feat: 更新后端平台模块、前端组件及构建产物，新增工作计划文档

- 更新 backend 平台 API endpoints 及 platform_store - 更新前端 ComputeNodesView、DataProcessCreateView、FineTuneCreateView 等组件 - 更新前端 API 模块（compute、fineTune） - 重构 frontend/dist 构建产物（新 hash） - 新增 docs/2026-07-24-work-plan.md 工作计划文档 Co-Authored-By: Claude <noreply@anthropic.com>
2026-07-24 10:27:52 +08:00
parent b28cfbc6fa
commit 2b10c013ce
151 changed files with 925 additions and 181 deletions
--- a/backend/app/api/v1/endpoints/platform.py
+++ b/backend/app/api/v1/endpoints/platform.py
@@ -5,7 +5,7 @@ import json
 from pathlib import Path
 from typing import Any

-from fastapi import APIRouter, Body, File, HTTPException, Query, UploadFile
+from fastapi import APIRouter, BackgroundTasks, Body, File, HTTPException, Query, UploadFile
 from fastapi.responses import PlainTextResponse

 from app.core.config import get_settings
@@ -41,6 +41,63 @@ def _node_for_compute_job_record(job_id: str) -> dict[str, Any] | None:
    return next((node for node in store.compute_nodes() if node["id"] == record.get("node_id")), None)


+def _training_diagnostics(errors: list[str], warnings: list[str] | None = None, log_text: str = "") -> list[dict[str, str]]:
+    source_items = [*errors, *(warnings or [])]
+    if log_text:
+        source_items.append(log_text)
+    text = "\n".join(source_items).lower()
+    diagnostics: list[dict[str, str]] = []
+    rules = [
+        (
+            ["dataset columns missing", "keyerror", "history", "instruction", "input", "output", "messages"],
+            "训练数据字段不匹配",
+            "请检查所选数据集格式是否与训练模板一致。Alpaca 格式通常需要 instruction/input/output；ShareGPT 格式通常需要 messages。",
+        ),
+        (
+            ["dataset file not found", "dataset_dir", "no uploaded file"],
+            "训练数据文件不可用",
+            "请确认数据集已上传文件，并且应用服务可以将数据同步到目标算力节点的数据目录。",
+        ),
+        (
+            ["model_name_or_path path not available", "base_model", "model path", "no such file"],
+            "基座模型路径不可用",
+            "请在模型管理中检查本地模型路径，确保该路径在算力服务器或 Compute 容器挂载目录内真实存在。",
+        ),
+        (
+            ["cuda out of memory", "outofmemoryerror", "显存", "memory"],
+            "GPU 显存不足",
+            "请降低 batch_size、cutoff_len、LoRA rank，启用 4bit 量化，或选择更高显存的算力节点。",
+        ),
+        (
+            ["training command not found", "llamafactory-cli"],
+            "训练框架命令不可用",
+            "请检查 Compute 镜像是否包含 LLaMA-Factory，或确认 llamafactory-cli 已在容器 PATH 中。",
+        ),
+        (
+            ["llama_factory_home not found"],
+            "LLaMA-Factory 目录不可用",
+            "请检查 Compute 服务的 LLAMA_FACTORY_HOME 配置和宿主机挂载路径。",
+        ),
+        (
+            ["no available compute node", "not schedulable", "disabled", "capacity full"],
+            "暂无可调度算力节点",
+            "请检查算力节点是否启用、状态是否在线、并行任务数是否已满，或手动调整节点权重/标签。",
+        ),
+    ]
+    for keywords, title, suggestion in rules:
+        if any(keyword in text for keyword in keywords):
+            diagnostics.append({"level": "error", "title": title, "suggestion": suggestion})
+    if not diagnostics and (errors or log_text):
+        diagnostics.append(
+            {
+                "level": "error",
+                "title": "训练任务异常",
+                "suggestion": "请查看预检错误和训练日志原文，优先确认模型路径、数据集格式、GPU 显存和 LLaMA-Factory 参数。",
+            }
+        )
+    return diagnostics
+
+
 async def _submit_fine_tune_task(store: Any, payload: dict[str, Any]) -> dict[str, Any]:
    task_id = str(payload.get("task_id") or payload.get("id") or "")
    if task_id and get_settings().compute_mode != "simulator":
@@ -68,6 +125,25 @@ async def _fine_tune_preflight(
    sync_resources: bool = False,
 ) -> dict[str, Any]:
    node, job_payload = store.prepare_compute_job_payload(task_id, payload or {})
+    return await _fine_tune_preflight_with_job_payload(node, job_payload, validate=validate, sync_resources=sync_resources, store=store)
+
+
+async def _fine_tune_preflight_payload(
+    store: Any,
+    payload: dict[str, Any],
+    validate: bool = True,
+) -> dict[str, Any]:
+    node, job_payload = store.prepare_compute_job_payload_from_payload(payload)
+    return await _fine_tune_preflight_with_job_payload(node, job_payload, validate=validate, sync_resources=False, store=store)
+
+
+async def _fine_tune_preflight_with_job_payload(
+    node: dict[str, Any],
+    job_payload: dict[str, Any],
+    validate: bool,
+    sync_resources: bool,
+    store: Any,
+) -> dict[str, Any]:
    sync_results: list[dict[str, Any]] = []
    sync_errors: list[str] = []
    if sync_resources and get_settings().compute_mode != "simulator":
@@ -105,6 +181,7 @@ async def _fine_tune_preflight(
        "valid": bool(preview.get("valid", not errors)) and not errors,
        "errors": errors,
        "warnings": warnings,
+        "diagnostics": _training_diagnostics(errors, warnings),
        "node": {
            "id": node.get("id"),
            "code": node.get("code"),
@@ -618,6 +695,26 @@ async def start_fine_tune(payload: dict[str, Any] = Body(...)) -> dict[str, Any]
        raise fail(502, f"submit compute job failed: {exc}")


