fix: not watching hidden files when triggering hot reload.

Fix incorrect page count calculation.

.editorconfig

@@ -15,6 +15,9 @@ trim_trailing_whitespace = true
 [project.json]
 indent_size = 2
 
+[*.{yaml,yml}]
+indent_size = 2
+
 # C# and Visual Basic files
 [*.{cs,vb}]
 charset = utf-8-bom

.gitea/workflows/build.yaml

@@ -1,33 +1,36 @@
 name: Build blog docker image
 on:
-    push:
+  push:
-        branches:
+    branches:
-            - master
+      - master
 jobs:
-    Build-Blog-Image:
+  Build-Blog-Image:
-        runs-on: archlinux
+    runs-on: archlinux
-        steps:
+    steps:
-            -   uses: https://mirrors.rrricardo.top/actions/checkout.git@v4
+      - name: Check out code.
-                name: Check out code
+        uses: https://mirrors.rrricardo.top/actions/checkout.git@v4
-                with:
+        with:
-                    lfs: true
+          lfs: true
-            -   name: Build project
+      - name: Build project.
-                run: |
+        run: |
-                    cd YaeBlog
+          cd YaeBlog
-                    dotnet publish
+          dotnet publish
-            -   name: Build docker image
+      - name: Build docker image.
-                run: |
+        run: |
-                    cd YaeBlog
+          proxy
-                    podman build . -t registry.cn-beijing.aliyuncs.com/jackfiled/blog:latest --build-arg COMMIT_ID=$(git rev-parse --short=10 HEAD)
+          podman pull mcr.microsoft.com/dotnet/aspnet:9.0
-            -   name: Workaround to make sure podman login succeed
+          unproxy
-                run: |
+          cd YaeBlog
-                    mkdir /root/.docker
+          podman build . -t ccr.ccs.tencentyun.com/jackfiled/blog --build-arg COMMIT_ID=$(git rev-parse --short=10 HEAD)
-            -   name: Login aliyun docker registry
+      - name: Workaround to make sure podman-login working.
-                uses: https://mirrors.rrricardo.top/actions/podman-login.git@v1
+        run: |
-                with:
+          mkdir /root/.docker
-                    registry: registry.cn-beijing.aliyuncs.com
+      - name: Login tencent cloud docker registry.
-                    username: 初冬的朝阳
+        uses: https://mirrors.rrricardo.top/actions/podman-login.git@v1
-                    password: ${{ secrets.ALIYUN_PASSWORD }}
+        with:
-                    auth_file_path: /etc/containers/auth.json
+          registry: ccr.ccs.tencentyun.com
-            -   name: Push docker image
+          username: 100044380877
-                run: podman push registry.cn-beijing.aliyuncs.com/jackfiled/blog:latest
+          password: ${{ secrets.TENCENT_REGISTRY_PASSWORD }}
+          auth_file_path: /etc/containers/auth.json
+      - name: Push docker image.
+        run: podman push ccr.ccs.tencentyun.com/jackfiled/blog:latest

YaeBlog/Pages/BlogIndex.razor

@@ -39,6 +39,11 @@
     {
         _page = Page ?? 1;
         _pageCount = Contents.Count / EssaysPerPage + 1;
+        (_pageCount, int reminder) = int.DivRem(Contents.Count, EssaysPerPage);
+        if (reminder > 0)
+        {
+            _pageCount += 1;
+        }
 
         if (EssaysPerPage * _page > Contents.Count + EssaysPerPage)
         {

YaeBlog/Services/BlogHotReloadService.cs

@@ -16,11 +16,11 @@ public sealed class BlogHotReloadService(
 
         await rendererService.RenderAsync(true);
 
-        Task[] reloadTasks = [FileWatchTask(stoppingToken)];
+        Task[] reloadTasks = [WatchFileAsync(stoppingToken)];
         await Task.WhenAll(reloadTasks);
     }
 
