;
|
|
160
|
+
/**
|
|
161
|
+
* How the Dockerfile is obtained for the image build. Defaults to using `./Dockerfile` from the source code.
|
|
162
|
+
*/
|
|
163
|
+
dockerfile?: pulumi.Input<inputs.ArtifactSpecContainerGroupContainerImageBuildConfigDockerfile>;
|
|
164
|
+
}
|
|
165
|
+
export interface ArtifactSpecContainerGroupContainerImageBuildConfigCodeRef {
|
|
166
|
+
/**
|
|
167
|
+
* Files API catalog ID (24-character hex).
|
|
168
|
+
*/
|
|
169
|
+
catalogId: pulumi.Input<string>;
|
|
170
|
+
/**
|
|
171
|
+
* Files API catalog version ID (24-character hex).
|
|
172
|
+
*/
|
|
173
|
+
catalogVersionId: pulumi.Input<string>;
|
|
174
|
+
}
|
|
175
|
+
export interface ArtifactSpecContainerGroupContainerImageBuildConfigDockerfile {
|
|
176
|
+
/**
|
|
177
|
+
* Entrypoint baked into the generated Dockerfile CMD. Required when source is `generated`.
|
|
178
|
+
*/
|
|
179
|
+
entrypoints?: pulumi.Input<pulumi.Input<string>[]>;
|
|
180
|
+
/**
|
|
181
|
+
* Execution environment ID for the base Docker image. Required when source is `generated`.
|
|
182
|
+
*/
|
|
183
|
+
executionEnvironmentId?: pulumi.Input<string>;
|
|
184
|
+
/**
|
|
185
|
+
* Execution environment version ID that pins the base image. Required when source is `generated`.
|
|
186
|
+
*/
|
|
187
|
+
executionEnvironmentVersionId?: pulumi.Input<string>;
|
|
188
|
+
/**
|
|
189
|
+
* Relative path to the Dockerfile in the source code. Used when source is `provided`. Defaults to `./Dockerfile`.
|
|
190
|
+
*/
|
|
191
|
+
path?: pulumi.Input<string>;
|
|
192
|
+
/**
|
|
193
|
+
* How the Dockerfile is obtained: `provided` (from source code) or `generated` (from an execution environment). Defaults to `provided`.
|
|
194
|
+
*/
|
|
195
|
+
source?: pulumi.Input<string>;
|
|
196
|
+
}
|
|
151
197
|
export interface ArtifactSpecContainerGroupContainerLivenessProbe {
|
|
152
198
|
/**
|
|
153
199
|
* Minimum consecutive failures for the probe to be considered failed.
|
|
@@ -177,6 +223,10 @@ export interface ArtifactSpecContainerGroupContainerLivenessProbe {
|
|
|
177
223
|
* Scheme to use for connecting to the host (HTTP or HTTPS).
|
|
178
224
|
*/
|
|
179
225
|
scheme?: pulumi.Input<string>;
|
|
226
|
+
/**
|
|
227
|
+
* Minimum consecutive successes for the probe to be considered successful after having failed.
|
|
228
|
+
*/
|
|
229
|
+
successThreshold?: pulumi.Input<number>;
|
|
180
230
|
/**
|
|
181
231
|
* Number of seconds after which the probe times out.
|
|
182
232
|
*/
|
|
@@ -211,6 +261,10 @@ export interface ArtifactSpecContainerGroupContainerReadinessProbe {
|
|
|
211
261
|
* Scheme to use for connecting to the host (HTTP or HTTPS).
|
|
212
262
|
*/
|
|
213
263
|
scheme?: pulumi.Input<string>;
|
|
264
|
+
/**
|
|
265
|
+
* Minimum consecutive successes for the probe to be considered successful after having failed.
|
|
266
|
+
*/
|
|
267
|
+
successThreshold?: pulumi.Input<number>;
|
|
214
268
|
/**
|
|
215
269
|
* Number of seconds after which the probe times out.
|
|
216
270
|
*/
|
|
@@ -245,6 +299,10 @@ export interface ArtifactSpecContainerGroupContainerStartupProbe {
|
|
|
245
299
|
* Scheme to use for connecting to the host (HTTP or HTTPS).
