fix(api): map upstream timeouts to 504 responses

ts/api/handlers/chat.ts

@@ -6,6 +6,7 @@ import * as http from 'node:http';
 import type { IApiError, IChatCompletionRequest } from '../../interfaces/api.ts';
 import { ClusterCoordinator } from '../../cluster/coordinator.ts';
 import { ContainerManager } from '../../containers/container-manager.ts';
+import { UpstreamTimeoutError } from '../../containers/base-container.ts';
 import { API_SERVER } from '../../constants.ts';
 import { logger } from '../../logger.ts';
 import { ModelRegistry } from '../../models/registry.ts';
@@ -86,6 +87,11 @@ export class ChatHandler {
 
       await this.proxyChatRequest(req, res, ensured.location.endpoint, requestBody);
     } catch (error) {
+      if (error instanceof UpstreamTimeoutError) {
+        this.sendError(res, 504, error.message, 'upstream_timeout');
+        return;
+      }
+
       const message = error instanceof Error ? error.message : String(error);
       logger.error(`Chat completion error: ${message}`);
       this.sendError(res, 500, `Chat completion failed: ${message}`, 'server_error');
@@ -166,6 +172,11 @@ export class ChatHandler {
       headers: this.buildForwardHeaders(req),
       body: JSON.stringify(body),
       signal: controller.signal,
+    }).catch((error) => {
+      if (error instanceof Error && error.name === 'AbortError') {
+        throw new UpstreamTimeoutError();
+      }
+      throw error;
     }).finally(() => clearTimeout(timeout));
 
     if (body.stream) {

ts/api/handlers/embeddings.ts

@@ -11,6 +11,7 @@ import type {
 } from '../../interfaces/api.ts';
 import { ClusterCoordinator } from '../../cluster/coordinator.ts';
 import { ContainerManager } from '../../containers/container-manager.ts';
+import { UpstreamTimeoutError } from '../../containers/base-container.ts';
 import { API_SERVER } from '../../constants.ts';
 import { logger } from '../../logger.ts';
 import { ModelRegistry } from '../../models/registry.ts';
@@ -93,6 +94,11 @@ export class EmbeddingsHandler {
       });
       res.end(text);
     } catch (error) {
+      if (error instanceof UpstreamTimeoutError) {
+        this.sendError(res, 504, error.message, 'upstream_timeout');
+        return;
+      }
+
       const message = error instanceof Error ? error.message : String(error);
       logger.error(`Embeddings error: ${message}`);
       this.sendError(res, 500, `Embeddings generation failed: ${message}`, 'server_error');
@@ -224,6 +230,11 @@ export class EmbeddingsHandler {
         ...init,
         signal: controller.signal,
       });
+    } catch (error) {
+      if (error instanceof Error && error.name === 'AbortError') {
+        throw new UpstreamTimeoutError();
+      }
+      throw error;
     } finally {
       clearTimeout(timeout);
     }

ts/containers/base-container.ts

@@ -24,6 +24,13 @@ export type TModelPullProgress = (progress: {
   percent?: number;
 }) => void;
 
+export class UpstreamTimeoutError extends Error {
+  constructor(message: string = 'Upstream request timed out') {
+    super(message);
+    this.name = 'UpstreamTimeoutError';
+  }
+}
+
 /**
  * Abstract base class for AI model containers
  */
@@ -181,6 +188,11 @@ export abstract class BaseContainer {
       });
 
       return response;
+    } catch (error) {
+      if (error instanceof Error && error.name === 'AbortError') {
+        throw new UpstreamTimeoutError();
+      }
+      throw error;
     } finally {
       clearTimeout(timeoutId);
     }