-    private async Task FileWatchTask(CancellationToken token)
+    private async Task WatchFileAsync(CancellationToken token)
     {
         while (!token.IsCancellationRequested)
         {
@@ -33,6 +33,15 @@ public sealed class BlogHotReloadService(
                 break;
             }
 
+            FileInfo changeFileInfo = new(changeFile);
+
+            if (changeFileInfo.Name.StartsWith('.'))
+            {
+                // Ignore dot-started file and directory.
+                logger.LogDebug("Ignore hidden file: {}.", changeFile);
+                continue;
+            }
+
             logger.LogInformation("{} changed, re-rendering.", changeFile);
             essayContentService.Clear();
             await rendererService.RenderAsync(true);

YaeBlog/Services/EssayScanService.cs

@@ -109,6 +109,12 @@ public partial class EssayScanService : IEssayScanService
         {
             foreach (BlogResult blog in fileContents)
             {
+                if (blog.BlogContent.Length < 4)
+                {
+                    // Even not contains a legal header.
+                    continue;
+                }
+
                 int endPos = blog.BlogContent.IndexOf("---", 4, StringComparison.Ordinal);
                 if (!blog.BlogContent.StartsWith("---") || endPos is -1 or 0)
                 {
@@ -121,14 +127,14 @@ public partial class EssayScanService : IEssayScanService
                 try
                 {
                     MarkdownMetadata metadata = _yamlDeserializer.Deserialize<MarkdownMetadata>(metadataString);
-                    _logger.LogDebug("Scan metadata title: '{}' for {}.", metadata.Title, blog.BlogFile.Name);
+                    _logger.LogDebug("Scan metadata title: '{title}' for {name}.", metadata.Title, blog.BlogFile.Name);
 
                     contents.Add(new BlogContent(blog.BlogFile, metadata, blog.BlogContent[(endPos + 3)..], isDraft,
                         blog.Images, blog.NotFoundImages));
                 }
                 catch (YamlException e)
                 {
-                    _logger.LogWarning("Failed to parser metadata from {} due to {}, skipping", blog.BlogFile.Name, e);
+                    _logger.LogWarning("Failed to parser metadata from {name} due to {exception}, skipping", blog.BlogFile.Name, e);
                 }
             }
         });

YaeBlog/source/drafts/hpc-2025-heterogeneous-quantum-computing.md

@@ -1,10 +0,0 @@
----
-title: High Performance Computing 25 SP Quantum Computing
-date: 2025-06-12T19:26:24.6668760+08:00
-tags: 
-  - 高性能计算
-  - 学习资料
----
-
-
-<!--more-->

YaeBlog/source/posts/hpc-2025-cuda.md

@@ -1,12 +1,13 @@
 ---
 title: High Performance Computing 25 SP NVIDIA
-date: 2025-04-24T19:02:36.1077330+08:00
+date: 2025-08-31T13:50:42.8639950+08:00
 tags:
 - 高性能计算
 - 学习资料
 ---
 
 
+
 Fxxk you, NVIDIA!
 
 <!--more-->

YaeBlog/source/posts/hpc-2025-non-stored-program-computing.md

@@ -1,11 +1,13 @@
 ---
-title: High Performance Computing 2025 SP Non Stored Program Computing
+title: High Performance Computing 25 SP Non Stored Program Computing
-date: 2025-05-29T18:29:28.6155560+08:00
+date: 2025-08-31T13:51:17.5260660+08:00
 tags:
 - 高性能计算
 - 学习资料
 ---
 
+
+
 No Von Neumann Machines.
 
 <!--more-->
@@ -60,7 +62,7 @@ There are two types of semi-custom ASICs:
 
 The Standard cell based ASICs is also called as **Cell-based ASIC(CBIC)**.
 
-![image-20250815093113115](./hpc-2025-non-stored-program-computing/image-20250815093113115.png)
+![image-20250815093113115](./hpc-2025-non-stored-program-computing/image-20250815093113115.webp)
 
 > The *gate* is used a unit to measure the ability of semiconductor to store logical elements.
 