|
|
246
300
|
*/
|
|
247
301
|
scheme?: pulumi.Input<string>;
|
|
302
|
+
/**
|
|
303
|
+
* Minimum consecutive successes for the probe to be considered successful after having failed.
|
|
304
|
+
*/
|
|
305
|
+
successThreshold?: pulumi.Input<number>;
|
|
248
306
|
/**
|
|
249
307
|
* Number of seconds after which the probe times out.
|
|
250
308
|
*/
|
|
@@ -1484,29 +1542,25 @@ export interface WorkloadRuntimeContainerGroupAutoscaling {
|
|
|
1484
1542
|
*/
|
|
1485
1543
|
enabled?: pulumi.Input<boolean>;
|
|
1486
1544
|
/**
|
|
1487
|
-
*
|
|
1545
|
+
* Maximum number of replicas. Defaults to `1`.
|
|
1488
1546
|
*/
|
|
1489
|
-
|
|
1490
|
-
}
|
|
1491
|
-
export interface WorkloadRuntimeContainerGroupAutoscalingPolicy {
|
|
1547
|
+
maxReplicaCount?: pulumi.Input<number>;
|
|
1492
1548
|
/**
|
|
1493
|
-
*
|
|
1549
|
+
* Minimum number of replicas. Set to `0` to allow scale-to-zero. Defaults to `0`.
|
|
1494
1550
|
*/
|
|
1495
|
-
|
|
1551
|
+
minReplicaCount?: pulumi.Input<number>;
|
|
1496
1552
|
/**
|
|
1497
|
-
*
|
|
1498
|
-
*/
|
|
1499
|
-
minCount: pulumi.Input<number>;
|
|
1500
|
-
/**
|
|
1501
|
-
* Policy priority when multiple policies are defined.
|
|
1553
|
+
* Scaling policies that define when and how to scale.
|
|
1502
1554
|
*/
|
|
1503
|
-
|
|
1555
|
+
policies: pulumi.Input<pulumi.Input<inputs.WorkloadRuntimeContainerGroupAutoscalingPolicy>[]>;
|
|
1556
|
+
}
|
|
1557
|
+
export interface WorkloadRuntimeContainerGroupAutoscalingPolicy {
|
|
1504
1558
|
/**
|
|
1505
|
-
* Metric used for scaling decisions: `cpuAverageUtilization`, `httpRequestsConcurrency`, `gpuCacheUtilization`, or `gpuRequestQueueDepth`.
|
|
1559
|
+
* Metric used for scaling decisions: `cpuAverageUtilization`, `httpRequestsConcurrency`, `gpuCacheUtilization`, or `gpuRequestQueueDepth`. Custom metric names (e.g. `vllm:kv_cache_usage_perc`) are supported for NIM artifacts only.
|
|
1506
1560
|
*/
|
|
1507
1561
|
scalingMetric: pulumi.Input<string>;
|
|
1508
1562
|
/**
|
|
1509
|
-
* Target value for the scaling metric.
|
|
1563
|
+
* Target value for the scaling metric. Must be non-negative.
|
|
1510
1564
|
*/
|
|
1511
1565
|
target: pulumi.Input<number>;
|
|
1512
1566
|
}
|
package/types/output.d.ts
CHANGED
|
@@ -97,9 +97,13 @@ export interface ArtifactSpecContainerGroupContainer {
|
|
|
97
97
|
*/
|
|
98
98
|
environmentVars?: outputs.ArtifactSpecContainerGroupContainerEnvironmentVar[];
|
|
99
99
|
/**
|
|
100
|
-
*
|
|
100
|
+
* Configuration for server-side image builds from source code.
|
|
101
101
|
*/
|
|
102
|
-
|
|
102
|
+
imageBuildConfig?: outputs.ArtifactSpecContainerGroupContainerImageBuildConfig;
|
|
103
|
+
/**
|
|
104
|
+
* Docker image URI. Omit when using `imageBuildConfig` on draft artifacts; required when status is `locked` and `imageBuildConfig` is set.