+@router.post("/fine-tune/preflight")
+async def fine_tune_create_preflight(payload: dict[str, Any] = Body(...)) -> dict[str, Any]:
+    try:
+        return ok(await _fine_tune_preflight_payload(get_platform_store(), payload, validate=True))
+    except RuntimeError as exc:
+        return ok({"valid": False, "errors": [str(exc)], "warnings": [], "diagnostics": _training_diagnostics([str(exc)])})
+    except Exception as exc:  # noqa: BLE001 - expose compute validation errors to training create page
+        raise fail(502, f"compute preflight failed: {exc}")
+
+
+@router.post("/fine-tune/command-preview")
+async def fine_tune_create_command_preview(payload: dict[str, Any] = Body(...)) -> dict[str, Any]:
+    try:
+        return ok(await _fine_tune_preflight_payload(get_platform_store(), payload, validate=False))
+    except RuntimeError as exc:
+        return ok({"valid": False, "errors": [str(exc)], "warnings": [], "diagnostics": _training_diagnostics([str(exc)])})
+    except Exception as exc:  # noqa: BLE001
+        raise fail(502, f"compute command preview failed: {exc}")
+
+
@router.post("/fine-tune/{task_id}/preflight")
 async def fine_tune_preflight(task_id: str, payload: dict[str, Any] | None = Body(default=None)) -> dict[str, Any]:
    try:
@@ -685,6 +782,32 @@ async def fine_tune_logs(
    return ok({"job_id": task.get("compute_job_id") or "", "source": "task", "file": task.get("log_file") or "", "content": content, "size": f"{len(content.encode('utf-8'))} B"})


+@router.get("/fine-tune/{task_id}/diagnostics")
+async def fine_tune_diagnostics(task_id: str) -> dict[str, Any]:
+    store = get_platform_store()
+    try:
+        task = store.task(task_id)
+    except KeyError:
+        raise fail(404, "fine tune task not found")
+    log_text = ""
+    node = _node_for_task(task)
+    if node and task.get("compute_job_id"):
+        try:
+            logs = await ComputeNodeClient(node["api_base_url"]).job_logs(task["compute_job_id"], 1000, None, None)
+            log_text = str(logs.get("content") or "")
+        except Exception:
+            log_text = ""
+    errors = [str(task.get("failure_reason") or "")] if task.get("failure_reason") else []
+    return ok(
+        {
+            "task_id": task_id,
+            "status": task.get("status"),
+            "failure_reason": task.get("failure_reason") or "",
+            "diagnostics": _training_diagnostics(errors, [], log_text),
+        }
+    )
+
+
@router.put("/fine-tune/{task_id}")
 async def update_fine_tune(task_id: str, payload: dict[str, Any] = Body(...)) -> dict[str, Any]:
    try:
@@ -1055,6 +1178,14 @@ async def compute_node_replicas(node_id: str) -> dict[str, Any]:
    return ok(get_platform_store().replicas(node_id))


