feat(admin-perf): full Vue SPA migration + slow-query/memory monitoring gaps
Remove Jinja2 template fallback (1249 lines) — /admin/performance now serves Vue SPA exclusively via send_from_directory. Backend: - Add _SLOW_QUERY_WAITING counter with get_slow_query_waiting_count() - Record slow-path latency in read_sql_df_slow/iter via record_query_latency() - Extend metrics_history schema with slow_query_active, slow_query_waiting, worker_rss_bytes columns + ALTER TABLE migration for existing DBs - Add cleanup_archive_logs() with configurable ARCHIVE_LOG_DIR/KEEP_COUNT - Integrate archive cleanup into MetricsHistoryCollector 50-min cycle Frontend: - Add slow_query_active and slow_query_waiting StatCards to connection pool - Add slow_query_active trend line to pool trend chart - Add Worker memory (RSS MB) trend chart with preprocessing - Update modernization gate check path to frontend style.css Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
egg
@@ -160,6 +160,8 @@
|
||||
<StatCard :value="poolTotalConnections" label="總連線數" />
|
||||
<StatCard :value="perfDetail.db_pool.status.max_capacity" label="最大容量" />
|
||||
<StatCard :value="poolOverflowDisplay" label="溢出連線" />
|
||||
<StatCard :value="perfDetail.db_pool.status.slow_query_active" label="慢查詢執行中" />
|
||||
<StatCard :value="perfDetail.db_pool.status.slow_query_waiting" label="慢查詢排隊中" />
|
||||
<StatCard :value="perfDetail.db_pool.config?.pool_size" label="池大小" />
|
||||
<StatCard :value="perfDetail.db_pool.config?.pool_recycle" label="回收週期 (s)" />
|
||||
<StatCard :value="perfDetail.db_pool.config?.pool_timeout" label="逾時 (s)" />
|
||||
@@ -175,6 +177,15 @@
|
||||
:series="poolTrendSeries"
|
||||
/>
|
||||
|
||||
<!-- Worker Memory Trend -->
|
||||
<TrendChart
|
||||
v-if="historyData.length > 1"
|
||||
title="Worker 記憶體趨勢"
|
||||
:snapshots="historyData"
|
||||
:series="memoryTrendSeries"
|
||||
yAxisLabel="MB"
|
||||
/>
|
||||
|
||||
<!-- Worker Control -->
|
||||
<section class="panel">
|
||||
<h2 class="panel-title">Worker 控制</h2>
|
||||
@@ -450,7 +461,12 @@ async function loadWorkerStatus() {
|
||||
async function loadPerformanceHistory() {
|
||||
try {
|
||||
const res = await apiGet('/admin/api/performance-history', { params: { minutes: 30 } });
|
||||
historyData.value = res?.data?.snapshots || [];
|
||||
const snapshots = res?.data?.snapshots || [];
|
||||
// Pre-process: convert worker_rss_bytes to worker_rss_mb for trend chart
|
||||
historyData.value = snapshots.map((s) => ({
|
||||
...s,
|
||||
worker_rss_mb: s.worker_rss_bytes ? Math.round(s.worker_rss_bytes / 1048576 * 10) / 10 : 0,
|
||||
}));
|
||||
} catch (e) {
|
||||
console.error('Failed to load performance history:', e);
|
||||
}
|
||||
@@ -460,6 +476,7 @@ async function loadPerformanceHistory() {
|
||||
const poolTrendSeries = [
|
||||
{ name: '飽和度', key: 'pool_saturation', color: '#6366f1' },
|
||||
{ name: '使用中', key: 'pool_checked_out', color: '#f59e0b' },
|
||||
{ name: '慢查詢執行中', key: 'slow_query_active', color: '#ef4444' },
|
||||
];
|
||||
|
||||
const latencyTrendSeries = [
|
||||
@@ -478,6 +495,10 @@ const hitRateTrendSeries = [
|
||||
{ name: 'L2 命中率', key: 'rc_l2_hit_rate', color: '#f59e0b' },
|
||||
];
|
||||
|
||||
const memoryTrendSeries = [
|
||||
{ name: 'RSS (MB)', key: 'worker_rss_mb', color: '#8b5cf6' },
|
||||
];
|
||||
|
||||
async function refreshAll() {
|
||||
loading.value = true;
|
||||
try {
|
||||
|
||||
@@ -0,0 +1,2 @@
|
||||
schema: spec-driven
|
||||
created: 2026-02-26
|
||||
@@ -0,0 +1,81 @@
|
||||
## Context
|
||||
|
||||
2026-02-25 的 server crash 暴露出 pool 隔離架構變更後的監控盲區。event_fetcher 和 lineage_engine 已遷移到 `read_sql_df_slow`（獨立連線 + semaphore），但 metrics_history 快照只記錄 pool 相關指標，slow query 並行數、排隊數、Worker RSS 完全無歷史紀錄。
|
||||
|
||||
同時 `/admin/performance` 仍保留 1249 行 Jinja template 作為 Vue SPA fallback，但 SPA 已是唯一使用的版本（build artifact 存在於 `static/dist/admin-performance.html`），兩套 UI 增加維護成本且 Jinja 版功能遠不及 SPA。
|
||||
|
||||
`logs/archive/` 目錄累積 rotated log 檔案無自動清理，是唯一會無限增長的儲存。
|
||||
|
||||
## Goals / Non-Goals
|
||||
|
||||
**Goals:**
|
||||
- 移除 Jinja fallback，統一為 Vue SPA 單一架構
|
||||
- 讓 slow query 並行數、排隊數、Worker RSS 成為可觀測的歷史趨勢指標
|
||||
- 讓 P50/P95/P99 反映所有查詢路徑（pool + slow path）
|
||||
- 解決 archive log 無限增長問題
|
||||
|
||||
**Non-Goals:**
|
||||
- 不修改 `/admin/pages`（仍為 Jinja template）
|
||||
- 不新增 async job queue 面板（P1，後續 change 處理）
|
||||
- 不新增 event cache hit/miss 計數器（P2）
|
||||
- 不增加即時告警或 webhook 通知機制
|
||||
|
||||
## Decisions
|
||||
|
||||
### D1：SQLite schema migration 策略
|
||||
|
||||
**選擇**：啟動時執行 `ALTER TABLE ADD COLUMN IF NOT EXISTS`（容錯 "duplicate column" error）
|
||||
|
||||
**替代方案**：version table + migration script → 過度工程，SQLite 只有 3 天保留，加欄是向後相容的
|
||||
|
||||
**理由**：新欄位 nullable，舊 row 自動為 NULL，不影響既有查詢。MetricsHistoryStore.initialize() 已在啟動時執行 CREATE TABLE IF NOT EXISTS，加入 ALTER TABLE 語句自然整合。
|
||||
|
||||
### D2：RSS 記憶體取得方式
|
||||
|
||||
**選擇**：`resource.getrusage(resource.RUSAGE_SELF).ru_maxrss * 1024`（Python stdlib，Linux 上單位為 KB）
|
||||
|
||||
**替代方案 A**：讀取 `/proc/self/status` VmRSS → 平台相依，解析 overhead
|
||||
**替代方案 B**：`psutil.Process().memory_info().rss` → 需新增外部依賴
|
||||
|
||||
**理由**：`resource` 模組為 Python 標準庫，無需額外依賴。`ru_maxrss` 在 Linux 上返回 KB，乘以 1024 轉為 bytes。
|
||||
|
||||
### D3：Semaphore 排隊計數器實作
|
||||
|
||||
**選擇**：在 `read_sql_df_slow()` 的 semaphore.acquire() 前後遞增/遞減 `_SLOW_QUERY_WAITING` atomic counter
|
||||
|
||||
**流程**：
|
||||
```
|
||||
_SLOW_QUERY_WAITING += 1
|
||||
acquired = semaphore.acquire(timeout=60)
|
||||
_SLOW_QUERY_WAITING -= 1
|
||||
if not acquired: raise RuntimeError
|
||||
_SLOW_QUERY_ACTIVE += 1
|
||||
... execute query ...
