trainfabric 0.1.46 → 0.1.47

package/dist/index.cjs

@@ -8626,9 +8626,25 @@ function encodeRunId(runId) {
 }
 function isTransientQuoteError(error) {
   const message = error instanceof Error ? error.message : String(error);
-  return /status (502|503|504)\b/i.test(message);
+  return /status (502|503|504|429)\b|fetch failed|network|ECONNRESET|ETIMEDOUT/i.test(message);
 }
 var RunsClient = class extends ResourceClient {
+  async retryQuoteOperation(operation) {
+    const attempts = Math.max(1, Number(process.env.TRAINFABRIC_QUOTE_RETRY_ATTEMPTS ?? 5));
+    let lastError;
+    for (let attempt = 1; attempt <= attempts; attempt += 1) {
+      try {
+        return await operation();
+      } catch (error) {
+        lastError = error;
+        if (!isTransientQuoteError(error) || attempt === attempts) {
+          throw error;
+        }
+        await wait(Math.min(5e3, 400 * 2 ** (attempt - 1)));
+      }
+    }
+    throw lastError instanceof Error ? lastError : new Error("Quote request failed.");
+  }
   describeEmptyQuoteFailure(input) {
     const compute = input.compute && input.compute !== "auto" && input.compute !== "cpu-sim" ? input.compute : void 0;
     const requestedAccelerator = compute?.acceleratorClass;
@@ -8643,7 +8659,15 @@ var RunsClient = class extends ResourceClient {
     if (constraints.length === 0) {
       return "No quote options were returned for the requested run.";
     }
-    return `No quote options were returned for the requested run under the current hardware constraints (${constraints.join(", ")}). Omit the accelerator constraint or request a larger GPU memory floor.`;
+    const suggestions = [];
+    if (requestedAccelerator) {
+      suggestions.push("omit the accelerator constraint");
+    }
+    if (requestedMemory) {
+      suggestions.push("lower --min-memory if it excludes available GPUs");
+    }
+    suggestions.push("try a different mode or retry when capacity changes");
+    return `No quote options were returned for the requested run under the current hardware constraints (${constraints.join(", ")}). ${suggestions.join(", ")}.`;
   }
   profile(input) {
     const payload = jobProfileSchema.parse({
@@ -8679,11 +8703,11 @@ var RunsClient = class extends ResourceClient {
     if (profileSessionId) {
       return this.waitForProfile(profileSessionId);
     }
-    const session = await this.profile(input);
-    return this.waitForProfile(session.id);
+    const session = await this.retryQuoteOperation(() => this.profile(input));
+    return this.retryQuoteOperation(() => this.waitForProfile(session.id));
   }
   async quote(input) {
-    const detail = await this.ensureReadyProfile(input);
+    const detail = await this.retryQuoteOperation(() => this.ensureReadyProfile(input));
     const payload = runQuotesSchema.parse({
       ...input,
       projectId: input.projectId ?? this.parent.projectId,
@@ -8692,7 +8716,7 @@ var RunsClient = class extends ResourceClient {
     return this.requestQuoteWithRetry("/v1/runs/quotes", payload);
   }
   async quoteLegacy(input) {
-    const detail = await this.ensureReadyProfile(input);
+    const detail = await this.retryQuoteOperation(() => this.ensureReadyProfile(input));
     const payload = runQuoteSchema.parse({
       ...input,
       projectId: input.projectId ?? this.parent.projectId,
@@ -8701,20 +8725,7 @@ var RunsClient = class extends ResourceClient {
     return this.requestQuoteWithRetry("/v1/runs/quote", payload);
   }
   async requestQuoteWithRetry(pathname, payload) {
-    const attempts = Math.max(1, Number(process.env.TRAINFABRIC_QUOTE_RETRY_ATTEMPTS ?? 3));
-    let lastError;
-    for (let attempt = 1; attempt <= attempts; attempt += 1) {
-      try {
-        return await this.requestPost(pathname, payload);
-      } catch (error) {
-        lastError = error;
-        if (!isTransientQuoteError(error) || attempt === attempts) {
-          throw error;
-        }
-        await wait(500 * attempt);
-      }
-    }
-    throw lastError instanceof Error ? lastError : new Error("Quote request failed.");
+    return this.retryQuoteOperation(() => this.requestPost(pathname, payload));
   }
   async create(input) {
     const projectId = input.projectId ?? this.parent.projectId;
@@ -9351,7 +9362,7 @@ function buildComputeSpec(options) {
 
 // src/index.ts
 var DEFAULT_TRAINFABRIC_API_URL2 = "https://api.trainfabric.com";
-var CLI_VERSION = "0.1.46";
+var CLI_VERSION = "0.1.47";
 var CONFIG_DIR = import_node_path3.default.join(import_node_os2.default.homedir(), ".trainfabric");
 var CONFIG_PATH = import_node_path3.default.join(CONFIG_DIR, "config.json");
 var FALLBACK_SECRET_PATH = import_node_path3.default.join(CONFIG_DIR, "session.enc");
@@ -9829,10 +9840,18 @@ function assertQuoteOptions(bundle, options) {
   if (constraints.length === 0) {
     throw new Error("No quote options were returned for the requested run.");
   }
+  const suggestions = [];
+  if (options.accelerator) {
+    suggestions.push("omit --accelerator");
+  }
+  if (options.minMemory) {
+    suggestions.push("adjust --min-memory if it excludes available GPUs");
+  }
+  suggestions.push("try a different --mode or retry when capacity changes");
   throw new Error(
     `No quote options were returned for the requested run under the current hardware constraints (${constraints.join(
       ", "
-    )}). Try a different --mode, omit --accelerator, adjust --min-memory if it excludes available GPUs, or retry when capacity changes.`
+    )}). ${suggestions.join(", ")}.`
   );
 }
 function filterQuoteBundleForRequestedMode(bundle, options) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "trainfabric",
-  "version": "0.1.46",
+  "version": "0.1.47",
   "description": "Trainfabric CLI for launching GPU training jobs on the hosted Trainfabric backend.",
   "type": "module",
   "main": "dist/index.cjs",