+@router.get("/compute/sync-jobs/{sync_id}")
+async def compute_sync_job_detail(sync_id: str) -> dict[str, Any]:
+    try:
+        return ok(get_platform_store().sync_job(sync_id))
+    except KeyError:
+        raise fail(404, "sync job not found")
+
+
@router.get("/compute/nodes/{node_id}/replicas/drift")
 async def compute_node_replica_drift(node_id: str) -> dict[str, Any]:
    store = get_platform_store()
@@ -1091,35 +1222,18 @@ async def compute_node_replica_drift(node_id: str) -> dict[str, Any]:
    return ok({"node_id": node_id, "items": items, "drifted": len([item for item in items if item.get("sync_status") == "drifted"])})


-@router.post("/compute/nodes/{node_id}/replicas/repair")
-async def compute_node_replica_repair(node_id: str, payload: dict[str, Any] | None = Body(default=None)) -> dict[str, Any]:
+async def _run_resource_replica_repair(sync_id: str, node_id: str, payload: dict[str, Any], replicas_to_repair: list[dict[str, Any]]) -> None:
    store = get_platform_store()
+    store.update_sync_job(sync_id, "running", 5)
    node = next((item for item in store.compute_nodes() if item["id"] == node_id), None)
    if not node:
-        raise fail(404, "compute node not found")
-    payload = payload or {}
-    replica_ids = payload.get("replica_ids") or [
-        item["id"] for item in store.replicas(node_id) if item.get("sync_status") in {"drifted", "failed", "repair_pending"}
-    ]
-    updated = store.mark_resource_replica_repair_pending([str(item) for item in replica_ids])
-    sync_id = store.create_sync_job(
-        node_id,
-        {
-            "resources": [
-                {
-                    "resource_type": item.get("resource_type"),
-                    "resource_id": item.get("resource_id"),
-                    "replica_id": item.get("id"),
-                    "target_path": item.get("local_path"),
-                }
-                for item in updated
-            ]
-        },
-    )
+        store.update_sync_job(sync_id, "failed", 100, completed=True)
+        return
    client = ComputeNodeClient(node["api_base_url"])
    repaired = []
    failures = []
-    for replica in updated:
+    total = max(len(replicas_to_repair), 1)
+    for index, replica in enumerate(replicas_to_repair, start=1):
        replica_id = str(replica["id"])
        resource_type = str(replica.get("resource_type") or "")
        resource_id = str(replica.get("resource_id") or "")
@@ -1187,8 +1301,46 @@ async def compute_node_replica_repair(node_id: str, payload: dict[str, Any] | No
            error = str(exc)
            failures.append({"replica_id": replica_id, "resource_type": resource_type, "resource_id": resource_id, "error": error})
            repaired.append(store.update_resource_replica_sync_result(replica_id, False, None, int(replica.get("byte_size") or 0), "", error))
+        progress = min(95, 5 + int(index / total * 90))
+        store.update_sync_job(sync_id, "running", progress)
    store.update_sync_job(sync_id, "failed" if failures else "completed", 100 if not failures else 99, completed=True)
-    return ok({"sync": store.sync_job(sync_id), "replicas": repaired, "failed": failures})
+
+
+@router.post("/compute/nodes/{node_id}/replicas/repair")
+async def compute_node_replica_repair(
+    node_id: str,
+    background_tasks: BackgroundTasks,
+    payload: dict[str, Any] | None = Body(default=None),
+) -> dict[str, Any]:
+    store = get_platform_store()
+    node = next((item for item in store.compute_nodes() if item["id"] == node_id), None)
+    if not node:
+        raise fail(404, "compute node not found")
+    payload = payload or {}
+    replica_ids = payload.get("replica_ids") or [
+        item["id"] for item in store.replicas(node_id) if item.get("sync_status") in {"drifted", "failed", "repair_pending"}
+    ]
+    updated = store.mark_resource_replica_repair_pending([str(item) for item in replica_ids])
+    sync_id = store.create_sync_job(
+        node_id,
+        {
+            "operation": "repair",
+            "resources": [
+                {
+                    "resource_type": item.get("resource_type"),
+                    "resource_id": item.get("resource_id"),
+                    "replica_id": item.get("id"),
+                    "target_path": item.get("local_path"),
+                }
+                for item in updated
+            ],
+        },
+    )
+    if not updated:
+        store.update_sync_job(sync_id, "completed", 100, completed=True)
+        return ok({"sync": store.sync_job(sync_id), "replicas": [], "failed": [], "async": False})
+    background_tasks.add_task(_run_resource_replica_repair, sync_id, node_id, payload, updated)
+    return ok({"sync": store.sync_job(sync_id), "replicas": updated, "failed": [], "async": True})


@router.get("/compute/nodes/{node_id}/engines")