|
||||
_SLOW_QUERY_ACTIVE -= 1
|
||||
```
|
||||
|
||||
**理由**：與既有 `_SLOW_QUERY_ACTIVE` 模式一致，使用 threading.Lock 保護。
|
||||
|
||||
### D4：Archive log cleanup 整合位置
|
||||
|
||||
**選擇**：整合到 `MetricsHistoryCollector._run()` 的 cleanup cycle（每 ~100 intervals ≈ 50 分鐘）
|
||||
|
||||
**替代方案**：獨立 cron job → 需額外 crontab 配置，不自包含
|
||||
|
||||
**理由**：已有 daemon thread 定期 cleanup SQLite，加入 archive cleanup 邏輯一致且自包含。
|
||||
|
||||
### D5：移除 Jinja fallback 的安全性
|
||||
|
||||
**選擇**：直接移除 fallback，admin_routes.py 改為只 `send_from_directory(dist_dir, "admin-performance.html")`
|
||||
|
||||
**理由**：
|
||||
- Vue SPA build artifact 已存在（`static/dist/admin-performance.html`，2026-02-26 更新）
|
||||
- `frontend/package.json` build script 已包含 admin-performance entry
|
||||
- CI/deploy 流程必包含 `npx vite build`
|
||||
- 若 build 失敗，`/health/frontend-shell` 已有 asset readiness 檢查可偵測
|
||||
|
||||
## Risks / Trade-offs
|
||||
|
||||
- **[Risk] Build 失敗時 /admin/performance 返回 404** → 既有 `/health/frontend-shell` 檢查 + deploy script 驗證。移除 fallback 反而讓問題更早暴露。
|
||||
- **[Risk] ALTER TABLE 在 SQLite 大表上可能慢** → metrics_history 最多 50K rows，ALTER TABLE 即時完成。
|
||||
- **[Trade-off] `ru_maxrss` 是 peak RSS，非 current RSS** → 在 Linux 上 `ru_maxrss` 是 process lifetime 的 max RSS。改用 `/proc/self/status` 的 VmRSS 可取得 current，但需 file I/O。鑑於每 30 秒收集一次且 max RSS 更能反映記憶體壓力，接受此 trade-off。若日後需要 current RSS，可改讀 `/proc/self/status`。
|
||||
@@ -0,0 +1,34 @@
|
||||
## Why
|
||||
|
||||
2026-02-25 server crash 暴露出管理員效能監控頁面在 pool 隔離架構變更後的關鍵盲區：slow query 並行數、slow-path 延遲、Worker 記憶體等核心指標既未收集也未顯示，導致 crash 前完全無法觀測系統真實負載。同時，`/admin/performance` 仍保留 1249 行的 Jinja template 作為 fallback，與已完成的 Vue SPA 遷移架構不一致，增加維護成本。
|
||||
|
||||
## What Changes
|
||||
|
||||
- **移除** Jinja template `templates/admin/performance.html`，`/admin/performance` 路由直接服務 Vue SPA（`static/dist/admin-performance.html`），不再有 fallback 邏輯
|
||||
- **新增** `slow_query_active`、`slow_query_waiting`、`worker_rss_bytes` 三個欄位到 `metrics_history.sqlite` 快照，含 SQLite schema migration
|
||||
- **新增** semaphore 排隊計數器（`_SLOW_QUERY_WAITING`），追蹤等待 slow query semaphore 的 thread 數量
|
||||
- **修正** `read_sql_df_slow()` 和 `read_sql_df_slow_iter()` 將查詢延遲記錄到 `QueryMetrics`，使 P50/P95/P99 反映所有查詢路徑
|
||||
- **新增** Vue SPA 連線池區塊顯示「慢查詢執行中」「慢查詢排隊中」指標 + 連線池趨勢圖加入 slow_query_active 線 + Worker 記憶體趨勢圖
|
||||
- **新增** archive log 自動清理機制，整合到既有 `MetricsHistoryCollector` 的 cleanup cycle
|
||||
|
||||
## Capabilities
|
||||
|
||||
### New Capabilities
|
||||
|
||||
- `slow-query-observability`: 追蹤 slow query 並行數、排隊數、延遲，寫入 metrics history 並在前端顯示趨勢
|
||||
- `worker-memory-tracking`: 追蹤 Worker RSS 記憶體，寫入 metrics history 並在前端顯示趨勢
|
||||
- `archive-log-rotation`: logs/archive/ 目錄的自動清理機制，防止檔案無限增長
|
||||
|
||||
### Modified Capabilities
|
||||
|
||||
- `admin-performance-spa`: 移除 Jinja template fallback，完全遷移至 Vue SPA，新增 slow query 與記憶體監控面板
|
||||
- `metrics-history-trending`: 擴充 snapshot schema 加入 slow_query_active、slow_query_waiting、worker_rss_bytes
|
||||
- `connection-pool-monitoring`: 新增 semaphore 排隊計數器，slow-path 延遲納入 QueryMetrics
|
||||
|
||||
## Impact
|
||||
|
||||
- **後端**：`core/database.py`（排隊計數器 + latency 記錄）、`core/metrics_history.py`（schema 擴充 + archive cleanup）、`routes/admin_routes.py`（移除 fallback）
|
||||
- **前端**：`frontend/src/admin-performance/App.vue`（新面板 + 趨勢圖）→ 需 rebuild
|
||||
- **刪除**：`templates/admin/performance.html`（1249 行）
|
||||
- **資料**：既有 `metrics_history.sqlite` 需 ALTER TABLE 加欄（向後相容，新欄位 nullable）
|
||||
- **測試**：既有 `test_performance_integration.py` 已測試 SPA 路徑，無需修改；需新增 schema migration 測試
|
||||
@@ -0,0 +1,37 @@
|
||||
## MODIFIED Requirements
|
||||
|
||||
### Requirement: Vue 3 SPA page replaces Jinja2 template
|
||||
The `/admin/performance` route SHALL serve the Vite-built `admin-performance.html` static file directly. The Jinja2 template fallback SHALL be removed. If the SPA build artifact does not exist, the server SHALL return a standard HTTP error (no fallback rendering).
|
||||
|
||||
#### Scenario: Page loads as Vue SPA
|
||||
- **WHEN** user navigates to `/admin/performance`
|
||||
- **THEN** the server SHALL return the Vite-built `admin-performance.html` static file via `send_from_directory`
|
||||
|
||||
#### Scenario: Portal-shell integration
|
||||
- **WHEN** the portal-shell renders `/admin/performance`
|
||||
- **THEN** it SHALL load the page as a native Vue SPA (not an external iframe)
|
||||
|
||||
#### Scenario: Build artifact missing
|
||||
- **WHEN** the SPA build artifact `admin-performance.html` does not exist in `static/dist/`
|
||||
- **THEN** the server SHALL return an HTTP error (no Jinja2 fallback)
|
||||
|
||||
### Requirement: Connection pool panel
|
||||
The dashboard SHALL display connection pool saturation as a GaugeBar and stat cards showing checked_out, checked_in, overflow, max_capacity, pool_size, pool_recycle, pool_timeout, direct connection count, slow_query_active, and slow_query_waiting.
|
||||
|
||||
#### Scenario: Pool under normal load
|
||||
- **WHEN** pool saturation is below 80%
|
||||
- **THEN** the GaugeBar SHALL display in a normal color (green/blue)
|
||||
|
||||
#### Scenario: Pool near saturation
|
||||
- **WHEN** pool saturation exceeds 80%
|
||||
- **THEN** the GaugeBar SHALL display in a warning color (yellow/orange/red)
|
||||
|
||||
#### Scenario: Slow query metrics displayed
|
||||
- **WHEN** `db_pool.status` includes `slow_query_active` and `slow_query_waiting`
|
||||
- **THEN** the panel SHALL display StatCards for both values
|
||||
|
||||
## REMOVED Requirements
|
||||
|
||||
### Requirement: Jinja2 template fallback for performance page
|
||||
**Reason**: The Vue SPA is the sole UI. Maintaining a 1249-line Jinja template as fallback adds maintenance burden and feature divergence.
|
||||
**Migration**: Delete `templates/admin/performance.html`. The route handler serves the SPA directly.
|
||||
@@ -0,0 +1,30 @@
|
||||
## ADDED Requirements
|
||||
|
||||
### Requirement: Automatic archive log cleanup
|
||||
The system SHALL provide a `cleanup_archive_logs()` function in `core/metrics_history.py` that deletes old rotated log files from `logs/archive/`, keeping the most recent N files per log type (access, error, watchdog, rq_worker, startup).
|
||||
|
||||
#### Scenario: Cleanup keeps recent files
|
||||
- **WHEN** `cleanup_archive_logs()` is called with `keep_per_type=20` and there are 30 access_*.log files
|
||||
- **THEN** 10 oldest access_*.log files SHALL be deleted, keeping the 20 most recent by modification time
|
||||
|
||||
#### Scenario: No excess files
|
||||
- **WHEN** `cleanup_archive_logs()` is called and each type has fewer than `keep_per_type` files
|
||||
- **THEN** no files SHALL be deleted
|
||||
|
||||
#### Scenario: Archive directory missing
|
||||
- **WHEN** `cleanup_archive_logs()` is called and the archive directory does not exist
|
||||
- **THEN** the function SHALL return 0 without error
|
||||
|
||||
### Requirement: Archive cleanup integrated into collector cycle
|
||||
The `MetricsHistoryCollector` SHALL call `cleanup_archive_logs()` alongside the existing SQLite cleanup, running approximately every 50 minutes (every 100 collection intervals).
|
||||
|
||||
#### Scenario: Periodic cleanup executes
|
||||
- **WHEN** the cleanup counter reaches 100 intervals
|
||||
- **THEN** both SQLite metrics cleanup and archive log cleanup SHALL execute
|
||||
|
||||
### Requirement: Archive cleanup configuration
|
||||
The archive log cleanup SHALL be configurable via environment variables: `ARCHIVE_LOG_DIR` (default: `logs/archive`) and `ARCHIVE_LOG_KEEP_COUNT` (default: 20).
|
||||
|
||||
#### Scenario: Custom keep count
|
||||
- **WHEN** `ARCHIVE_LOG_KEEP_COUNT=10` is set
|
||||
- **THEN** cleanup SHALL keep only the 10 most recent files per type
|
||||
@@ -0,0 +1,29 @@
|
||||
## MODIFIED Requirements
|
||||
|
||||
### Requirement: Connection pool status in performance detail
|
||||
The performance-detail API SHALL include `db_pool` section with `status` (checked_out, checked_in, overflow, max_capacity, saturation, slow_query_active, slow_query_waiting) from `get_pool_status()` and `config` (pool_size, max_overflow, pool_timeout, pool_recycle) from `get_pool_runtime_config()`.
