@infernetprotocol/deploy-providers 0.1.22 → 0.1.25

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 CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@infernetprotocol/deploy-providers",
3
- "version": "0.1.22",
3
+ "version": "0.1.25",
4
4
  "description": "Adapters for deploying Infernet provider nodes to cloud GPU services (RunPod, vast.ai, Lambda, etc.).",
5
5
  "license": "MIT",
6
6
  "type": "module",
@@ -25,7 +25,7 @@
25
25
  },
26
26
  "repository": {
27
27
  "type": "git",
28
- "url": "git+https://github.com/profullstack/infernet-protocol.git",
28
+ "url": "git+https://github.com/InfernetProtocol/infernet-protocol.git",
29
29
  "directory": "packages/deploy-providers"
30
30
  }
31
31
  }
@@ -119,3 +119,261 @@ export async function destroyDeployment({ apiKey, deploymentId }) {
119
119
  await doFetch({ apiKey, path: `/droplets/${deploymentId}`, method: "DELETE" });
120
120
  return { ok: true };
121
121
  }
122
+
123
+ // ---- IPIP-0019 DeployProvider class adapter ----
124
+
125
+ import { DeployProvider, NotSupportedError } from "./providers/base.js";
126
+ import { canonicalize, aliasesFor } from "./gpu-normalize.js";
127
+ import * as state from "./state.js";
128
+ import { composeInstallScript, pollVllmHealth } from "./vllm-bootstrap.js";
129
+
130
+ const DEFAULT_INSTALLER =
131
+ process.env.INFERNET_INSTALLER_URL
132
+ ?? "https://infernetprotocol.com/install.sh";
133
+
134
+ // DO publishes monthly prices; convert to hourly with a 730-hour month.
135
+ const HOURS_PER_MONTH = 730;
136
+
137
+ export class DigitalOceanProvider extends DeployProvider {
138
+ constructor(config = {}) {
139
+ super({ ...config, providerId: "digitalocean" });
140
+ }
141
+
142
+ async _fetch(path, { method = "GET", body } = {}) {
143
+ return doFetch({ apiKey: this.config.apiKey, path, method, body });
144
+ }
145
+
146
+ async validateAuth() {
147
+ const data = await this._fetch("/account");
148
+ if (!data?.account) throw new Error("DigitalOcean: API key invalid");
149
+ return { ok: true, accountInfo: data.account };
150
+ }
151
+
152
+ async listGpuTypes() {
153
+ const data = await this._fetch("/sizes?per_page=200");
154
+ const sizes = (data?.sizes ?? []).filter((s) => s.slug?.startsWith("gpu-"));
155
+ return sizes.map((s) => ({
156
+ id: s.slug,
157
+ name: s.description ?? s.slug,
158
+ vramGb: this._inferVramFromSlug(s.slug),
159
+ pricePerHour: (s.price_monthly ?? 0) / HOURS_PER_MONTH
160
+ }));
161
+ }
162
+
163
+ _inferVramFromSlug(slug) {
164
+ const m = slug?.match(/-(\d+)gb/i);
165
+ return m ? Number(m[1]) : null;
166
+ }
167
+
168
+ async listRegions() {
169
+ const data = await this._fetch("/regions");
170
+ return (data?.regions ?? [])
171
+ .filter((r) => r.available)
172
+ .map((r) => ({ id: r.slug, name: r.name }));
173
+ }
174
+
175
+ async findOffers(request = {}) {
176
+ const data = await this._fetch("/sizes?per_page=200");
177
+ const sizes = (data?.sizes ?? []).filter((s) => s.slug?.startsWith("gpu-"));
178
+
179
+ const wantCanonical = request.gpu ? canonicalize(request.gpu) : null;
180
+ const aliases = wantCanonical
181
+ ? [wantCanonical, ...aliasesFor(wantCanonical)].map((a) => a.toLowerCase())
182
+ : null;
183
+
184
+ const offers = [];
185
+ for (const s of sizes) {
186