|
|
105
|
+
*/
|
|
106
|
+
imageUri?: string;
|
|
103
107
|
/**
|
|
104
108
|
* Container liveness check configuration.
|
|
105
109
|
*/
|
|
@@ -135,11 +139,11 @@ export interface ArtifactSpecContainerGroupContainerEnvironmentVar {
|
|
|
135
139
|
*/
|
|
136
140
|
key?: string;
|
|
137
141
|
/**
|
|
138
|
-
* Name of the environment variable.
|
|
142
|
+
* Name of the environment variable. Required when source is "string" or "dr-credential". Optional for "api-key": when omitted, the platform injects the token as DATAROBOT*API*TOKEN.
|
|
139
143
|
*/
|
|
140
|
-
name
|
|
144
|
+
name?: string;
|
|
141
145
|
/**
|
|
142
|
-
* Source type: "string" for plain text values, "dr-credential" for DataRobot credentials. Defaults to "string".
|
|
146
|
+
* Source type: "string" for plain text values, "dr-credential" for DataRobot credentials, "api-key" for a platform-managed DataRobot API token. Defaults to "string".
|
|
143
147
|
*/
|
|
144
148
|
source: string;
|
|
145
149
|
/**
|
|
@@ -147,6 +151,48 @@ export interface ArtifactSpecContainerGroupContainerEnvironmentVar {
|
|
|
147
151
|
*/
|
|
148
152
|
value?: string;
|
|
149
153
|
}
|
|
154
|
+
export interface ArtifactSpecContainerGroupContainerImageBuildConfig {
|
|
155
|
+
/**
|
|
156
|
+
* Reference to source code in the DataRobot catalog. Optional at create; required before image build or lock.
|
|
157
|
+
*/
|
|
158
|
+
codeRef?: outputs.ArtifactSpecContainerGroupContainerImageBuildConfigCodeRef;
|
|
159
|
+
/**
|
|
160
|
+
* How the Dockerfile is obtained for the image build. Defaults to using `./Dockerfile` from the source code.
|
|
161
|
+
*/
|
|
162
|
+
dockerfile?: outputs.ArtifactSpecContainerGroupContainerImageBuildConfigDockerfile;
|
|
163
|
+
}
|
|
164
|
+
export interface ArtifactSpecContainerGroupContainerImageBuildConfigCodeRef {
|
|
165
|
+
/**
|
|
166
|
+
* Files API catalog ID (24-character hex).
|
|
167
|
+
*/
|
|
168
|
+
catalogId: string;
|
|
169
|
+
/**
|
|
170
|
+
* Files API catalog version ID (24-character hex).
|
|
171
|
+
*/
|
|
172
|
+
catalogVersionId: string;
|
|
173
|
+
}
|
|
174
|
+
export interface ArtifactSpecContainerGroupContainerImageBuildConfigDockerfile {
|
|
175
|
+
/**
|
|
176
|
+
* Entrypoint baked into the generated Dockerfile CMD. Required when source is `generated`.
|
|
177
|
+
*/
|
|
178
|
+
entrypoints?: string[];
|
|
179
|
+
/**
|
|
180
|
+
* Execution environment ID for the base Docker image. Required when source is `generated`.
|
|
181
|
+
*/
|
|
182
|
+
executionEnvironmentId?: string;
|
|
183
|
+
/**
|
|
184
|
+
* Execution environment version ID that pins the base image. Required when source is `generated`.
|
|
185
|
+
*/
|
|
186
|
+
executionEnvironmentVersionId?: string;
|
|
187
|
+
/**
|
|
188
|
+
* Relative path to the Dockerfile in the source code. Used when source is `provided`. Defaults to `./Dockerfile`.
|
|
189
|
+
*/
|
|
190
|
+
path: string;
|
|
191
|
+
/**
|
|
192
|
+
* How the Dockerfile is obtained: `provided` (from source code) or `generated` (from an execution environment). Defaults to `provided`.