@@ -84,7 +86,7 @@ Depending on the structure, the standard PLD can be divided into:
 - Programmable Logic Array(PLA): A programmable array of AND gates feeding a programmable of OR gates.
 - Complex Programmable Logic Device(CPLD) and Field Programmable Gate Array(FPGA): complex enough to be called as *architecture*.
 
-![image-20250817183832472](./hpc-2025-non-stored-program-computing/image-20250817183832472.png)
+![image-20250817183832472](./hpc-2025-non-stored-program-computing/image-20250817183832472.webp)
 
 
 
@@ -96,7 +98,7 @@ Depending on the structure, the standard PLD can be divided into:
 
 ### FPGA Architecture
 
-![image-20250817184419856](./hpc-2025-non-stored-program-computing/image-20250817184419856.png)
+![image-20250817184419856](./hpc-2025-non-stored-program-computing/image-20250817184419856.webp)
 
 #### Configurable Logic Block(CLB) Architecture
 
@@ -116,7 +118,7 @@ LUT is a ram with data width of 1 bit and the content is programmed at power up.
 
 The below figure shows LUT working:
 
-![image-20250817185111521](./hpc-2025-non-stored-program-computing/image-20250817185111521.png)
+![image-20250817185111521](./hpc-2025-non-stored-program-computing/image-20250817185111521.webp)
 
 The configuration memory holds the output of truth table entries, so that when the FPGA is restarting it will run with the same *program*.
 
@@ -126,7 +128,7 @@ And as the truth table entries are just bits, the program of FPGA is called as *
 
 Let the input signal as address, the LUT will be configured as a RAM. Normally, LUT mode performs read operations, the address decoders can generate clock signal to latches for writing operation.
 
-![image-20250817185859510](./hpc-2025-non-stored-program-computing/image-20250817185859510.png)
+![image-20250817185859510](./hpc-2025-non-stored-program-computing/image-20250817185859510.webp)
 
 #### Routing Architecture
 
@@ -134,7 +136,7 @@ The logic blocks are connected to each though programmable routing network. And
 
 Horizontal and vertical mesh or wire segments interconnection by programmable switches called programmable interconnect points(PIPs).
 
-![image-20250817192006784](./hpc-2025-non-stored-program-computing/image-20250817192006784.png)
+![image-20250817192006784](./hpc-2025-non-stored-program-computing/image-20250817192006784.webp)
 
 These PIPs are implemented using a transmission gate controlled by a memory bits from the configuration memory.
 
@@ -146,7 +148,7 @@ Several types of PIPs are used in the FPGA:
 - Non-decoded MUX: n wire segments each with a configuration bit.
 - Compound cross-point: 6 breakpoint PIPs and can isolate two isolated signal nets.
 
-![image-20250817194355228](./hpc-2025-non-stored-program-computing/image-20250817194355228.png)
+![image-20250817194355228](./hpc-2025-non-stored-program-computing/image-20250817194355228.webp)
 
 #### Input/Output Architecture
 
@@ -158,7 +160,7 @@ The programmable Input/Output cells consists of three parts:
 - Routing resources.
 - Programmable I/O voltage and current levels.
 
-![image-20250817195139631](./hpc-2025-non-stored-program-computing/image-20250817195139631.png)
+![image-20250817195139631](./hpc-2025-non-stored-program-computing/image-20250817195139631.webp)
 
 #### Fine-grained and Coarse-grained Architecture
 
@@ -186,9 +188,9 @@ Three types of interconnected devices have been commonly used to connect there w
 
 ### FPGA Design Flow
 
-![image-20250817195714935](./hpc-2025-non-stored-program-computing/image-20250817195714935.png)
+![image-20250817195714935](./hpc-2025-non-stored-program-computing/image-20250817195714935.webp)
 
-![image-20250817200350750](./hpc-2025-non-stored-program-computing/image-20250817200350750.png)
+![image-20250817200350750](./hpc-2025-non-stored-program-computing/image-20250817200350750.webp)
 
 The FPGA configuration techniques contains:
 
@@ -222,7 +224,7 @@ The OpenCL is not an traditional hardare description language. And OpenCL needs
 
 The follow figure shows how the OpenCL-FPGA compiler turns an vector adding function into the circuit.
 