+ const slug = s.slug.toLowerCase();
187
+ const desc = (s.description ?? "").toLowerCase();
188
+ if (aliases && !aliases.some((a) => slug.includes(a) || desc.includes(a))) continue;
189
+ const vramGb = this._inferVramFromSlug(s.slug);
190
+ if (request.vramMin && vramGb && vramGb < request.vramMin) continue;
191
+ const pricePerHour = (s.price_monthly ?? 0) / HOURS_PER_MONTH;
192
+ if (request.maxPricePerHour != null && pricePerHour > request.maxPricePerHour) continue;
193
+
194
+ for (const region of s.regions ?? []) {
195
+ if (request.region && region !== request.region) continue;
196
+ offers.push({
197
+ providerId: "digitalocean",
198
+ offerId: `${s.slug}@${region}`,
199
+ gpu: { name: s.description ?? s.slug, count: 1, vramGb: vramGb ?? 0 },
200
+ cpu: { cores: s.vcpus },
201
+ ramGb: s.memory ? Math.round(s.memory / 1024) : undefined,
202
+ diskGb: s.disk,
203
+ region,
204
+ pricePerHour,
205
+ deployStyle: "vm-ssh",
206
+ available: true,
207
+ raw: { slug: s.slug, region, size: s }
208
+ });
209
+ }
210
+ }
211
+ return offers;
212
+ }
213
+
214
+ _buildUserData(envMap, opts = {}) {
215
+ if (opts.engine === "vllm") {
216
+ return composeInstallScript({
217
+ infernetEnv: envMap,
218
+ vllmConfig: opts.vllmConfig ?? {}
219
+ });
220
+ }
221
+ const exports = Object.entries(envMap)
222
+ .map(([k, v]) => `export ${k}=${JSON.stringify(String(v))}`)
223
+ .join("\n");
224
+ return [
225
+ "#!/bin/bash",
226
+ "set -eux",
227
+ exports,
228
+ `curl -fsSL ${DEFAULT_INSTALLER} | bash`,
229
+ "infernet register || true",
230
+ "infernet start || true"
231
+ ].join("\n");
232
+ }
233
+
234
+ async createNode(request = {}) {
235
+ const {
236
+ gpu, name, size, region, image, sshKeyIds = [],
237
+ env = {}, controlPlaneUrl = "", engine = "ollama", model = null,
238
+ hourlyPrice = 0, vramGb = 0
239
+ } = request;
240
+
241
+ // Resolve size/region from canonical GPU if not given explicitly.
242
+ let finalSize = size;
243
+ let finalRegion = region;
244
+ if (!finalSize) {
245
+ const offers = await this.findOffers({ gpu, region });
246
+ if (offers.length === 0) throw new Error(`digitalocean: no GPU droplet matched gpu="${gpu}"${region ? ` in region=${region}` : ""}`);
247
+ offers.sort((a, b) => a.pricePerHour - b.pricePerHour);
248
+ finalSize = offers[0].raw.slug;
249
+ finalRegion = offers[0].raw.region;
250
+ }
251
+ if (!finalRegion) throw new Error("digitalocean: createNode requires region");
252
+
253
+ const localId = state.generateNodeId();
254
+ const dropletName = name ?? localId;
255
+
256
+ const userData = this._buildUserData({
257
+ INFERNET_NODE_NAME: dropletName,
258
+ INFERNET_CONTROL_PLANE_URL: controlPlaneUrl,
259
+ INFERNET_ENGINE: engine,
260
+ ...(model ? { INFERNET_MODEL: model } : {}),
261
+ ...env
262
+ }, { engine, vllmConfig: { model, ...(request.vllm ?? {}) } });
263
+
264
+ const result = await createDeployment({
265
+ apiKey: this.config.apiKey,
266
+ name: dropletName,
267
+ size: finalSize,
268
+ region: finalRegion,
269
+ userData, image, sshKeyIds
270
+ });
271
+
272
+ const record = {
273
+ id: localId,
274
+ provider: "digitalocean",
275
+ providerNodeId: result.deploymentId,
276
+ name: dropletName,
277
+ gpu: gpu ?? finalSize,
278
+ gpuCount: 1,
279
+ vramGb,
280
+ region: finalRegion,