|
|
193
|
+
*/
|
|
194
|
+
source: string;
|
|
195
|
+
}
|
|
150
196
|
export interface ArtifactSpecContainerGroupContainerLivenessProbe {
|
|
151
197
|
/**
|
|
152
198
|
* Minimum consecutive failures for the probe to be considered failed.
|
|
@@ -176,6 +222,10 @@ export interface ArtifactSpecContainerGroupContainerLivenessProbe {
|
|
|
176
222
|
* Scheme to use for connecting to the host (HTTP or HTTPS).
|
|
177
223
|
*/
|
|
178
224
|
scheme: string;
|
|
225
|
+
/**
|
|
226
|
+
* Minimum consecutive successes for the probe to be considered successful after having failed.
|
|
227
|
+
*/
|
|
228
|
+
successThreshold: number;
|
|
179
229
|
/**
|
|
180
230
|
* Number of seconds after which the probe times out.
|
|
181
231
|
*/
|
|
@@ -210,6 +260,10 @@ export interface ArtifactSpecContainerGroupContainerReadinessProbe {
|
|
|
210
260
|
* Scheme to use for connecting to the host (HTTP or HTTPS).
|
|
211
261
|
*/
|
|
212
262
|
scheme: string;
|
|
263
|
+
/**
|
|
264
|
+
* Minimum consecutive successes for the probe to be considered successful after having failed.
|
|
265
|
+
*/
|
|
266
|
+
successThreshold: number;
|
|
213
267
|
/**
|
|
214
268
|
* Number of seconds after which the probe times out.
|
|
215
269
|
*/
|
|
@@ -244,6 +298,10 @@ export interface ArtifactSpecContainerGroupContainerStartupProbe {
|
|
|
244
298
|
* Scheme to use for connecting to the host (HTTP or HTTPS).
|
|
245
299
|
*/
|
|
246
300
|
scheme: string;
|
|
301
|
+
/**
|
|
302
|
+
* Minimum consecutive successes for the probe to be considered successful after having failed.
|
|
303
|
+
*/
|
|
304
|
+
successThreshold: number;
|
|
247
305
|
/**
|
|
248
306
|
* Number of seconds after which the probe times out.
|
|
249
307
|
*/
|
|
@@ -1449,11 +1507,11 @@ export interface GetArtifactSpecContainerGroupContainerEnvironmentVar {
|
|
|
1449
1507
|
*/
|
|
1450
1508
|
key: string;
|
|
1451
1509
|
/**
|
|
1452
|
-
* Name of the environment variable.
|
|
1510
|
+
* Name of the environment variable. May be absent for "api-key" entries, in which case the token is injected as DATAROBOT*API*TOKEN.
|
|
1453
1511
|
*/
|
|
1454
1512
|
name: string;
|
|
1455
1513
|
/**
|
|
1456
|
-
* Source type: "string" for plain text values, "dr-credential" for DataRobot credentials.
|
|
1514
|
+
* Source type: "string" for plain text values, "dr-credential" for DataRobot credentials, "api-key" for a platform-managed DataRobot API token.
|
|
1457
1515
|
*/
|
|
1458
1516
|
source: string;
|
|
1459
1517
|
/**
|
|
@@ -1546,6 +1604,10 @@ export interface GetArtifactSpecContainerGroupContainerLivenessProbe {
|
|
|
1546
1604
|
* Scheme to use for connecting to the host (HTTP or HTTPS).
|
|
1547
1605
|
*/
|
|
1548
1606
|
scheme: string;
|
|
1607
|
+
/**
|
|
1608
|
+
* Minimum consecutive successes for the probe to be considered successful after having failed.
|
|
1609
|
+
*/
|
|
1610
|
+
successThreshold: number;
|
|
1549
1611
|
/**
|
|
1550
1612
|
* Number of seconds after which the probe times out.
|
|
1551
1613
|
*/
|
|
@@ -1580,6 +1642,10 @@ export interface GetArtifactSpecContainerGroupContainerReadinessProbe {
|
|
|
1580
1642
|
* Scheme to use for connecting to the host (HTTP or HTTPS).
|
|
1581
1643
|
*/
|
|
1582
1644
|
scheme: string;
|
|
1645
|
+
/**
|
|
1646
|
+
* Minimum consecutive successes for the probe to be considered successful after having failed.