|
||||
|
||||
#### Scenario: Pool status retrieved
|
||||
- **WHEN** the API is called
|
||||
- **THEN** `db_pool.status` SHALL contain current pool utilization metrics including `slow_query_active` and `slow_query_waiting`, and `db_pool.config` SHALL contain the pool configuration values
|
||||
|
||||
#### Scenario: Saturation calculation
|
||||
- **WHEN** the pool has 8 checked_out connections and max_capacity is 30
|
||||
- **THEN** saturation SHALL be reported as approximately 26.7%
|
||||
|
||||
#### Scenario: Slow query waiting included
|
||||
- **WHEN** 2 threads are waiting for the slow query semaphore
|
||||
- **THEN** `db_pool.status.slow_query_waiting` SHALL be 2
|
||||
|
||||
## ADDED Requirements
|
||||
|
||||
### Requirement: Slow-path query latency included in QueryMetrics
|
||||
The `read_sql_df_slow()` and `read_sql_df_slow_iter()` functions SHALL call `record_query_latency()` with the total elapsed time upon completion, ensuring P50/P95/P99 percentiles reflect queries from all paths (pooled and slow/direct).
|
||||
|
||||
#### Scenario: Slow query latency recorded
|
||||
- **WHEN** `read_sql_df_slow()` completes a query in 8.5 seconds
|
||||
- **THEN** `record_query_latency(8.5)` SHALL be called and the value SHALL appear in subsequent `get_percentiles()` results
|
||||
|
||||
#### Scenario: Slow iter latency recorded
|
||||
- **WHEN** `read_sql_df_slow_iter()` completes streaming in 45 seconds
|
||||
- **THEN** `record_query_latency(45.0)` SHALL be called in the finally block
|
||||
@@ -0,0 +1,54 @@
|
||||
## MODIFIED Requirements
|
||||
|
||||
### Requirement: SQLite metrics history store
|
||||
The system SHALL provide a `MetricsHistoryStore` class in `core/metrics_history.py` that persists metrics snapshots to a SQLite database (`logs/metrics_history.sqlite` by default). The store SHALL use thread-local connections and a write lock, following the `LogStore` pattern in `core/log_store.py`. The schema SHALL include columns for `slow_query_active` (INTEGER), `slow_query_waiting` (INTEGER), and `worker_rss_bytes` (INTEGER) in addition to the existing pool, Redis, route cache, and latency columns.
|
||||
|
||||
#### Scenario: Write and query snapshots
|
||||
- **WHEN** `write_snapshot(data)` is called with pool/redis/route_cache/latency/slow_query/memory metrics
|
||||
- **THEN** a row SHALL be inserted into `metrics_snapshots` with the current ISO 8601 timestamp, worker PID, and all metric columns
|
||||
|
||||
#### Scenario: Query by time range
|
||||
- **WHEN** `query_snapshots(minutes=30)` is called
|
||||
- **THEN** it SHALL return all rows from the last 30 minutes, ordered by timestamp ascending, including the new columns
|
||||
|
||||
#### Scenario: Retention cleanup
|
||||
- **WHEN** `cleanup()` is called
|
||||
- **THEN** rows older than `METRICS_HISTORY_RETENTION_DAYS` (default 3) SHALL be deleted, and total rows SHALL be capped at `METRICS_HISTORY_MAX_ROWS` (default 50000)
|
||||
|
||||
#### Scenario: Thread safety
|
||||
- **WHEN** multiple threads write snapshots concurrently
|
||||
- **THEN** the write lock SHALL serialize writes and prevent database corruption
|
||||
|
||||
#### Scenario: Schema migration for existing databases
|
||||
- **WHEN** the store initializes on an existing database without the new columns
|
||||
- **THEN** it SHALL execute ALTER TABLE ADD COLUMN for each missing column, tolerating "duplicate column" errors
|
||||
|
||||
### Requirement: Background metrics collector
|
||||
The system SHALL provide a `MetricsHistoryCollector` class that runs a daemon thread collecting metrics snapshots at a configurable interval (default 30 seconds, via `METRICS_HISTORY_INTERVAL` env var). The collector SHALL include `slow_query_active`, `slow_query_waiting`, and `worker_rss_bytes` in each snapshot.
|
||||
|
||||
#### Scenario: Automatic collection
|
||||
- **WHEN** the collector is started via `start_metrics_history(app)`
|
||||
- **THEN** it SHALL collect pool status (including slow_query_active and slow_query_waiting), Redis info, route cache status, query latency metrics, and worker RSS memory every interval and write them to the store
|
||||
|
||||
#### Scenario: Graceful shutdown
|
||||
- **WHEN** `stop_metrics_history()` is called
|
||||
- **THEN** the collector thread SHALL stop within one interval period
|
||||
|
||||
#### Scenario: Subsystem unavailability
|
||||
- **WHEN** a subsystem (e.g., Redis) is unavailable during collection
|
||||
- **THEN** the collector SHALL write null/0 for those fields and continue collecting other metrics
|
||||
|
||||
### Requirement: Frontend trend charts
|
||||
The system SHALL display 5 trend chart panels in the admin performance dashboard using vue-echarts VChart line/area charts: connection pool saturation, query latency (P50/P95/P99), Redis memory, cache hit rates, and worker memory.
|
||||
|
||||
#### Scenario: Trend charts with data
|
||||
- **WHEN** historical snapshots contain more than 1 data point
|
||||
- **THEN** the dashboard SHALL display trend charts for: connection pool saturation (including slow_query_active), query latency (P50/P95/P99), Redis memory, cache hit rates, and worker memory (RSS in MB)
|
||||
|
||||
#### Scenario: Trend charts without data
|
||||
- **WHEN** historical snapshots are empty or contain only 1 data point
|
||||
- **THEN** the trend charts SHALL NOT be displayed (hidden via `v-if`)
|
||||
|
||||
#### Scenario: Auto-refresh
|
||||
- **WHEN** the dashboard auto-refreshes
|
||||
- **THEN** historical data SHALL also be refreshed alongside real-time metrics
|
||||
@@ -0,0 +1,49 @@
|
||||
## ADDED Requirements
|
||||
|
||||
### Requirement: Slow query active count in metrics history snapshots
|
||||
The `MetricsHistoryCollector` SHALL include `slow_query_active` in each 30-second snapshot, recording the number of slow queries currently executing via dedicated connections.
|
||||
|
||||
#### Scenario: Snapshot includes slow_query_active
|
||||
- **WHEN** the collector writes a snapshot while 3 slow queries are executing
|
||||
- **THEN** the `slow_query_active` column SHALL contain the value 3
|
||||
|
||||
#### Scenario: No slow queries active
|
||||
- **WHEN** the collector writes a snapshot while no slow queries are executing
|
||||
- **THEN** the `slow_query_active` column SHALL contain the value 0
|
||||
|
||||
### Requirement: Slow query waiting count tracked and persisted
|
||||
The system SHALL maintain a thread-safe counter `_SLOW_QUERY_WAITING` in `database.py` that tracks the number of threads currently waiting to acquire the slow query semaphore. This counter SHALL be included in `get_pool_status()` and persisted to metrics history snapshots.
|
||||
|
||||
#### Scenario: Counter increments on semaphore wait
|
||||
- **WHEN** a thread enters `read_sql_df_slow()` and the semaphore is full
|
||||
- **THEN** `_SLOW_QUERY_WAITING` SHALL be incremented before `semaphore.acquire()` and decremented after acquire completes (success or timeout)
|
||||
|
||||
#### Scenario: Counter in pool status API
|
||||
- **WHEN** `get_pool_status()` is called
|
||||
- **THEN** the returned dict SHALL include `slow_query_waiting` with the current waiting thread count
|
||||
|
||||
#### Scenario: Counter persisted to metrics history
|
||||
- **WHEN** the collector writes a snapshot
|
||||
- **THEN** the `slow_query_waiting` column SHALL reflect the count at snapshot time
|
||||
|
||||
### Requirement: Slow-path query latency recorded in QueryMetrics
|
||||
The `read_sql_df_slow()` and `read_sql_df_slow_iter()` functions SHALL call `record_query_latency()` with the elapsed query time, so that P50/P95/P99 metrics reflect all query paths (pool + slow).
|
||||
|
||||
#### Scenario: Slow query latency appears in percentiles
|
||||
- **WHEN** a `read_sql_df_slow()` call completes in 5.2 seconds
|
||||
- **THEN** `record_query_latency(5.2)` SHALL be called and the latency SHALL appear in subsequent `get_percentiles()` results
|
||||
|
||||
#### Scenario: Slow iter latency recorded on completion
|
||||
- **WHEN** a `read_sql_df_slow_iter()` generator completes after yielding all batches in 120 seconds total
|
||||
- **THEN** `record_query_latency(120.0)` SHALL be called in the finally block
|
||||
|
||||
### Requirement: Slow query metrics displayed in Vue SPA
|
||||
The admin performance Vue SPA SHALL display `slow_query_active` and `slow_query_waiting` as StatCards in the connection pool panel, and include `slow_query_active` as a trend line in the connection pool trend chart.
|
||||
|
||||
#### Scenario: StatCards display current values
|
||||
- **WHEN** the performance-detail API returns `db_pool.status.slow_query_active = 4` and `db_pool.status.slow_query_waiting = 2`
|
||||
- **THEN** the connection pool panel SHALL display StatCards showing "慢查詢執行中: 4" and "慢查詢排隊中: 2"
|
||||
|
||||
#### Scenario: Trend chart includes slow_query_active
|
||||
- **WHEN** historical snapshots contain `slow_query_active` data points
|
||||
- **THEN** the connection pool trend chart SHALL include a "慢查詢執行中" line series
|
||||
@@ -0,0 +1,23 @@
|
||||
## ADDED Requirements
|
||||
|
||||
### Requirement: Worker RSS memory in metrics history snapshots
|
||||
The `MetricsHistoryCollector` SHALL include `worker_rss_bytes` in each 30-second snapshot, recording the current worker process peak RSS memory using Python's `resource.getrusage()`.