281
+ hourlyPrice,
282
+ engine,
283
+ model: model ?? undefined,
284
+ status: "creating",
285
+ createdAt: new Date().toISOString(),
286
+ controlPlaneUrl
287
+ };
288
+ await state.saveNode(record);
289
+ return record;
290
+ }
291
+
292
+ async waitUntilReady(node, timeoutMs = 5 * 60 * 1000) {
293
+ if (!node?.providerNodeId) throw new Error("digitalocean: waitUntilReady requires node.providerNodeId");
294
+ const deadline = Date.now() + timeoutMs;
295
+ while (Date.now() < deadline) {
296
+ const live = await getDeployment({ apiKey: this.config.apiKey, deploymentId: node.providerNodeId });
297
+ if ((live?.status ?? "").toLowerCase() === "active" && live?.endpoint) {
298
+ const endpointUrl = `http://${live.endpoint}:46337`;
299
+ return state.updateNode(node.id, { status: "running", endpointUrl, ip: live.endpoint });
300
+ }
301
+ await new Promise((r) => setTimeout(r, 5000));
302
+ }
303
+ await state.updateNode(node.id, { status: "error", error_log: "droplet did not reach ACTIVE in time" });
304
+ throw new Error(`DigitalOcean droplet ${node.providerNodeId} did not reach ACTIVE within ${Math.round(timeoutMs / 1000)}s`);
305
+ }
306
+
307
+ async bootstrapNode(node, request = {}) {
308
+ if (!node?.endpointUrl && !node?.ip) return { ok: false, reason: "no endpointUrl/ip on node" };
309
+ if ((request.engine ?? node.engine) === "vllm") {
310
+ return pollVllmHealth(node.ip ?? node.endpointUrl, { port: request.vllm?.port ?? 8000 });
311
+ }
312
+ const healthEndpoint = `${node.endpointUrl}/health`;
313
+ const deadline = Date.now() + 5 * 60 * 1000;
314
+ let lastErr = null;
315
+ while (Date.now() < deadline) {
316
+ try {
317
+ const ctrl = new AbortController();
318
+ const t = setTimeout(() => ctrl.abort(), 5000);
319
+ const res = await fetch(healthEndpoint, { signal: ctrl.signal });
320
+ clearTimeout(t);
321
+ if (res.ok) return { ok: true, healthEndpoint };
322
+ lastErr = `HTTP ${res.status}`;
323
+ } catch (err) { lastErr = err?.message ?? String(err); }
324
+ await new Promise((r) => setTimeout(r, 10000));
325
+ }
326
+ return { ok: false, reason: lastErr ?? "bootstrap health-check timed out", healthEndpoint };
327
+ }
328
+
329
+ async getNode(nodeId) {
330
+ const local = await state.loadNode(nodeId);
331
+ if (!local || local.provider !== "digitalocean") throw new Error(`digitalocean: no local record for nodeId=${nodeId}`);
332
+ try {
333
+ const live = await getDeployment({ apiKey: this.config.apiKey, deploymentId: local.providerNodeId });
334
+ if (live?.status) local.liveStatus = live.status;
335
+ } catch { /* best-effort */ }
336
+ return local;
337
+ }
338
+
339
+ async listNodes() {
340
+ const all = await state.listNodes();
341
+ return all.filter((n) => n.provider === "digitalocean");
342
+ }
343
+
344
+ async stopNode(nodeId) {
345
+ const local = await state.loadNode(nodeId);
346
+ if (!local || local.provider !== "digitalocean") throw new Error(`digitalocean: no local record for nodeId=${nodeId}`);
347
+ await this._fetch(`/droplets/${local.providerNodeId}/actions`, { method: "POST", body: { type: "power_off" } });
348
+ await state.updateNode(nodeId, { status: "stopped" });
349
+ return { ok: true };
350
+ }
351
+
352
+ async startNode(nodeId) {
353
+ const local = await state.loadNode(nodeId);