|
|
1647
|
+
*/
|
|
1648
|
+
successThreshold: number;
|
|
1583
1649
|
/**
|
|
1584
1650
|
* Number of seconds after which the probe times out.
|
|
1585
1651
|
*/
|
|
@@ -1652,6 +1718,10 @@ export interface GetArtifactSpecContainerGroupContainerStartupProbe {
|
|
|
1652
1718
|
* Scheme to use for connecting to the host (HTTP or HTTPS).
|
|
1653
1719
|
*/
|
|
1654
1720
|
scheme: string;
|
|
1721
|
+
/**
|
|
1722
|
+
* Minimum consecutive successes for the probe to be considered successful after having failed.
|
|
1723
|
+
*/
|
|
1724
|
+
successThreshold: number;
|
|
1655
1725
|
/**
|
|
1656
1726
|
* Number of seconds after which the probe times out.
|
|
1657
1727
|
*/
|
|
@@ -1859,11 +1929,11 @@ export interface GetArtifactsArtifactSpecContainerGroupContainerEnvironmentVar {
|
|
|
1859
1929
|
*/
|
|
1860
1930
|
key: string;
|
|
1861
1931
|
/**
|
|
1862
|
-
* Name of the environment variable.
|
|
1932
|
+
* Name of the environment variable. May be absent for "api-key" entries, in which case the token is injected as DATAROBOT*API*TOKEN.
|
|
1863
1933
|
*/
|
|
1864
1934
|
name: string;
|
|
1865
1935
|
/**
|
|
1866
|
-
* Source type: "string" for plain text values, "dr-credential" for DataRobot credentials.
|
|
1936
|
+
* Source type: "string" for plain text values, "dr-credential" for DataRobot credentials, "api-key" for a platform-managed DataRobot API token.
|
|
1867
1937
|
*/
|
|
1868
1938
|
source: string;
|
|
1869
1939
|
/**
|
|
@@ -1956,6 +2026,10 @@ export interface GetArtifactsArtifactSpecContainerGroupContainerLivenessProbe {
|
|
|
1956
2026
|
* Scheme to use for connecting to the host (HTTP or HTTPS).
|
|
1957
2027
|
*/
|
|
1958
2028
|
scheme: string;
|
|
2029
|
+
/**
|
|
2030
|
+
* Minimum consecutive successes for the probe to be considered successful after having failed.
|
|
2031
|
+
*/
|
|
2032
|
+
successThreshold: number;
|
|
1959
2033
|
/**
|
|
1960
2034
|
* Number of seconds after which the probe times out.
|
|
1961
2035
|
*/
|
|
@@ -1990,6 +2064,10 @@ export interface GetArtifactsArtifactSpecContainerGroupContainerReadinessProbe {
|
|
|
1990
2064
|
* Scheme to use for connecting to the host (HTTP or HTTPS).
|
|
1991
2065
|
*/
|
|
1992
2066
|
scheme: string;
|
|
2067
|
+
/**
|
|
2068
|
+
* Minimum consecutive successes for the probe to be considered successful after having failed.
|
|
2069
|
+
*/
|
|
2070
|
+
successThreshold: number;
|
|
1993
2071
|
/**
|
|
1994
2072
|
* Number of seconds after which the probe times out.
|
|
1995
2073
|
*/
|
|
@@ -2062,6 +2140,10 @@ export interface GetArtifactsArtifactSpecContainerGroupContainerStartupProbe {
|
|
|
2062
2140
|
* Scheme to use for connecting to the host (HTTP or HTTPS).
|
|
2063
2141
|
*/
|
|
2064
2142
|
scheme: string;
|
|
2143
|
+
/**
|
|
2144
|
+
* Minimum consecutive successes for the probe to be considered successful after having failed.
|
|
2145
|
+
*/
|
|
2146
|
+
successThreshold: number;
|
|
2065
2147
|
/**
|
|
2066
2148
|
* Number of seconds after which the probe times out.
