feat(cluster,api,models,cli): add cluster-aware model catalog deployments and request routing

2026-04-20 23:00:50 +00:00
parent 83cacd0cf1
commit 4f2266e1b7
55 changed files with 3970 additions and 1630 deletions
@@ -0,0 +1,192 @@
+import * as fs from 'node:fs/promises';
+import { CLUSTER, PATHS } from '../constants.ts';
+import type { IModelGridConfig } from '../interfaces/config.ts';
+import { logger } from '../logger.ts';
+
+export class ClusterHandler {
+  public async status(): Promise<void> {
+    const response = await this.request('/_cluster/status');
+    if (!response) {
+      return;
+    }
+
+    logger.log(JSON.stringify(response, null, 2));
+  }
+
+  public async nodes(): Promise<void> {
+    const response = await this.request('/_cluster/nodes');
+    if (!response) {
+      return;
+    }
+
+    logger.log(JSON.stringify(response, null, 2));
+  }
+
+  public async models(): Promise<void> {
+    const response = await this.request('/_cluster/status');
+    if (!response || typeof response !== 'object' || !('models' in response)) {
+      return;
+    }
+
+    logger.log(JSON.stringify((response as { models: unknown }).models, null, 2));
+  }
+
+  public async desired(): Promise<void> {
+    const response = await this.request('/_cluster/desired');
+    if (!response) {
+      return;
+    }
+
+    logger.log(JSON.stringify(response, null, 2));
+  }
+
+  public async ensure(model: string): Promise<void> {
+    if (!model) {
+      logger.error('Model ID is required');
+      return;
+    }
+
+    const response = await this.request('/_cluster/models/ensure', {
+      method: 'POST',
+      body: { model },
+    });
+    if (!response) {
+      return;
+    }
+
+    logger.log(JSON.stringify(response, null, 2));
+  }
+
+  public async scale(model: string, desiredReplicas: number): Promise<void> {
+    if (!model || Number.isNaN(desiredReplicas)) {
+      logger.error('Model ID and desired replica count are required');
+      return;
+    }
+
+    const response = await this.request('/_cluster/models/desired', {
+      method: 'POST',
+      body: { model, desiredReplicas },
+    });
+    if (!response) {
+      return;
+    }
+
+    logger.log(JSON.stringify(response, null, 2));
+  }
+
+  public async clear(model: string): Promise<void> {
+    if (!model) {
+      logger.error('Model ID is required');
+      return;
+    }
+
+    const response = await this.request('/_cluster/models/desired/remove', {
+      method: 'POST',
+      body: { model },
+    });
+    if (!response) {
+      return;
+    }
+
+    logger.log(JSON.stringify(response, null, 2));
+  }
+
+  public async cordon(nodeName: string): Promise<void> {
+    await this.updateNodeState('/_cluster/nodes/cordon', nodeName);
+  }
+
+  public async uncordon(nodeName: string): Promise<void> {
+    await this.updateNodeState('/_cluster/nodes/uncordon', nodeName);
+  }
+
+  public async drain(nodeName: string): Promise<void> {
+    await this.updateNodeState('/_cluster/nodes/drain', nodeName);
+  }
+
+  public async activate(nodeName: string): Promise<void> {
+    await this.updateNodeState('/_cluster/nodes/activate', nodeName);
+  }
+
+  private async request(
+    path: string,
+    options: {
+      method?: 'GET' | 'POST';
+      body?: unknown;
+    } = {},
+  ): Promise<unknown | null> {
+    const config = await this.readConfig();
+    if (!config) {
+      return null;
+    }
+
+    const endpoint = this.resolveEndpoint(config);
+    const headers: Record<string, string> = {
+      'Content-Type': 'application/json',
+    };
+
+    if (config.cluster.sharedSecret) {
+      headers[CLUSTER.AUTH_HEADER_NAME] = config.cluster.sharedSecret;
+    }
+
+    try {
+      const response = await fetch(`${endpoint}${path}`, {
+        method: options.method || 'GET',
+        headers,
+        body: options.body ? JSON.stringify(options.body) : undefined,
+      });
+
+      if (!response.ok) {
+        logger.error(`Cluster request failed: ${response.status} ${await response.text()}`);
+        return null;
+      }
+
+      return await response.json();
+    } catch (error) {
+      logger.error(
+        `Cluster request failed: ${error instanceof Error ? error.message : String(error)}`,
+      );
+      return null;
+    }
+  }
+
+  private async readConfig(): Promise<IModelGridConfig | null> {
+    try {
+      return JSON.parse(await fs.readFile(PATHS.CONFIG_FILE, 'utf-8')) as IModelGridConfig;
+    } catch (error) {
+      logger.error(
+        `Failed to read config: ${error instanceof Error ? error.message : String(error)}`,
+      );
+      return null;
+    }
+  }
+
+  private resolveEndpoint(config: IModelGridConfig): string {
+    if (config.cluster.controlPlaneUrl) {
+      return config.cluster.controlPlaneUrl;
+    }
+
+    if (config.cluster.advertiseUrl) {
+      return config.cluster.advertiseUrl;
+    }
+
+    const host = config.api.host === '0.0.0.0' ? '127.0.0.1' : config.api.host;
+    return `http://${host}:${config.api.port}`;
+  }
+
+  private async updateNodeState(path: string, nodeName: string): Promise<void> {
+    if (!nodeName) {
+      logger.error('Node name is required');
+      return;
+    }
+
+    const response = await this.request(path, {
+      method: 'POST',
+      body: { nodeName },
+    });
+    if (!response) {
+      return;
+    }
+
+    logger.log(JSON.stringify(response, null, 2));
+  }
+}