354
+ if (!local || local.provider !== "digitalocean") throw new Error(`digitalocean: no local record for nodeId=${nodeId}`);
355
+ await this._fetch(`/droplets/${local.providerNodeId}/actions`, { method: "POST", body: { type: "power_on" } });
356
+ await state.updateNode(nodeId, { status: "running" });
357
+ return { ok: true };
358
+ }
359
+
360
+ async destroyNode(nodeId) {
361
+ const local = await state.loadNode(nodeId);
362
+ if (!local || local.provider !== "digitalocean") throw new Error(`digitalocean: no local record for nodeId=${nodeId}`);
363
+ await destroyDeployment({ apiKey: this.config.apiKey, deploymentId: local.providerNodeId });
364
+ await state.updateNode(nodeId, { status: "destroyed" });
365
+ return { ok: true };
366
+ }
367
+
368
+ async logs(nodeId, _opts = {}) {
369
+ const local = await state.loadNode(nodeId).catch(() => null);
370
+ const err = new NotSupportedError("logs", "digitalocean");
371
+ const sshHint = local?.ip
372
+ ? `\nSSH and tail the daemon log:\n ssh root@${local.ip} 'tail -200 /var/log/infernet/daemon.log'`
373
+ : "";
374
+ err.message = `digitalocean: droplet logs are not exposed via the API.\nView at: https://cloud.digitalocean.com/droplets` + sshHint;
375
+ throw err;
376
+ }
377
+ }
378
+
379
+ export default DigitalOceanProvider;
package/src/index.js CHANGED
@@ -1,22 +1,51 @@
1
1
  /**
2
2
  * Deploy-provider registry.
3
3
  *
4
- * Each adapter implements (or partially implements):
5
- * meta — { id, label, keyUrl, sizesDoc, regionsDoc }
6
- * listGpuTypes — (apiKey) → catalog of available GPU types
7
- * searchOffers — (apiKey, criteria) → marketplace results (Vast.ai)
8
- * createDeployment — (apiKey, …) → { deploymentId, status, endpoint }
9
- * getDeployment — (apiKey, deploymentId) → { status, endpoint }
10
- * destroyDeployment — (apiKey, deploymentId) → { ok }
4
+ * Two surfaces are exposed for backwards compatibility during the
5
+ * IPIP-0019 rollout:
11
6
  *
12
- * All adapters are stateless — they never persist the user's cloud
13
- * API key. The CLI / web layer passes the key through and forgets it.
7
+ * adapters — legacy module-export shape (createDeployment,
8
+ * getDeployment, destroyDeployment). Used by the
9
+ * existing per-provider CLI commands. STABLE.
10
+ *
11
+ * providers — IPIP-0019 DeployProvider class registry. Each entry
12
+ * is a class extending ./providers/base.js → instantiate
13
+ * with `new providers.runpod({ apiKey })` and call
14
+ * findOffers / createNode / waitUntilReady / etc.
15
+ * Used by lifecycle commands and pricing-aware preset
16
+ * dispatch. THIS IS THE FORWARD-LOOKING SURFACE.
17
+ *
18
+ * Adding a new provider:
19
+ * 1. Drop packages/deploy-providers/src/providers/<id>.js exporting a
20
+ * class extending DeployProvider, with providerId === "<id>".
21
+ * 2. Import + register it in the `providers` map below.
22
+ * 3. Add it to PROVIDER_KEY_URLS so the CLI knows where to send users
23
+ * to mint an API key.
24
+ * 4. Add reliability / dx scores in pricing.js so it participates in
25
+ * cross-provider ranking.
26
+ *
27
+ * Adapters are stateless w.r.t. credentials — they never persist the
28
+ * user's cloud API key. The CLI / web layer passes the key through and
29
+ * forgets it after the request.
