🐛 Bugfix: Adapt frontend for k8s A2A port configuration

[xuyaqist]

#3267

[JasonW404]

内存限制从 4Gi 直接提升到 64Gi（16 倍），这个幅度异常大。如果是为了应对特定场景的内存峰值，建议添加注释说明原因。如果是内存泄漏导致的 OOM，应该排查根因而不是单纯提高限制。

[YehongPan]

🔍 Code Review Comments

1. [代码规范] 默认端口处理
应确认前端是否正确处理了 window.location.port 为空（默认端口 80/443）的情况，避免拼接出 host: 这样的无效 URL。

[YehongPan]

Code Review

• [代码规范] 应确认前端是否正确处理了 window.location.port 为空（默认端口 80/443）的情况，避免拼接出 host: 这样的无效 URL。

[WMC001]

A2A over k8s is a solid architectural addition. The k8s-native approach (in-cluster service discovery, sidecar proxy) is well-chosen. Please ensure the fallback logic when the k8s environment is unavailable is tested.

[WMC001]

Bug 1: wrong substring truncation in pod name test expectations

test/sdk/container/test_k8s_client.py lines 751-752 and 761-762 — the assertions for "tenant_123" and "tenant.123" assume 8-char truncation yields "tenant-12" and "tenant.1", but after _sanitize_k8s_name, both become "tenant-123" (9 chars), which truncates to "tenant-1" (8 chars). The tests will fail:

# Line 751-752:
assert name == "mcp-test-service-tenant-1-user-123-a1b2c3d4"  # tenant-1, not tenant-12

# Line 761-762:
assert name == "mcp-test-service-tenant-1-user-123-a1b2c3d4"  # tenant-1, not tenant.1

Same issue with "tenant 123" at line 772 — should be "tenant-1" not "tenant 1".

[WMC001]

Bug 2: minio_files is unconditionally set to None in assistant save

backend/services/conversation_management_service.py line 23 — the change from minio_files=request.minio_files to minio_files=None silently drops all file attachments from assistant messages. If assistants ever produce file attachments (e.g., generated images, exported artifacts), they will be lost. This seems like an intentional change, but confirm this is the desired behavior — otherwise it should be minio_files=request.minio_files or None.