-![image-20250829210329225](./hpc-2025-non-stored-program-computing/image-20250829210329225.png)
+![image-20250829210329225](./hpc-2025-non-stored-program-computing/image-20250829210329225.webp)
 
 The compiler generates three stages for this function:
 

YaeBlog/source/posts/hpc-2025-opencl.md

@@ -1,11 +1,12 @@
 ---
-title: High Performance Computing 2025 SP OpenCL Programming
+title: High Performance Computing 25 SP OpenCL Programming
-date: 2025-05-29T18:29:14.8444660+08:00
+date: 2025-08-31T13:51:02.0181970+08:00
 tags:
 - 高性能计算
 - 学习资料
 ---
 
+
 Open Computing Language.
 
 <!--more-->

YaeBlog/source/posts/hpc-2025-potpourri.md

@@ -1,11 +1,12 @@
 ---
 title: High Performance Computing 25 SP Potpourri
-date: 2025-06-12T18:45:49.2698190+08:00
+date: 2025-08-31T13:51:29.8809980+08:00
 tags:
 - 高性能计算
 - 学习资料
 ---
 
+
 Potpourri has a good taste.
 
 <!--more-->

YaeBlog/source/posts/hpc-2025-program-cuda.md

@@ -1,11 +1,12 @@
 ---
-title: High Performance Computing 2025 SP Programming CUDA
+title: High Performance Computing 25 SP Programming CUDA
-date: 2025-05-15T19:13:48.8893010+08:00
+date: 2025-08-31T13:50:53.6891520+08:00
 tags:
 - 高性能计算
 - 学习资料
 ---
 