|
||||
|
||||
#### Scenario: RSS recorded in snapshot
|
||||
- **WHEN** the collector writes a snapshot and the worker process has 256 MB peak RSS
|
||||
- **THEN** the `worker_rss_bytes` column SHALL contain approximately 268435456
|
||||
|
||||
#### Scenario: RSS collection failure
|
||||
- **WHEN** `resource.getrusage()` raises an exception
|
||||
- **THEN** the collector SHALL write NULL for `worker_rss_bytes` and continue collecting other metrics
|
||||
|
||||
### Requirement: Worker memory trend chart in Vue SPA
|
||||
The admin performance Vue SPA SHALL display a "Worker 記憶體趨勢" TrendChart showing RSS memory over time in megabytes.
|
||||
|
||||
#### Scenario: Memory trend displayed
|
||||
- **WHEN** historical snapshots contain `worker_rss_bytes` data with more than 1 data point
|
||||
- **THEN** the dashboard SHALL display a TrendChart with RSS values converted to MB
|
||||
|
||||
#### Scenario: No memory data
|
||||
- **WHEN** historical snapshots do not contain `worker_rss_bytes` data (all NULL)
|
||||
- **THEN** the trend chart SHALL show "趨勢資料不足" message
|
||||
@@ -0,0 +1,37 @@
|
||||
## 1. 後端：Semaphore 排隊計數器 + Slow-path latency
|
||||
|
||||
- [x] 1.1 在 `src/mes_dashboard/core/database.py` 新增 `_SLOW_QUERY_WAITING` counter 和 `get_slow_query_waiting_count()` 函數
|
||||
- [x] 1.2 修改 `read_sql_df_slow()` 在 semaphore.acquire() 前後遞增/遞減 `_SLOW_QUERY_WAITING`
|
||||
- [x] 1.3 修改 `read_sql_df_slow_iter()` 同上加入 waiting counter 邏輯
|
||||
- [x] 1.4 修改 `get_pool_status()` 回傳中加入 `slow_query_waiting` 欄位
|
||||
- [x] 1.5 在 `read_sql_df_slow()` finally block 呼叫 `record_query_latency(elapsed)`
|
||||
- [x] 1.6 在 `read_sql_df_slow_iter()` finally block 呼叫 `record_query_latency(elapsed)`
|
||||
|
||||
## 2. 後端：metrics_history schema 擴充 + archive cleanup
|
||||
|
||||
- [x] 2.1 在 `src/mes_dashboard/core/metrics_history.py` 的 schema 新增 `slow_query_active INTEGER`, `slow_query_waiting INTEGER`, `worker_rss_bytes INTEGER` 欄位
|
||||
- [x] 2.2 在 `MetricsHistoryStore.initialize()` 加入 ALTER TABLE ADD COLUMN migration（容錯 duplicate column）
|
||||
- [x] 2.3 更新 `COLUMNS` list 加入新欄位
|
||||
- [x] 2.4 更新 `write_snapshot()` 加入新欄位的讀取和 INSERT
|
||||
- [x] 2.5 更新 `_collect_snapshot()` 收集 `slow_query_active`、`slow_query_waiting`（從 `get_pool_status()`）和 `worker_rss_bytes`（從 `resource.getrusage()`）
|
||||
- [x] 2.6 新增 `cleanup_archive_logs(archive_dir, keep_per_type)` 函數，含 `ARCHIVE_LOG_DIR` 和 `ARCHIVE_LOG_KEEP_COUNT` env var 配置
|
||||
- [x] 2.7 在 `MetricsHistoryCollector._run()` 的 cleanup cycle 呼叫 `cleanup_archive_logs()`
|
||||
|
||||
## 3. 後端：移除 Jinja fallback
|
||||
|
||||
- [x] 3.1 修改 `src/mes_dashboard/routes/admin_routes.py` 的 `performance()` 路由，移除 Jinja fallback 邏輯（改為直接 `send_from_directory`）
|
||||
- [x] 3.2 刪除 `src/mes_dashboard/templates/admin/performance.html`
|
||||
- [x] 3.3 更新 `scripts/check_full_modernization_gates.py` 將 `/admin/performance` 的 gate check 從 template 路徑改為 `frontend/src/admin-performance/style.css`
|
||||
|
||||
## 4. 前端：Vue SPA 新增監控面板
|
||||
|
||||
- [x] 4.1 在 `frontend/src/admin-performance/App.vue` 連線池 section 新增 `slow_query_active` 和 `slow_query_waiting` StatCards
|
||||
- [x] 4.2 在 `poolTrendSeries` 加入 `slow_query_active` 趨勢線
|
||||
- [x] 4.3 新增 `memoryTrendSeries` 定義和 Worker 記憶體 TrendChart 組件
|
||||
- [x] 4.4 新增 `historyData` 預處理邏輯：將 `worker_rss_bytes` 轉為 `worker_rss_mb`
|
||||
|
||||
## 5. Build + 測試驗證
|
||||
|
||||
- [x] 5.1 執行 `cd frontend && npx vite build` 確認 build 成功
|
||||
- [x] 5.2 執行 `python -m pytest tests/ -v --tb=short` 確認既有測試通過
|
||||
- [x] 5.3 確認 `test_performance_page_loads` 測試通過（SPA 路徑驗證）
|
||||
@@ -1,100 +1,37 @@
|
||||
## ADDED Requirements
|
||||
|
||||
### Requirement: Vue 3 SPA page replaces Jinja2 template
|
||||
The `/admin/performance` route SHALL serve a Vue 3 SPA page built by Vite, replacing the existing Jinja2 server-rendered template. The SPA SHALL be registered as a Vite entry point and integrated into the portal-shell navigation as a `renderMode: 'native'` route.
|
||||
|
||||
#### Scenario: Page loads as Vue SPA
|
||||
- **WHEN** user navigates to `/admin/performance`
|
||||
- **THEN** the server SHALL return the Vite-built `admin-performance.html` static file (not a Jinja2 rendered template)
|
||||
|
||||
#### Scenario: Portal-shell integration
|
||||
- **WHEN** the portal-shell renders `/admin/performance`
|
||||
- **THEN** it SHALL load the page as a native Vue SPA (not an external iframe)
|
||||
|
||||
### Requirement: Status cards display system health
|
||||
The dashboard SHALL display 4 status cards in a horizontal grid: Database, Redis, Circuit Breaker, and Worker PID. Each card SHALL show a StatusDot indicator (healthy/degraded/error/disabled) with the current status value.
|
||||
|
||||
#### Scenario: All systems healthy
|
||||
- **WHEN** all backend systems report healthy status via `/admin/api/system-status`
|
||||
- **THEN** all 4 status cards SHALL display green StatusDot indicators with their respective values
|
||||
|
||||
#### Scenario: Redis disabled
|
||||
- **WHEN** Redis is disabled (`REDIS_ENABLED=false`)
|
||||
- **THEN** the Redis status card SHALL display a disabled StatusDot indicator and the Redis cache panel SHALL show a graceful degradation message
|
||||
|
||||
### Requirement: Query performance panel with ECharts
|
||||
The dashboard SHALL display query performance metrics (P50, P95, P99 latencies, total queries, slow queries) and an ECharts latency distribution chart, replacing the existing Chart.js implementation.
|
||||
|
||||
#### Scenario: Metrics loaded successfully
|
||||
- **WHEN** `/admin/api/metrics` returns valid performance data
|
||||
- **THEN** the panel SHALL display P50/P95/P99 latency values and render an ECharts bar chart showing latency distribution
|
||||
|
||||
#### Scenario: No metrics data
|
||||
- **WHEN** `/admin/api/metrics` returns empty or null metrics
|
||||
- **THEN** the panel SHALL display placeholder text indicating no data available
|
||||
|
||||
### Requirement: Redis cache detail panel
|
||||
The dashboard SHALL display a Redis cache detail panel showing memory usage (as a GaugeBar), connected clients, hit rate percentage, peak memory, and a namespace key distribution table.
|
||||
|
||||
#### Scenario: Redis active with data
|
||||
- **WHEN** `/admin/api/performance-detail` returns Redis data with namespace key counts
|
||||
- **THEN** the panel SHALL display a memory GaugeBar, hit rate, client count, and a table listing each namespace with its key count
|
||||
|
||||
#### Scenario: Redis disabled
|
||||
- **WHEN** Redis is disabled
|
||||
- **THEN** the Redis detail panel SHALL display a disabled state message without errors
|
||||
|
||||
### Requirement: Memory cache panel
|
||||
The dashboard SHALL display ProcessLevelCache statistics as grid cards (showing entries/max_size as a mini gauge and TTL) plus Route Cache telemetry (L1 hit rate, L2 hit rate, miss rate, total reads).
|
||||
|
||||
#### Scenario: Multiple caches registered
|
||||
- **WHEN** `/admin/api/performance-detail` returns process_caches with multiple entries
|
||||
- **THEN** the panel SHALL render one card per cache instance showing entries, max_size, TTL, and description
|
||||
|
||||
#### Scenario: Route cache telemetry
|
||||
- **WHEN** `/admin/api/performance-detail` returns route_cache data
|
||||
- **THEN** the panel SHALL display L1 hit rate, L2 hit rate, miss rate, and total reads
|
||||
|
||||
### Requirement: Connection pool panel
|
||||
The dashboard SHALL display connection pool saturation as a GaugeBar and stat cards showing checked_out, checked_in, overflow, max_capacity, pool_size, pool_recycle, pool_timeout, and direct connection count.