|
|
2067
2149
|
*/
|
|
@@ -2249,29 +2331,25 @@ export interface WorkloadRuntimeContainerGroupAutoscaling {
|
|
|
2249
2331
|
*/
|
|
2250
2332
|
enabled: boolean;
|
|
2251
2333
|
/**
|
|
2252
|
-
*
|
|
2334
|
+
* Maximum number of replicas. Defaults to `1`.
|
|
2253
2335
|
*/
|
|
2254
|
-
|
|
2255
|
-
}
|
|
2256
|
-
export interface WorkloadRuntimeContainerGroupAutoscalingPolicy {
|
|
2257
|
-
/**
|
|
2258
|
-
* Maximum number of replicas.
|
|
2259
|
-
*/
|
|
2260
|
-
maxCount: number;
|
|
2336
|
+
maxReplicaCount: number;
|
|
2261
2337
|
/**
|
|
2262
|
-
* Minimum number of replicas.
|
|
2338
|
+
* Minimum number of replicas. Set to `0` to allow scale-to-zero. Defaults to `0`.
|
|
2263
2339
|
*/
|
|
2264
|
-
|
|
2340
|
+
minReplicaCount: number;
|
|
2265
2341
|
/**
|
|
2266
|
-
*
|
|
2342
|
+
* Scaling policies that define when and how to scale.
|
|
2267
2343
|
*/
|
|
2268
|
-
|
|
2344
|
+
policies: outputs.WorkloadRuntimeContainerGroupAutoscalingPolicy[];
|
|
2345
|
+
}
|
|
2346
|
+
export interface WorkloadRuntimeContainerGroupAutoscalingPolicy {
|
|
2269
2347
|
/**
|
|
2270
|
-
* Metric used for scaling decisions: `cpuAverageUtilization`, `httpRequestsConcurrency`, `gpuCacheUtilization`, or `gpuRequestQueueDepth`.
|
|
2348
|
+
* Metric used for scaling decisions: `cpuAverageUtilization`, `httpRequestsConcurrency`, `gpuCacheUtilization`, or `gpuRequestQueueDepth`. Custom metric names (e.g. `vllm:kv_cache_usage_perc`) are supported for NIM artifacts only.
|
|
2271
2349
|
*/
|
|
2272
2350
|
scalingMetric: string;
|
|
2273
2351
|
/**
|
|
2274
|
-
* Target value for the scaling metric.
|
|
2352
|
+
* Target value for the scaling metric. Must be non-negative.
|
|
2275
2353
|
*/
|
|
2276
2354
|
target: number;
|
|
2277
2355
|
}
|
package/workload.d.ts
CHANGED
|
@@ -4,11 +4,7 @@ import * as outputs from "./types/output";
|
|
|
4
4
|
/**
|
|
5
5
|
* A Workload runs a containerized artifact in the cluster and exposes an inference endpoint.
|
|
6
6
|
*
|
|
7
|
-
*
|
|
8
|
-
*
|
|
9
|
-
* ```typescript
|
|
10
|
-
* import * as pulumi from "@pulumi/pulumi";
|
|
11
|
-
* ```
|
|
7
|
+
* Changes to `artifactId` or `runtime` trigger an in-place workload replacement via the Workload API. The workload ID and endpoint remain stable across artifact and runtime updates.
|
|
12
8
|
*/
|
|
13
9
|
export declare class Workload extends pulumi.CustomResource {
|
|
14
10
|
/**
|
|
@@ -27,7 +23,7 @@ export declare class Workload extends pulumi.CustomResource {
|
|
|
27
23
|
*/
|
|
28
24
|
static isInstance(obj: any): obj is Workload;
|
|
29
25
|
/**
|
|
30
|
-
* ID of the Artifact version to deploy. When using `datarobot.Artifact`, reference `datarobot_artifact.<name>.artifact_id` (not `.id`). Changing this value
|
|
26
|
+
* ID of the Artifact version to deploy. When using `datarobot.Artifact`, reference `datarobot_artifact.<name>.artifact_id` (not `.id`). Changing this value triggers an in-place workload replacement.