+
 Compute Unified Device Architecture
 
 <!--more-->

YaeBlog/source/posts/podman-dns-failed.md

@@ -0,0 +1,130 @@
+---
+title: Podman容器中特定域名解析失败问题排查
+date: 2025-10-22T17:42:07.7738936+08:00
+tags:
+- 技术笔记
+- Bug Collections
+---
+
+最难以置信的一集。
+
+<!--more-->
+
+## 问题现象
+
+在折腾我的Gitea  CI/CD管线的时候我，我尝试进行一项重要的变更：在运行CI任务时，任务中的容器支持从Docker out of Docker（DoD）迁移到Podman inside Podman（PiP）。但是在执行该项迁移的过程中，我遇到了如下一个非常神秘的Bug。
+
+在CI任务运行的过程中，我发现一些**特定的**、**稳定的**域名无法在任务容器中正常解析并请求。目前注意到的域名有如下两个：
+
+- dot.net
+- rcj.bupt-hpc.cn
+
+在容器内使用`curl http://dot.net -vvvvi`时，日志中重复出现如下内容。
+
+```
+10:35:43.261007 [0-x] == Info: [MULTI] [INIT] added to multi, mid=1, running=1, total=2
+10:35:43.261252 [0-x] == Info: [MULTI] [INIT] multi_wait(fds=1, timeout=0) tinternal=0
+10:35:43.261533 [0-x] == Info: [MULTI] [INIT] -> [SETUP]
+10:35:43.261690 [0-x] == Info: [MULTI] [SETUP] -> [CONNECT]
+10:35:43.261838 [0-x] == Info: [READ] client_reset, clear readers
+10:35:43.262042 [0-0] == Info: [MULTI] [CONNECT] [CPOOL] added connection 0. The cache now contains 1 members
+10:35:43.262315 [0-0] == Info: [DNS] init threaded resolve of rcj.bupt-hpc.cn:9090
+10:35:43.262597 [0-0] == Info: [DNS] resolve thread started for of rcj.bupt-hpc.cn:9090
+10:35:43.262854 [0-0] == Info: [MULTI] [CONNECT] [TIMEOUT] set ASYNC_NAME to expire in 1000ns
+10:35:43.263081 [0-0] == Info: [MULTI] [CONNECT] -> [RESOLVING]
+10:35:43.263254 [0-0] == Info: [MULTI] [RESOLVING] multi_wait pollset[fd=4 IN], timeouts=1
+10:35:43.263500 [0-0] == Info: [MULTI] [RESOLVING] [TIMEOUT] ASYNC_NAME expires in 355ns
+10:35:43.263743 [0-0] == Info: [MULTI] [RESOLVING] multi_wait(fds=2, timeout=1) tinternal=1
+10:35:43.265041 [0-0] == Info: [DNS] Curl_resolv_check() -> 0, missing
+10:35:43.265218 [0-0] == Info: [MULTI] [RESOLVING] multi_wait pollset[fd=4 IN], timeouts=1
+10:35:43.265497 [0-0] == Info: [MULTI] [RESOLVING] [TIMEOUT] ASYNC_NAME expires in 1545ns
+10:35:43.265736 [0-0] == Info: [MULTI] [RESOLVING] multi_wait(fds=2, timeout=2) tinternal=2
+10:35:43.268052 [0-0] == Info: [DNS] Curl_resolv_check() -> 0, missing
+10:35:43.268236 [0-0] == Info: [MULTI] [RESOLVING] multi_wait pollset[fd=4 IN], timeouts=1
+10:35:43.268471 [0-0] == Info: [MULTI] [RESOLVING] [TIMEOUT] ASYNC_NAME expires in 3582ns
+10:35:43.268688 [0-0] == Info: [MULTI] [RESOLVING] multi_wait(fds=2, timeout=4) tinternal=4
+```
+
+## 初步假设与排除过程
+
+这个问题非常奇怪，在问题排查初期，我进行了如下的操作。
+
+1. 首先单独测试容器中的域名解析，使用`dig rcj.bupt-hpc.cn`进行解析测试，解析的结果符合预期，使用的DNS服务器是当前的容器的Podman网关地址10.17.14.1。
+2. 独立启动容器，而不是使用CI管线启动的容器，使用`podman run --rm -it --privileged ccr.ccs.tencentyun.com/jackfiled/runner-base:latest bash -l`创建容器，对上述出现问题的两个域名上的服务进行请求，没有问题，排除镜像本身的问题。