|
||||
|
||||
#### Scenario: Pool under normal load
|
||||
- **WHEN** pool saturation is below 80%
|
||||
- **THEN** the GaugeBar SHALL display in a normal color (green/blue)
|
||||
|
||||
#### Scenario: Pool near saturation
|
||||
- **WHEN** pool saturation exceeds 80%
|
||||
- **THEN** the GaugeBar SHALL display in a warning color (yellow/orange/red)
|
||||
|
||||
### Requirement: Worker control panel
|
||||
The dashboard SHALL display worker PID, uptime, cooldown status, and provide a restart button with a confirmation modal.
|
||||
|
||||
#### Scenario: Restart worker
|
||||
- **WHEN** user clicks the restart button and confirms in the modal
|
||||
- **THEN** the system SHALL POST to `/admin/api/worker/restart` and display the result
|
||||
|
||||
#### Scenario: Restart during cooldown
|
||||
- **WHEN** worker is in cooldown period
|
||||
- **THEN** the restart button SHALL be disabled with a cooldown indicator
|
||||
|
||||
### Requirement: System logs panel with filtering and pagination
|
||||
The dashboard SHALL display system logs with level filtering, text search, and pagination controls.
|
||||
|
||||
#### Scenario: Filter by log level
|
||||
- **WHEN** user selects a specific log level filter
|
||||
- **THEN** only logs matching that level SHALL be displayed
|
||||
|
||||
#### Scenario: Paginate logs
|
||||
- **WHEN** logs exceed the page size
|
||||
- **THEN** pagination controls SHALL allow navigating between pages
|
||||
|
||||
### Requirement: Auto-refresh with toggle
|
||||
The dashboard SHALL auto-refresh all panels every 30 seconds using `useAutoRefresh`. The user SHALL be able to toggle auto-refresh on/off and manually trigger a refresh.
|
||||
|
||||
#### Scenario: Auto-refresh enabled
|
||||
- **WHEN** auto-refresh is enabled (default)
|
||||
- **THEN** all panels SHALL refresh their data every 30 seconds via `Promise.all` parallel fetch
|
||||
|
||||
#### Scenario: Manual refresh
|
||||
- **WHEN** user clicks the manual refresh button
|
||||
- **THEN** all panels SHALL immediately refresh their data
|
||||
## MODIFIED Requirements
|
||||
|
||||
### Requirement: Vue 3 SPA page replaces Jinja2 template
|
||||
The `/admin/performance` route SHALL serve the Vite-built `admin-performance.html` static file directly. The Jinja2 template fallback SHALL be removed. If the SPA build artifact does not exist, the server SHALL return a standard HTTP error (no fallback rendering).
|
||||
|
||||
#### Scenario: Page loads as Vue SPA
|
||||
- **WHEN** user navigates to `/admin/performance`
|
||||
- **THEN** the server SHALL return the Vite-built `admin-performance.html` static file via `send_from_directory`
|
||||
|
||||
#### Scenario: Portal-shell integration
|
||||
- **WHEN** the portal-shell renders `/admin/performance`
|
||||
- **THEN** it SHALL load the page as a native Vue SPA (not an external iframe)
|
||||
|
||||
#### Scenario: Build artifact missing
|
||||
- **WHEN** the SPA build artifact `admin-performance.html` does not exist in `static/dist/`
|
||||
- **THEN** the server SHALL return an HTTP error (no Jinja2 fallback)
|
||||
|
||||
### Requirement: Connection pool panel
|
||||
The dashboard SHALL display connection pool saturation as a GaugeBar and stat cards showing checked_out, checked_in, overflow, max_capacity, pool_size, pool_recycle, pool_timeout, direct connection count, slow_query_active, and slow_query_waiting.
|
||||
|
||||
#### Scenario: Pool under normal load
|
||||
- **WHEN** pool saturation is below 80%
|
||||
- **THEN** the GaugeBar SHALL display in a normal color (green/blue)
|
||||
|
||||
#### Scenario: Pool near saturation
|
||||
- **WHEN** pool saturation exceeds 80%
|
||||
- **THEN** the GaugeBar SHALL display in a warning color (yellow/orange/red)
|
||||
|
||||
#### Scenario: Slow query metrics displayed
|
||||
- **WHEN** `db_pool.status` includes `slow_query_active` and `slow_query_waiting`
|
||||
- **THEN** the panel SHALL display StatCards for both values
|
||||
|
||||
## REMOVED Requirements
|
||||
|
||||
### Requirement: Jinja2 template fallback for performance page
|
||||
**Reason**: The Vue SPA is the sole UI. Maintaining a 1249-line Jinja template as fallback adds maintenance burden and feature divergence.
|
||||
**Migration**: Delete `templates/admin/performance.html`. The route handler serves the SPA directly.
|
||||
|
||||
30
openspec/specs/archive-log-rotation/spec.md
Normal file
30
openspec/specs/archive-log-rotation/spec.md
Normal file
@@ -0,0 +1,30 @@
|
||||
## ADDED Requirements
|
||||
|
||||
### Requirement: Automatic archive log cleanup
|
||||
The system SHALL provide a `cleanup_archive_logs()` function in `core/metrics_history.py` that deletes old rotated log files from `logs/archive/`, keeping the most recent N files per log type (access, error, watchdog, rq_worker, startup).
|
||||
|
||||
#### Scenario: Cleanup keeps recent files
|
||||
- **WHEN** `cleanup_archive_logs()` is called with `keep_per_type=20` and there are 30 access_*.log files
|
||||
- **THEN** 10 oldest access_*.log files SHALL be deleted, keeping the 20 most recent by modification time
|
||||
|
||||
#### Scenario: No excess files
|
||||
- **WHEN** `cleanup_archive_logs()` is called and each type has fewer than `keep_per_type` files
|
||||
- **THEN** no files SHALL be deleted
|
||||
|
||||
#### Scenario: Archive directory missing
|
||||
- **WHEN** `cleanup_archive_logs()` is called and the archive directory does not exist
|
||||
- **THEN** the function SHALL return 0 without error
|
||||
|
||||
### Requirement: Archive cleanup integrated into collector cycle
|
||||
The `MetricsHistoryCollector` SHALL call `cleanup_archive_logs()` alongside the existing SQLite cleanup, running approximately every 50 minutes (every 100 collection intervals).
|
||||
|
||||
#### Scenario: Periodic cleanup executes
|
||||
- **WHEN** the cleanup counter reaches 100 intervals
|
||||
- **THEN** both SQLite metrics cleanup and archive log cleanup SHALL execute
|
||||
|
||||
### Requirement: Archive cleanup configuration
|
||||
The archive log cleanup SHALL be configurable via environment variables: `ARCHIVE_LOG_DIR` (default: `logs/archive`) and `ARCHIVE_LOG_KEEP_COUNT` (default: 20).
|
||||
|
||||
#### Scenario: Custom keep count
|
||||
- **WHEN** `ARCHIVE_LOG_KEEP_COUNT=10` is set
|
||||
- **THEN** cleanup SHALL keep only the 10 most recent files per type
|
||||
@@ -1,27 +1,29 @@
|
||||
## ADDED Requirements
|
||||
|
||||
### Requirement: Connection pool status in performance detail
|
||||
The performance-detail API SHALL include `db_pool` section with `status` (checked_out, checked_in, overflow, max_capacity, saturation) from `get_pool_status()` and `config` (pool_size, max_overflow, pool_timeout, pool_recycle) from `get_pool_runtime_config()`.
|
||||
|
||||
#### Scenario: Pool status retrieved
|
||||
- **WHEN** the API is called
|
||||
- **THEN** `db_pool.status` SHALL contain current pool utilization metrics and `db_pool.config` SHALL contain the pool configuration values
|
||||
|
||||
#### Scenario: Saturation calculation
|
||||
- **WHEN** the pool has 8 checked_out connections and max_capacity is 30
|
||||
- **THEN** saturation SHALL be reported as approximately 26.7%
|
||||
|
||||
### Requirement: Direct Oracle connection counter
|
||||
The system SHALL maintain a thread-safe monotonic counter in `database.py` that increments each time `get_db_connection()` or `read_sql_df_slow()` successfully creates a direct (non-pooled) Oracle connection.
|
||||
|
||||
#### Scenario: Counter increments on direct connection
|
||||
- **WHEN** `get_db_connection()` successfully creates a connection
|
||||
- **THEN** the direct connection counter SHALL increment by 1
|
||||
|
||||
#### Scenario: Counter in performance detail
|
||||
- **WHEN** the performance-detail API is called
|
||||
- **THEN** `direct_connections` SHALL contain `total_since_start` (counter value) and `worker_pid` (current process PID)
|
||||
|
||||
#### Scenario: Counter is per-worker
|
||||
- **WHEN** multiple gunicorn workers are running
|
||||
- **THEN** each worker SHALL maintain its own independent counter, and the API SHALL return the counter for the responding worker
|
||||
## MODIFIED Requirements
|
||||
|
||||
### Requirement: Connection pool status in performance detail
|
||||
The performance-detail API SHALL include `db_pool` section with `status` (checked_out, checked_in, overflow, max_capacity, saturation, slow_query_active, slow_query_waiting) from `get_pool_status()` and `config` (pool_size, max_overflow, pool_timeout, pool_recycle) from `get_pool_runtime_config()`.
|
||||
|
||||
#### Scenario: Pool status retrieved
|
||||
- **WHEN** the API is called
|
||||
- **THEN** `db_pool.status` SHALL contain current pool utilization metrics including `slow_query_active` and `slow_query_waiting`, and `db_pool.config` SHALL contain the pool configuration values
|
||||
|
||||
#### Scenario: Saturation calculation
|
||||
- **WHEN** the pool has 8 checked_out connections and max_capacity is 30
|
||||
- **THEN** saturation SHALL be reported as approximately 26.7%
|
||||
|
||||
#### Scenario: Slow query waiting included
|
||||
- **WHEN** 2 threads are waiting for the slow query semaphore
|
||||
- **THEN** `db_pool.status.slow_query_waiting` SHALL be 2
|
||||
|
||||
## ADDED Requirements
|
||||
|
||||
### Requirement: Slow-path query latency included in QueryMetrics
|
||||
The `read_sql_df_slow()` and `read_sql_df_slow_iter()` functions SHALL call `record_query_latency()` with the total elapsed time upon completion, ensuring P50/P95/P99 percentiles reflect queries from all paths (pooled and slow/direct).
