feat(all): 重炼 crates/common 核心组件、上线 Web 运维看板与 Docker 容器化部署

2026-07-28 10:31:57 +08:00
commit 4b4238d702
71 changed files with 163402 additions and 0 deletions
@@ -0,0 +1,149 @@
+use super::AppState;
+use axum::{
+    extract::{Multipart, State},
+    response::IntoResponse,
+    Json,
+};
+use common::models::{GridPointParams, ModelSummary, TaskReport, TaskStatus};
+use serde_json::json;
+use std::path::Path;
+use tokio::fs;
+use tracing::{info, warn};
+
+use axum::http::StatusCode;
+
+pub async fn claim_task(State(state): State<AppState>) -> impl IntoResponse {
+    match state.queue.pop_task().await {
+        Ok(Some(task)) => {
+            if let Err(e) = state.db.mark_grid_point_running(&task.point_name).await {
+                warn!("领用任务 {} 后同步变更为 running 状态遇到异常: {}. 后置 stale 定时自取检索引索将介入修复维护", task.task_id, e);
+            }
+            (StatusCode::OK, Json(json!({"status": "ok", "task": task}))).into_response()
+        }
+        Ok(None) => (StatusCode::OK, Json(json!({"status": "empty", "task": null}))).into_response(),
+        Err(e) => (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            Json(json!({"status": "error", "message": format!("领用任务失败: {}", e)})),
+        )
+            .into_response(),
+    }
+}
+
+pub async fn report_task(
+    State(state): State<AppState>,
+    mut multipart: Multipart,
+) -> impl IntoResponse {
+    let mut report_json: Option<TaskReport> = None;
+    let mut seed_file_data: Option<Vec<u8>> = None;
+
+    while let Ok(Some(field)) = multipart.next_field().await {
+        let field_name = field.name().unwrap_or("").to_string();
+        if field_name == "report" {
+            if let Ok(bytes) = field.bytes().await {
+                if let Ok(report) = serde_json::from_slice::<TaskReport>(&bytes) {
+                    report_json = Some(report);
+                }
+            }
+        } else if field_name == "seed_file" {
+            if let Ok(bytes) = field.bytes().await {
+                seed_file_data = Some(bytes.to_vec());
+            }
+        }
+    }
+
+    let report = match report_json {
+        Some(r) => r,
+        None => {
+            return (
+                StatusCode::BAD_REQUEST,
+                Json(json!({"status": "error", "message": "请求中缺少 report 字段"})),
+            )
+                .into_response();
+        }
+    };
+
+    let name = report.point_name.clone();
+
+    if name.is_empty() || name.starts_with('.') || !name.chars().all(|c| c.is_ascii_alphanumeric() || c == '.' || c == '_' || c == '-' || c == '+' || c == '@') {
+        warn!("拒绝可能包含路径穿越或特殊非常规编码号攻击的网格点名称请求: {}", name);
+        return (
+            StatusCode::BAD_REQUEST,
+            Json(json!({"status": "error", "message": "非法的网格点名称参数"})),
+        )
+            .into_response();
+    }
+
+    let params = match extract_params(&report) {
+        Some(p) => p,
+        None => {
+            warn!("网格点 {} 汇报数据解析失败: 无法解析 params 或 summary_json", name);
+            return (
+                StatusCode::BAD_REQUEST,
+                Json(json!({"status": "error", "message": "无法解析 params 或 summary_json"})),
+            )
+                .into_response();
+        }
+    };
+
+    // Record in DB
+    if let Err(e) = state.db.record_task_report(&report).await {
+        warn!("记录网格点 {} 任务结果到数据库失败: {}", name, e);
+        return (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            Json(json!({"status": "error", "message": format!("记录数据库失败: {}", e)})),
+        )
+            .into_response();
+    }
+
+    // Clean up task from task_queue table to prevent queue DB bloat
+    if let Err(e) = state.queue.remove_task(&report.task_id.to_string()).await {
+        tracing::warn!("从任务队列中清理已上报任务记录 {} 失败: {}", report.task_id, e);
+    }
+
+    // 采用原子写入模式保持 conv.json 与核心二进制数据完整落地后才揭晓真实文件名
+    let model_dir = Path::new(&state.results_dir).join(&name);
+    if fs::create_dir_all(&model_dir).await.is_ok() {
+        let conv_tmp = model_dir.join(format!("conv.json.{}.tmp", uuid::Uuid::new_v4().simple()));
+        let conv_path = model_dir.join("conv.json");
+        if fs::write(&conv_tmp, &report.summary_json).await.is_ok() {
+            let _ = fs::rename(&conv_tmp, &conv_path).await;
+        }
+
+        // Save seed file .7 using atomic temporary writing strategy
+        if report.converged && !report.atmosphere_has_nan {
+            if let Some(bytes) = seed_file_data {
+                let seed_tmp = model_dir.join(format!("{}.7.{}.tmp", name, uuid::Uuid::new_v4().simple()));
+                let seed_path = model_dir.join(format!("{}.7", name));
+                if fs::write(&seed_tmp, bytes).await.is_ok() {
+                    if fs::rename(&seed_tmp, &seed_path).await.is_ok() {
+                        info!("成功保持原子写入落地并保存网格点 {} 的收敛种子文件: {}", name, seed_path.display());
+                        let _ = state
+                            .db
+                            .insert_seed(&params, &seed_path.to_string_lossy())
+                            .await;
+                    }
+                }
+            }
+        }
+    }
+
+    if report.status == TaskStatus::Failed || report.status == TaskStatus::Timeout || report.atmosphere_has_nan {
+        // Task did not succeed -> check if seed_step fallback should be triggered
+        info!("网格点 {} 计算未成功完成，检查种子回退机制...", name);
+        if let Err(e) = state.scheduler.trigger_seed_step_fallback(&params).await {
+            warn!("网格点 {} 触发种子回退机制失败: {}", name, e);
+        }
+    }
+
+    (StatusCode::OK, Json(json!({"status": "ok", "message": "上报成功"}))).into_response()
+}
+
+fn extract_params(report: &TaskReport) -> Option<GridPointParams> {
+    if let Some(ref p) = report.params {
+        return Some(p.clone());
+    }
+    serde_json::from_str::<ModelSummary>(&report.summary_json)
+        .ok()
+        .map(|summary| summary.params)
+}
+