14
30
  */
15
31
 
16
32
  import * as runpod from "./runpod.js";
17
33
  import * as digitalocean from "./digitalocean.js";
18
34
  import * as vast from "./vast.js";
19
35
 
36
+ import { DeployProvider, NotSupportedError } from "./providers/base.js";
37
+ import RunPodProvider from "./providers/runpod.js";
38
+ import TensorDockProvider from "./providers/tensordock.js";
39
+ import LambdaProvider from "./providers/lambda.js";
40
+ import HyperstackProvider from "./providers/hyperstack.js";
41
+ import CrusoeProvider from "./providers/crusoe.js";
42
+ import DataCrunchProvider from "./providers/datacrunch.js";
43
+ import VoltageParkProvider from "./providers/voltagepark.js";
44
+ import { VastProvider } from "./vast.js";
45
+ import { DigitalOceanProvider } from "./digitalocean.js";
46
+
47
+ // ---- Legacy module-export adapters (stable, used by existing CLI) ----
48
+
20
49
  export const adapters = {
21
50
  runpod,
22
51
  digitalocean,
@@ -29,10 +58,64 @@ export function getAdapter(name) {
29
58
  return adapter;
30
59
  }
31
60
 
61
+ export { runpod, digitalocean, vast };
62
+
63
+ // ---- IPIP-0019 DeployProvider class registry (forward-looking) ----
64
+
65
+ export const providers = {
66
+ runpod: RunPodProvider,
67
+ tensordock: TensorDockProvider,
68
+ lambda: LambdaProvider,
69
+ hyperstack: HyperstackProvider,
70
+ crusoe: CrusoeProvider,
71
+ datacrunch: DataCrunchProvider,
72
+ voltagepark: VoltageParkProvider,
73
+ vast: VastProvider,
74
+ digitalocean: DigitalOceanProvider
75
+ };
76
+
77
+ /**
78
+ * Instantiate a DeployProvider class by id. Throws if the id is unknown
79
+ * or if the apiKey is missing.
80
+ */
81
+ export function getProvider(id, config) {
82
+ const Cls = providers[id];
83
+ if (!Cls) {
84
+ const known = Object.keys(providers).join(", ") || "(none registered)";
85
+ throw new Error(`Unknown DeployProvider id: ${id}. Known: ${known}`);
86
+ }
87
+ return new Cls(config);
88
+ }
89
+
90
+ /** Provider ids that have a class implementation registered. */
91
+ export function listProviderIds() {
92
+ return Object.keys(providers);
93
+ }
94
+
95
+ export {
96
+ DeployProvider,
97
+ NotSupportedError,
98
+ RunPodProvider,
99
+ TensorDockProvider,
100
+ LambdaProvider,
101
+ HyperstackProvider,
102
+ CrusoeProvider,
103
+ DataCrunchProvider,
104
+ VoltageParkProvider,
105
+ VastProvider,
106
+ DigitalOceanProvider
107
+ };
108
+
109
+ // ---- Provider key URLs (where to mint an API key) ----
110
+
32
111
  export const PROVIDER_KEY_URLS = {
33
112
  digitalocean: digitalocean.meta.keyUrl,
34
- vast: vast.meta.keyUrl,
35
- runpod: "https://www.runpod.io/console/user/settings"
113
+ vast: vast.meta.keyUrl,
114
+ runpod: "https://www.runpod.io/console/user/settings",
115
+ tensordock: "https://dashboard.tensordock.com/api",
116
+ lambda: "https://cloud.lambdalabs.com/api-keys",
117
+ hyperstack: "https://infrahub.nexgencloud.com/dashboard/api-keys",
118
+ crusoe: "https://console.crusoecloud.com/security/tokens",
119
+ datacrunch: "https://cloud.datacrunch.io/account/api",
120
+ voltagepark: "https://dashboard.voltagepark.com/api-keys"
36
121
  };
37
-
38
- export { runpod, digitalocean, vast };
package/src/pricing.js CHANGED
@@ -26,7 +26,12 @@ const PROVIDER_SCORES = Object.freeze({
26
26
  lambda: { reliability: 0.90, dx: 0.80 },
27
27
  tensordock: { reliability: 0.75, dx: 0.70 },
28
28
  digitalocean: { reliability: 0.90, dx: 0.85 },
29
- vast: { reliability: 0.65, dx: 0.65 }
29
+ vast: { reliability: 0.65, dx: 0.65 },
30
+ // "Stable tier" — initial guesses, refine from observed deploy success.
31
+ hyperstack: { reliability: 0.85, dx: 0.80 },
32
+ crusoe: { reliability: 0.88, dx: 0.80 },
33
+ datacrunch: { reliability: 0.82, dx: 0.75 },
34
+ voltagepark: { reliability: 0.80, dx: 0.70 }
30
35
  });
31
36
 
32
37
  export function isValidPreset(name) {