|
||||
|
||||
#### Scenario: Slow query latency recorded
|
||||
- **WHEN** `read_sql_df_slow()` completes a query in 8.5 seconds
|
||||
- **THEN** `record_query_latency(8.5)` SHALL be called and the value SHALL appear in subsequent `get_percentiles()` results
|
||||
|
||||
#### Scenario: Slow iter latency recorded
|
||||
- **WHEN** `read_sql_df_slow_iter()` completes streaming in 45 seconds
|
||||
- **THEN** `record_query_latency(45.0)` SHALL be called in the finally block
|
||||
|
||||
@@ -1,65 +1,54 @@
|
||||
## ADDED Requirements
|
||||
|
||||
### Requirement: SQLite metrics history store
|
||||
The system SHALL provide a `MetricsHistoryStore` class in `core/metrics_history.py` that persists metrics snapshots to a SQLite database (`logs/metrics_history.sqlite` by default). The store SHALL use thread-local connections and a write lock, following the `LogStore` pattern in `core/log_store.py`.
|
||||
|
||||
#### Scenario: Write and query snapshots
|
||||
- **WHEN** `write_snapshot(data)` is called with pool/redis/route_cache/latency metrics
|
||||
- **THEN** a row SHALL be inserted into `metrics_snapshots` with the current ISO 8601 timestamp and worker PID
|
||||
|
||||
#### Scenario: Query by time range
|
||||
- **WHEN** `query_snapshots(minutes=30)` is called
|
||||
- **THEN** it SHALL return all rows from the last 30 minutes, ordered by timestamp ascending
|
||||
|
||||
#### Scenario: Retention cleanup
|
||||
- **WHEN** `cleanup()` is called
|
||||
- **THEN** rows older than `METRICS_HISTORY_RETENTION_DAYS` (default 3) SHALL be deleted, and total rows SHALL be capped at `METRICS_HISTORY_MAX_ROWS` (default 50000)
|
||||
|
||||
#### Scenario: Thread safety
|
||||
- **WHEN** multiple threads write snapshots concurrently
|
||||
- **THEN** the write lock SHALL serialize writes and prevent database corruption
|
||||
|
||||
### Requirement: Background metrics collector
|
||||
The system SHALL provide a `MetricsHistoryCollector` class that runs a daemon thread collecting metrics snapshots at a configurable interval (default 30 seconds, via `METRICS_HISTORY_INTERVAL` env var).
|
||||
|
||||
#### Scenario: Automatic collection
|
||||
- **WHEN** the collector is started via `start_metrics_history(app)`
|
||||
- **THEN** it SHALL collect pool status, Redis info, route cache status, and query latency metrics every interval and write them to the store
|
||||
|
||||
#### Scenario: Graceful shutdown
|
||||
- **WHEN** `stop_metrics_history()` is called
|
||||
- **THEN** the collector thread SHALL stop within one interval period
|
||||
|
||||
#### Scenario: Subsystem unavailability
|
||||
- **WHEN** a subsystem (e.g., Redis) is unavailable during collection
|
||||
- **THEN** the collector SHALL write null/0 for those fields and continue collecting other metrics
|
||||
|
||||
### Requirement: Performance history API endpoint
|
||||
The system SHALL expose `GET /admin/api/performance-history` that returns historical metrics snapshots.
|
||||
|
||||
#### Scenario: Query with time range
|
||||
- **WHEN** the API is called with `?minutes=30`
|
||||
- **THEN** it SHALL return `{"success": true, "data": {"snapshots": [...], "count": N}}`
|
||||
|
||||
#### Scenario: Time range bounds
|
||||
- **WHEN** `minutes` is less than 1 or greater than 180
|
||||
- **THEN** it SHALL be clamped to the range [1, 180]
|
||||
|
||||
#### Scenario: Admin authentication
|
||||
- **WHEN** the API is called without admin authentication
|
||||
- **THEN** it SHALL be rejected by the `@admin_required` decorator
|
||||
|
||||
### Requirement: Frontend trend charts
|
||||
The system SHALL display 4 trend chart panels in the admin performance dashboard using vue-echarts VChart line/area charts.
|
||||
|
||||
#### Scenario: Trend charts with data
|
||||
- **WHEN** historical snapshots contain more than 1 data point
|
||||
- **THEN** the dashboard SHALL display trend charts for: connection pool saturation, query latency (P50/P95/P99), Redis memory, and cache hit rates
|
||||
|
||||
#### Scenario: Trend charts without data
|
||||
- **WHEN** historical snapshots are empty or contain only 1 data point
|
||||
- **THEN** the trend charts SHALL NOT be displayed (hidden via `v-if`)
|
||||
|
||||
#### Scenario: Auto-refresh
|
||||
- **WHEN** the dashboard auto-refreshes
|
||||
- **THEN** historical data SHALL also be refreshed alongside real-time metrics
|
||||
## MODIFIED Requirements
|
||||
|
||||
### Requirement: SQLite metrics history store
|
||||
The system SHALL provide a `MetricsHistoryStore` class in `core/metrics_history.py` that persists metrics snapshots to a SQLite database (`logs/metrics_history.sqlite` by default). The store SHALL use thread-local connections and a write lock, following the `LogStore` pattern in `core/log_store.py`. The schema SHALL include columns for `slow_query_active` (INTEGER), `slow_query_waiting` (INTEGER), and `worker_rss_bytes` (INTEGER) in addition to the existing pool, Redis, route cache, and latency columns.
|
||||
|
||||
#### Scenario: Write and query snapshots
|
||||
- **WHEN** `write_snapshot(data)` is called with pool/redis/route_cache/latency/slow_query/memory metrics
|
||||
- **THEN** a row SHALL be inserted into `metrics_snapshots` with the current ISO 8601 timestamp, worker PID, and all metric columns
|
||||
|
||||
#### Scenario: Query by time range
|
||||
- **WHEN** `query_snapshots(minutes=30)` is called
|
||||
- **THEN** it SHALL return all rows from the last 30 minutes, ordered by timestamp ascending, including the new columns
|
||||
|
||||
#### Scenario: Retention cleanup
|
||||
- **WHEN** `cleanup()` is called
|
||||
- **THEN** rows older than `METRICS_HISTORY_RETENTION_DAYS` (default 3) SHALL be deleted, and total rows SHALL be capped at `METRICS_HISTORY_MAX_ROWS` (default 50000)
|
||||
|
||||
#### Scenario: Thread safety
|
||||
- **WHEN** multiple threads write snapshots concurrently
|
||||
- **THEN** the write lock SHALL serialize writes and prevent database corruption
|
||||
|
||||
#### Scenario: Schema migration for existing databases
|
||||
- **WHEN** the store initializes on an existing database without the new columns
|
||||
- **THEN** it SHALL execute ALTER TABLE ADD COLUMN for each missing column, tolerating "duplicate column" errors
|
||||
|
||||
### Requirement: Background metrics collector
|
||||
The system SHALL provide a `MetricsHistoryCollector` class that runs a daemon thread collecting metrics snapshots at a configurable interval (default 30 seconds, via `METRICS_HISTORY_INTERVAL` env var). The collector SHALL include `slow_query_active`, `slow_query_waiting`, and `worker_rss_bytes` in each snapshot.
|
||||
|
||||
#### Scenario: Automatic collection
|
||||
- **WHEN** the collector is started via `start_metrics_history(app)`
|
||||
- **THEN** it SHALL collect pool status (including slow_query_active and slow_query_waiting), Redis info, route cache status, query latency metrics, and worker RSS memory every interval and write them to the store
|
||||
|
||||
#### Scenario: Graceful shutdown
|
||||
- **WHEN** `stop_metrics_history()` is called
|
||||
- **THEN** the collector thread SHALL stop within one interval period
|
||||
|
||||
#### Scenario: Subsystem unavailability
|
||||
- **WHEN** a subsystem (e.g., Redis) is unavailable during collection
|
||||
- **THEN** the collector SHALL write null/0 for those fields and continue collecting other metrics
|
||||
|
||||
### Requirement: Frontend trend charts
|
||||
The system SHALL display 5 trend chart panels in the admin performance dashboard using vue-echarts VChart line/area charts: connection pool saturation, query latency (P50/P95/P99), Redis memory, cache hit rates, and worker memory.
|
||||
|
||||
#### Scenario: Trend charts with data
|
||||
- **WHEN** historical snapshots contain more than 1 data point
|
||||
- **THEN** the dashboard SHALL display trend charts for: connection pool saturation (including slow_query_active), query latency (P50/P95/P99), Redis memory, cache hit rates, and worker memory (RSS in MB)
|
||||
|
||||
#### Scenario: Trend charts without data
|
||||
- **WHEN** historical snapshots are empty or contain only 1 data point
|
||||
- **THEN** the trend charts SHALL NOT be displayed (hidden via `v-if`)
|
||||
|
||||
#### Scenario: Auto-refresh
|
||||
- **WHEN** the dashboard auto-refreshes
|
||||
- **THEN** historical data SHALL also be refreshed alongside real-time metrics
|
||||
|
||||
49
openspec/specs/slow-query-observability/spec.md
Normal file
49
openspec/specs/slow-query-observability/spec.md
Normal file
@@ -0,0 +1,49 @@
|
||||
## ADDED Requirements
|
||||
|
||||
### Requirement: Slow query active count in metrics history snapshots
|
||||
The `MetricsHistoryCollector` SHALL include `slow_query_active` in each 30-second snapshot, recording the number of slow queries currently executing via dedicated connections.