|
|
31
27
|
*/
|
|
32
28
|
readonly artifactId: pulumi.Output<string>;
|
|
33
29
|
/**
|
|
@@ -47,7 +43,7 @@ export declare class Workload extends pulumi.CustomResource {
|
|
|
47
43
|
*/
|
|
48
44
|
readonly name: pulumi.Output<string>;
|
|
49
45
|
/**
|
|
50
|
-
* Runtime configuration for the Workload.
|
|
46
|
+
* Runtime configuration for the Workload. Changes trigger an in-place workload replacement.
|
|
51
47
|
*/
|
|
52
48
|
readonly runtime: pulumi.Output<outputs.WorkloadRuntime>;
|
|
53
49
|
/**
|
|
@@ -68,7 +64,7 @@ export declare class Workload extends pulumi.CustomResource {
|
|
|
68
64
|
*/
|
|
69
65
|
export interface WorkloadState {
|
|
70
66
|
/**
|
|
71
|
-
* ID of the Artifact version to deploy. When using `datarobot.Artifact`, reference `datarobot_artifact.<name>.artifact_id` (not `.id`). Changing this value
|
|
67
|
+
* ID of the Artifact version to deploy. When using `datarobot.Artifact`, reference `datarobot_artifact.<name>.artifact_id` (not `.id`). Changing this value triggers an in-place workload replacement.
|
|
72
68
|
*/
|
|
73
69
|
artifactId?: pulumi.Input<string>;
|
|
74
70
|
/**
|
|
@@ -88,7 +84,7 @@ export interface WorkloadState {
|
|
|
88
84
|
*/
|
|
89
85
|
name?: pulumi.Input<string>;
|
|
90
86
|
/**
|
|
91
|
-
* Runtime configuration for the Workload.
|
|
87
|
+
* Runtime configuration for the Workload. Changes trigger an in-place workload replacement.
|
|
92
88
|
*/
|
|
93
89
|
runtime?: pulumi.Input<inputs.WorkloadRuntime>;
|
|
94
90
|
/**
|
|
@@ -101,7 +97,7 @@ export interface WorkloadState {
|
|
|
101
97
|
*/
|
|
102
98
|
export interface WorkloadArgs {
|
|
103
99
|
/**
|
|
104
|
-
* ID of the Artifact version to deploy. When using `datarobot.Artifact`, reference `datarobot_artifact.<name>.artifact_id` (not `.id`). Changing this value
|
|
100
|
+
* ID of the Artifact version to deploy. When using `datarobot.Artifact`, reference `datarobot_artifact.<name>.artifact_id` (not `.id`). Changing this value triggers an in-place workload replacement.
|
|
105
101
|
*/
|
|
106
102
|
artifactId: pulumi.Input<string>;
|
|
107
103
|
/**
|
|
@@ -117,7 +113,7 @@ export interface WorkloadArgs {
|
|
|
117
113
|
*/
|
|
118
114
|
name?: pulumi.Input<string>;
|
|
119
115
|
/**
|
|
120
|
-
* Runtime configuration for the Workload.
|
|
116
|
+
* Runtime configuration for the Workload. Changes trigger an in-place workload replacement.
|
|
121
117
|
*/
|
|
122
118
|
runtime: pulumi.Input<inputs.WorkloadRuntime>;
|
|
123
119
|
}
|
package/workload.js
CHANGED
|
@@ -8,11 +8,7 @@ const utilities = require("./utilities");
|
|
|
8
8
|
/**
|
|
9
9
|
* A Workload runs a containerized artifact in the cluster and exposes an inference endpoint.
|
|
10
10
|
*
|
|
11
|
-
*
|
|
12
|
-
*
|
|
13
|
-
* ```typescript
|
|
14
|
-
* import * as pulumi from "@pulumi/pulumi";
|
|
15
|
-
* ```
|
|
11
|
+
* Changes to `artifactId` or `runtime` trigger an in-place workload replacement via the Workload API. The workload ID and endpoint remain stable across artifact and runtime updates.