+
+此时已经束手无策了，遂尝试AI辅助进行排查。
+
+## AI辅助下的问题排查
+
+### dig工具和libc的解析机制之前存在差异
+
+AI首先提到，上面测试的dig工具和其他应用使用的libc解析机制之间存在差异：
+
+- `dig` 是直接使用 DNS 协议向 DNS 服务器发查询请求（绕过系统的解析器）。
+- `curl`（以及大多数应用）依赖系统 C 库（如 glibc）的 `getaddrinfo()` 函数，它会读取 `/etc/nsswitch.conf` 和 `/etc/resolv.conf`，并可能使用 `systemd-resolved`、`nscd` 等服务。
+
+于是需要使用`getent`工具测试libc库的DNS解析是否存在问题，测试发现解析结果正常。
+
+### IPv6 优先导致超时或失败
+
+AI还提到，如果系统尝试先用 IPv6 解析（AAAA 记录），但网络不支持 IPv6，可能导致 `curl` 卡住或失败，而 `dig` 默认查 A 记录。
+
+于是在使用`curl`时手动指定使用的网络协议，的确当使用`curl -4 http://dot.net`时可以正常的完成请求。
+
+虽然这一方法可以解决问题，但是其给出的理由似乎无法解释我遇到的问题，问题域名中的rcj.bupt-hpc.cn并没有任何的AAAA解析记录，执行 `getent ahosts rcj.bupt-hpc.cn` 仅返回 IPv4 地址，无 AAAA 记录，说明系统解析器未返回 IPv6 地址，`curl -4` 的成功不能用IPv6解释。
+
+### DNS服务器
+
+在尝试AI给出的几个建议之后，我也开始尝试修改容器内的DNS服务器，进行这个尝试是因为我发现，使用`podman run`启动容器时，容器内部使用的DNS服务器是10.3.9.4这一主机使用的DNS服务器，而在CI管线中启动的容器使用的DNS服务器是172.17.14.1这一容器网关。
+
+于是在CI启动的容器中，我通过直接写入`/etc/resolv.conf`文件的方式修改DNS服务器为10.3.9.4，在修改之后，针对两个域名的网络请求就恢复了正常。
+
+在将上述的新发现同步给AI之后，AI又给出了一些新的思路。
+
+### c-ares 异步 DNS 解析器兼容性问题？ 
+
+- 某些编译版本的 `curl`（尤其是 Alpine、某些容器镜像）会链接 `c-ares` 库，**绕过 glibc 的 `getaddrinfo`**，直接读取 `/etc/resolv.conf` 并发 DNS 请求。
+- `c-ares` **不读取 `/etc/nsswitch.conf`**，也不支持 mDNS、systemd-resolved 等机制。
+- 如果 `/etc/resolv.conf` 中的 nameserver（如 `172.17.14.1`）对 **某些域名响应异常**（如超时、截断、不响应 UDP 包），`c-ares` 会卡住。
+
+但是按照AI提供的步骤进行排查：
+
+- `curl --version` 输出中**未包含 `c-ares`**；
+- curl 使用 glibc 的 `getaddrinfo()` 或 libcurl threaded resolver；
+- **结论**：非 c-ares 导致。
+
+### UDP DNS 响应截断（>512 字节）？
+
+- `c-ares` 默认使用 **UDP** 查询 DNS。
+- 如果 Podman 的内置 DNS 代理（通常是 `netavark` + `aardvark-dns` 或旧版 `dnsmasq`）在处理 **特定域名** 的 UDP 请求时：
+  - 响应过大导致 UDP 截断（TC=1），但 `c-ares` 未自动重试 TCP；
+  - 或代理本身 bug 导致丢包；
+- 则 `c-ares` 会认为“无响应”，返回“missing”。
+
+而 `dig` 默认使用 UDP，但**当 TC=1时会自动使用TCP重试**，所以成功，而你的域名 `rcj.bupt-hpc.cn` 可能有较多 CNAME 或 TXT 记录，导致 UDP 响应超 512 字节。 
+
+但是我清楚的知道域名 `rcj.bupt-hpc.cn` 仅有一条 A 记录，DNS 响应极小，`dig` 与 `getent` 均能快速返回结果。
+
+## AI Conclusion：Podman 内置 DNS 代理（aardvark-dns）行为异常
+
+Podman 4.5.2 默认使用 **aardvark-dns** 作为容器网络的 DNS 代理，监听于网关地址（如 `172.10.0.1`）。其职责包括：
+
+- 解析容器名称（服务发现）；
+- 转发外部域名查询至上层 DNS。
+
+可能的故障点包括：
+
+1. aardvark-dns 对特定域名后缀（如 `.bupt-hpc.cn`）的转发逻辑存在缺陷；
+2. 响应格式不符合 glibc 或 libcurl 的预期（如缺少 AA 标志、TTL 异常）；
+3. 首次查询存在延迟或丢包，而 libcurl threaded resolver 未充分重试；
+4. 与宿主机 `/etc/resolv.conf` 中的 `search` 域交互异常，导致拼接错误查询。
+
+尽管 `dig` 能获取结果，但 `dig` 具有更强的容错性和自动重试机制（如 TC=1 时切 TCP），而 libcurl 的 resolver 更严格。 
+
+> 对于AI的这个结果，说实话我表示怀疑，我感觉还是和IPv6有关系，这样才能同时解释`curl -4`和更换DNS都可以运行的现象。
+
+## 临时缓解
+
+最终，我在CI启动容器的配置选项中添加`--dns 10.3.9.4`，不使用Podman启动的DNS服务器，暂时绕过了这个问题。
+
+不过这个问题并没有从本质上得到解决，甚至都还不知道背后的具体问题是什么，感觉会在后面攻击我，特此记录。