|
||||
|
||||
#### Scenario: Snapshot includes slow_query_active
|
||||
- **WHEN** the collector writes a snapshot while 3 slow queries are executing
|
||||
- **THEN** the `slow_query_active` column SHALL contain the value 3
|
||||
|
||||
#### Scenario: No slow queries active
|
||||
- **WHEN** the collector writes a snapshot while no slow queries are executing
|
||||
- **THEN** the `slow_query_active` column SHALL contain the value 0
|
||||
|
||||
### Requirement: Slow query waiting count tracked and persisted
|
||||
The system SHALL maintain a thread-safe counter `_SLOW_QUERY_WAITING` in `database.py` that tracks the number of threads currently waiting to acquire the slow query semaphore. This counter SHALL be included in `get_pool_status()` and persisted to metrics history snapshots.
|
||||
|
||||
#### Scenario: Counter increments on semaphore wait
|
||||
- **WHEN** a thread enters `read_sql_df_slow()` and the semaphore is full
|
||||
- **THEN** `_SLOW_QUERY_WAITING` SHALL be incremented before `semaphore.acquire()` and decremented after acquire completes (success or timeout)
|
||||
|
||||
#### Scenario: Counter in pool status API
|
||||
- **WHEN** `get_pool_status()` is called
|
||||
- **THEN** the returned dict SHALL include `slow_query_waiting` with the current waiting thread count
|
||||
|
||||
#### Scenario: Counter persisted to metrics history
|
||||
- **WHEN** the collector writes a snapshot
|
||||
- **THEN** the `slow_query_waiting` column SHALL reflect the count at snapshot time
|
||||
|
||||
### Requirement: Slow-path query latency recorded in QueryMetrics
|
||||
The `read_sql_df_slow()` and `read_sql_df_slow_iter()` functions SHALL call `record_query_latency()` with the elapsed query time, so that P50/P95/P99 metrics reflect all query paths (pool + slow).
|
||||
|
||||
#### Scenario: Slow query latency appears in percentiles
|
||||
- **WHEN** a `read_sql_df_slow()` call completes in 5.2 seconds
|
||||
- **THEN** `record_query_latency(5.2)` SHALL be called and the latency SHALL appear in subsequent `get_percentiles()` results
|
||||
|
||||
#### Scenario: Slow iter latency recorded on completion
|
||||
- **WHEN** a `read_sql_df_slow_iter()` generator completes after yielding all batches in 120 seconds total
|
||||
- **THEN** `record_query_latency(120.0)` SHALL be called in the finally block
|
||||
|
||||
### Requirement: Slow query metrics displayed in Vue SPA
|
||||
The admin performance Vue SPA SHALL display `slow_query_active` and `slow_query_waiting` as StatCards in the connection pool panel, and include `slow_query_active` as a trend line in the connection pool trend chart.
|
||||
|
||||
#### Scenario: StatCards display current values
|
||||
- **WHEN** the performance-detail API returns `db_pool.status.slow_query_active = 4` and `db_pool.status.slow_query_waiting = 2`
|
||||
- **THEN** the connection pool panel SHALL display StatCards showing "慢查詢執行中: 4" and "慢查詢排隊中: 2"
|
||||
|
||||
#### Scenario: Trend chart includes slow_query_active
|
||||
- **WHEN** historical snapshots contain `slow_query_active` data points
|
||||
- **THEN** the connection pool trend chart SHALL include a "慢查詢執行中" line series
|
||||
23
openspec/specs/worker-memory-tracking/spec.md
Normal file
23
openspec/specs/worker-memory-tracking/spec.md
Normal file
@@ -0,0 +1,23 @@
|
||||
## ADDED Requirements
|
||||
|
||||
### Requirement: Worker RSS memory in metrics history snapshots
|
||||
The `MetricsHistoryCollector` SHALL include `worker_rss_bytes` in each 30-second snapshot, recording the current worker process peak RSS memory using Python's `resource.getrusage()`.
|
||||
|
||||
#### Scenario: RSS recorded in snapshot
|
||||
- **WHEN** the collector writes a snapshot and the worker process has 256 MB peak RSS
|
||||
- **THEN** the `worker_rss_bytes` column SHALL contain approximately 268435456
|
||||
|
||||
#### Scenario: RSS collection failure
|
||||
- **WHEN** `resource.getrusage()` raises an exception
|
||||
- **THEN** the collector SHALL write NULL for `worker_rss_bytes` and continue collecting other metrics
|
||||
|
||||
### Requirement: Worker memory trend chart in Vue SPA
|
||||
The admin performance Vue SPA SHALL display a "Worker 記憶體趨勢" TrendChart showing RSS memory over time in megabytes.
|
||||
|
||||
#### Scenario: Memory trend displayed
|
||||
- **WHEN** historical snapshots contain `worker_rss_bytes` data with more than 1 data point
|
||||
- **THEN** the dashboard SHALL display a TrendChart with RSS values converted to MB
|
||||
|
||||
#### Scenario: No memory data
|
||||
- **WHEN** historical snapshots do not contain `worker_rss_bytes` data (all NULL)
|
||||
- **THEN** the trend chart SHALL show "趨勢資料不足" message
|
||||
@@ -86,7 +86,7 @@ def _route_css_targets() -> dict[str, list[Path]]:
|
||||
"/qc-gate": [ROOT / "frontend/src/qc-gate/style.css"],
|
||||
"/job-query": [ROOT / "frontend/src/job-query/style.css"],
|
||||
"/admin/pages": [ROOT / "src/mes_dashboard/templates/admin/pages.html"],
|
||||
"/admin/performance": [ROOT / "src/mes_dashboard/templates/admin/performance.html"],
|
||||
"/admin/performance": [ROOT / "frontend/src/admin-performance/style.css"],
|
||||
"/tables": [ROOT / "frontend/src/tables/style.css"],
|
||||
"/excel-query": [ROOT / "frontend/src/excel-query/style.css"],
|
||||
"/query-tool": [ROOT / "frontend/src/query-tool/style.css"],
|
||||
|
||||
@@ -233,6 +233,7 @@ def get_pool_status() -> Dict[str, Any]:
|
||||
"max_capacity": max_capacity,
|
||||
"saturation": saturation,
|
||||
"slow_query_active": get_slow_query_active_count(),
|
||||
"slow_query_waiting": get_slow_query_waiting_count(),
|
||||
}
|
||||
|
||||
|
||||
@@ -436,6 +437,7 @@ _DIRECT_CONN_LOCK = threading.Lock()
|
||||
# Slow-query concurrency control
|
||||
_SLOW_QUERY_SEMAPHORE: Optional[threading.Semaphore] = None
|
||||
_SLOW_QUERY_ACTIVE = 0
|
||||
_SLOW_QUERY_WAITING = 0
|
||||
_SLOW_QUERY_LOCK = threading.Lock()
|
||||
|
||||
|
||||
@@ -454,6 +456,11 @@ def get_slow_query_active_count() -> int:
|
||||
return _SLOW_QUERY_ACTIVE
|
||||
|
||||
|
||||
def get_slow_query_waiting_count() -> int:
|
||||
"""Return the number of threads waiting for the slow-query semaphore."""
|
||||
return _SLOW_QUERY_WAITING
|
||||
|
||||
|
||||
def get_direct_connection_count() -> int:
|
||||
"""Return total direct (non-pooled) connections since worker start."""
|
||||
return _DIRECT_CONN_COUNTER
|
||||
@@ -627,7 +634,8 @@ def read_sql_df_slow(
|
||||
Returns:
|
||||
DataFrame with query results.
|
||||
"""
|
||||
global _SLOW_QUERY_ACTIVE
|
||||
global _SLOW_QUERY_ACTIVE, _SLOW_QUERY_WAITING
|
||||
from mes_dashboard.core.metrics import record_query_latency
|
||||
|
||||
runtime = get_db_runtime_config()
|
||||
if timeout_seconds is None:
|
||||
@@ -636,7 +644,13 @@ def read_sql_df_slow(
|
||||
timeout_ms = timeout_seconds * 1000
|
||||
|
||||
sem = _get_slow_query_semaphore()
|
||||
acquired = sem.acquire(timeout=60)
|
||||
with _SLOW_QUERY_LOCK:
|
||||
_SLOW_QUERY_WAITING += 1
|
||||
try:
|
||||
acquired = sem.acquire(timeout=60)
|
||||
finally:
|
||||
with _SLOW_QUERY_LOCK:
|
||||
_SLOW_QUERY_WAITING -= 1
|
||||
if not acquired:
|
||||
raise RuntimeError(
|
||||
"Slow-query concurrency limit reached; try again later"
|
||||
@@ -690,6 +704,8 @@ def read_sql_df_slow(
|
||||
)
|
||||
raise
|
||||
finally:
|
||||
elapsed = time.time() - start_time
|
||||
record_query_latency(elapsed)
|
||||
if conn:
|
||||
try:
|
||||
conn.close()
|
||||
@@ -718,7 +734,8 @@ def read_sql_df_slow_iter(
|
||||
batch_size: Number of rows per fetchmany call. None = use
|
||||
DB_SLOW_FETCHMANY_SIZE from config (default 5000).