|
|
16
12
|
*/
|
|
17
13
|
class Workload extends pulumi.CustomResource {
|
|
18
14
|
/**
|
package/workload.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"workload.js","sourceRoot":"","sources":["../workload.ts"],"names":[],"mappings":";AAAA,wFAAwF;AACxF,iFAAiF;;;AAEjF,yCAAyC;AAGzC,yCAAyC;AAEzC
|
|
1
|
+
{"version":3,"file":"workload.js","sourceRoot":"","sources":["../workload.ts"],"names":[],"mappings":";AAAA,wFAAwF;AACxF,iFAAiF;;;AAEjF,yCAAyC;AAGzC,yCAAyC;AAEzC;;;;GAIG;AACH,MAAa,QAAS,SAAQ,MAAM,CAAC,cAAc;IAC/C;;;;;;;;OAQG;IACI,MAAM,CAAC,GAAG,CAAC,IAAY,EAAE,EAA2B,EAAE,KAAqB,EAAE,IAAmC;QACnH,OAAO,IAAI,QAAQ,CAAC,IAAI,EAAO,KAAK,kCAAO,IAAI,KAAE,EAAE,EAAE,EAAE,IAAG,CAAC;IAC/D,CAAC;IAKD;;;OAGG;IACI,MAAM,CAAC,UAAU,CAAC,GAAQ;QAC7B,IAAI,GAAG,KAAK,SAAS,IAAI,GAAG,KAAK,IAAI,EAAE;YACnC,OAAO,KAAK,CAAC;SAChB;QACD,OAAO,GAAG,CAAC,cAAc,CAAC,KAAK,QAAQ,CAAC,YAAY,CAAC;IACzD,CAAC;IAuCD,YAAY,IAAY,EAAE,WAA0C,EAAE,IAAmC;QACrG,IAAI,cAAc,GAAkB,EAAE,CAAC;QACvC,IAAI,GAAG,IAAI,IAAI,EAAE,CAAC;QAClB,IAAI,IAAI,CAAC,EAAE,EAAE;YACT,MAAM,KAAK,GAAG,WAAwC,CAAC;YACvD,cAAc,CAAC,YAAY,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC;YACpE,cAAc,CAAC,aAAa,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;YACtE,cAAc,CAAC,UAAU,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,SAAS,CAAC;YAChE,cAAc,CAAC,YAAY,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC;YACpE,cAAc,CAAC,MAAM,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC;YACxD,cAAc,CAAC,SAAS,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC;YAC9D,cAAc,CAAC,QAAQ,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC;SAC/D;aAAM;YACH,MAAM,IAAI,GAAG,WAAuC,CAAC;YACrD,IAAI,CAAC,CAAC,IAAI,IAAI,IAAI,CAAC,UAAU,KAAK,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE;gBACvD,MAAM,IAAI,KAAK,CAAC,wCAAwC,CAAC,CAAC;aAC7D;YACD,IAAI,CAAC,CAAC,IAAI,IAAI,IAAI,CAAC,OAAO,KAAK,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE;gBACpD,MAAM,IAAI,KAAK,CAAC,qCAAqC,CAAC,CAAC;aAC1D;YACD,cAAc,CAAC,YAAY,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC;YAClE,cAAc,CAAC,aAAa,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;YACpE,cAAc,CAAC,YAAY,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC;YAClE,cAAc,CAAC,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC;YACtD,cAAc,CAAC,SAAS,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC;YAC5D,cAAc,CAAC,UAAU,CAAC,GAAG,SAAS,CAAC,OAAO,CAAC;YAC/C,cAAc,CAAC,QAAQ,CAAC,GAAG,SAAS,CAAC,OAAO,CAAC;SAChD;QACD,IAAI,GAAG,MAAM,CAAC,YAAY,CAAC,SAAS,CAAC,oBAAoB,EAAE,EAAE,IAAI,CAAC,CAAC;QACnE,KAAK,CAAC,QAAQ,CAAC,YAAY,EAAE,IAAI,EAAE,cAAc,EAAE,IAAI,CAAC,CAAC;IAC7D,CAAC;;AA/FL,4BAgGC;AAlFG,gBAAgB;AACO,qBAAY,GAAG,mCAAmC,CAAC"}
|