|
||||
"""
|
||||
global _SLOW_QUERY_ACTIVE
|
||||
global _SLOW_QUERY_ACTIVE, _SLOW_QUERY_WAITING
|
||||
from mes_dashboard.core.metrics import record_query_latency
|
||||
|
||||
runtime = get_db_runtime_config()
|
||||
if timeout_seconds is None:
|
||||
@@ -730,7 +747,13 @@ def read_sql_df_slow_iter(
|
||||
batch_size = runtime["slow_fetchmany_size"]
|
||||
|
||||
sem = _get_slow_query_semaphore()
|
||||
acquired = sem.acquire(timeout=60)
|
||||
with _SLOW_QUERY_LOCK:
|
||||
_SLOW_QUERY_WAITING += 1
|
||||
try:
|
||||
acquired = sem.acquire(timeout=60)
|
||||
finally:
|
||||
with _SLOW_QUERY_LOCK:
|
||||
_SLOW_QUERY_WAITING -= 1
|
||||
if not acquired:
|
||||
raise RuntimeError(
|
||||
"Slow-query concurrency limit reached; try again later"
|
||||
@@ -788,6 +811,8 @@ def read_sql_df_slow_iter(
|
||||
)
|
||||
raise
|
||||
finally:
|
||||
elapsed = time.time() - start_time
|
||||
record_query_latency(elapsed)
|
||||
if conn:
|
||||
try:
|
||||
conn.close()
|
||||
|
||||
@@ -32,6 +32,9 @@ METRICS_HISTORY_INTERVAL = int(os.getenv('METRICS_HISTORY_INTERVAL', '30'))
|
||||
METRICS_HISTORY_RETENTION_DAYS = int(os.getenv('METRICS_HISTORY_RETENTION_DAYS', '3'))
|
||||
METRICS_HISTORY_MAX_ROWS = int(os.getenv('METRICS_HISTORY_MAX_ROWS', '50000'))
|
||||
|
||||
ARCHIVE_LOG_DIR = os.getenv('ARCHIVE_LOG_DIR', 'logs/archive')
|
||||
ARCHIVE_LOG_KEEP_COUNT = int(os.getenv('ARCHIVE_LOG_KEEP_COUNT', '20'))
|
||||
|
||||
# ============================================================
|
||||
# Database Schema
|
||||
# ============================================================
|
||||
@@ -54,10 +57,20 @@ CREATE TABLE IF NOT EXISTS metrics_snapshots (
|
||||
latency_p50_ms REAL,
|
||||
latency_p95_ms REAL,
|
||||
latency_p99_ms REAL,
|
||||
latency_count INTEGER
|
||||
latency_count INTEGER,
|
||||
slow_query_active INTEGER,
|
||||
slow_query_waiting INTEGER,
|
||||
worker_rss_bytes INTEGER
|
||||
);
|
||||
"""
|
||||
|
||||
# New columns added after initial schema — used for ALTER TABLE migration.
|
||||
_MIGRATION_COLUMNS = [
|
||||
("slow_query_active", "INTEGER"),
|
||||
("slow_query_waiting", "INTEGER"),
|
||||
("worker_rss_bytes", "INTEGER"),
|
||||
]
|
||||
|
||||
CREATE_INDEX_SQL = (
|
||||
"CREATE INDEX IF NOT EXISTS idx_metrics_ts ON metrics_snapshots(ts);"
|
||||
)
|
||||
@@ -69,9 +82,49 @@ COLUMNS = [
|
||||
"redis_used_memory", "redis_hit_rate",
|
||||
"rc_l1_hit_rate", "rc_l2_hit_rate", "rc_miss_rate",
|
||||
"latency_p50_ms", "latency_p95_ms", "latency_p99_ms", "latency_count",
|
||||
"slow_query_active", "slow_query_waiting", "worker_rss_bytes",
|
||||
]
|
||||
|
||||
|
||||
# ============================================================
|
||||
# Archive Log Cleanup
|
||||
# ============================================================
|
||||
|
||||
_ARCHIVE_LOG_PREFIXES = ("access_", "error_", "watchdog_", "rq_worker_", "startup_")
|
||||
|
||||
|
||||
def cleanup_archive_logs(
|
||||
archive_dir: str = ARCHIVE_LOG_DIR,
|
||||
keep_per_type: int = ARCHIVE_LOG_KEEP_COUNT,
|
||||
) -> int:
|
||||
"""Delete old rotated log files from the archive directory.
|
||||
|
||||
Keeps the most recent *keep_per_type* files per log type (by mtime).
|
||||
Returns the total number of files deleted.
|
||||
"""
|
||||
archive_path = Path(archive_dir)
|
||||
if not archive_path.is_dir():
|
||||
return 0
|
||||
|
||||
deleted = 0
|
||||
for prefix in _ARCHIVE_LOG_PREFIXES:
|
||||
files = sorted(
|
||||
(f for f in archive_path.iterdir() if f.name.startswith(prefix) and f.is_file()),
|
||||
key=lambda f: f.stat().st_mtime,
|
||||
reverse=True,
|
||||
)
|
||||
for old_file in files[keep_per_type:]:
|
||||
try:
|
||||
old_file.unlink()
|
||||
deleted += 1
|
||||
except OSError as exc:
|
||||
logger.warning("Failed to delete archive log %s: %s", old_file, exc)
|
||||
|
||||
if deleted > 0:
|
||||
logger.info("Cleaned up %d archive log files from %s", deleted, archive_dir)
|
||||
return deleted
|
||||
|
||||
|
||||
# ============================================================
|
||||
# Metrics History Store
|
||||
# ============================================================
|
||||
@@ -94,6 +147,14 @@ class MetricsHistoryStore:
|
||||
cursor = conn.cursor()
|
||||
cursor.execute(CREATE_TABLE_SQL)
|
||||
cursor.execute(CREATE_INDEX_SQL)
|
||||
# Migrate existing databases: add new columns if missing.
|
||||
for col_name, col_type in _MIGRATION_COLUMNS:
|
||||
try:
|
||||
cursor.execute(
|
||||
f"ALTER TABLE metrics_snapshots ADD COLUMN {col_name} {col_type}"
|
||||
)
|
||||
except sqlite3.OperationalError:
|
||||
pass # Column already exists — tolerate duplicate column error.
|
||||
conn.commit()
|
||||
self._initialized = True
|
||||
logger.info("Metrics history store initialized at %s", self.db_path)
|
||||
@@ -136,8 +197,9 @@ class MetricsHistoryStore:
|
||||
pool_overflow, pool_max_capacity,
|
||||
redis_used_memory, redis_hit_rate,
|
||||
rc_l1_hit_rate, rc_l2_hit_rate, rc_miss_rate,
|
||||
latency_p50_ms, latency_p95_ms, latency_p99_ms, latency_count)
|
||||
VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)
|
||||
latency_p50_ms, latency_p95_ms, latency_p99_ms, latency_count,
|
||||
slow_query_active, slow_query_waiting, worker_rss_bytes)
|
||||
VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)
|
||||
""",
|
||||
(
|
||||
ts, pid,
|
||||
@@ -155,6 +217,9 @@ class MetricsHistoryStore:
|
||||
lat.get("p95_ms"),
|
||||
lat.get("p99_ms"),
|
||||
lat.get("count"),
|
||||
data.get("slow_query_active"),
|
||||
data.get("slow_query_waiting"),
|
||||
data.get("worker_rss_bytes"),
|
||||
),
|
||||
)
|
||||
conn.commit()
|
||||
@@ -263,17 +328,36 @@ class MetricsHistoryCollector:
|
||||
if self._cleanup_counter >= 100:
|
||||
self._cleanup_counter = 0
|
||||
self._store.cleanup()
|
||||
try:
|
||||
cleanup_archive_logs()
|
||||
except Exception as exc:
|
||||
logger.debug("Archive log cleanup failed: %s", exc)
|
||||
|
||||
def _collect_snapshot(self) -> None:
|
||||
try:
|
||||
data: Dict[str, Any] = {}
|
||||
|
||||
# Pool status
|
||||
# Pool status (includes slow_query_active and slow_query_waiting)
|
||||
try:
|
||||
from mes_dashboard.core.database import get_pool_status
|
||||
data["pool"] = get_pool_status()
|
||||
pool_status = get_pool_status()
|
||||
data["pool"] = pool_status
|
||||
data["slow_query_active"] = pool_status.get("slow_query_active", 0)
|
||||
data["slow_query_waiting"] = pool_status.get("slow_query_waiting", 0)
|
||||
except Exception:
|
||||
data["pool"] = {}
|
||||
data["slow_query_active"] = 0
|
||||
data["slow_query_waiting"] = 0
|
||||
|
||||
# Worker RSS memory
|
||||
try:
|
||||
import resource
|
||||
# ru_maxrss is in KB on Linux
|
||||
data["worker_rss_bytes"] = resource.getrusage(
|
||||
resource.RUSAGE_SELF
|
||||
).ru_maxrss * 1024
|
||||
except Exception:
|
||||
data["worker_rss_bytes"] = 0
|
||||
|
||||
# Redis
|
||||
try:
|
||||
|
||||
@@ -71,10 +71,7 @@ _last_restart_request: float = 0.0
|
||||
def performance():
|
||||
"""Performance monitoring dashboard (Vue SPA)."""
|
||||
dist_dir = os.path.join(current_app.static_folder or "", "dist")
|
||||
dist_html = os.path.join(dist_dir, "admin-performance.html")
|
||||
if os.path.exists(dist_html):
|
||||
return send_from_directory(dist_dir, "admin-performance.html")
|
||||
return render_template("admin/performance.html")
|
||||
return send_from_directory(dist_dir, "admin-performance.html")
|
||||
|
||||
|
||||
@admin_bp.route("/api/system-status", methods=["GET"])
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
Reference in New Issue
Block a user