ntzyz's space

Four months ago I brought an HP Z620 workstation to run virtual machines at home with Hyper-V Server. Some days later I planned to replace the SSD in the workstation from Samsung 860EVO 500G to Intel DC P4510 2TB for more capacity and better performance. But after hardware upgrade I found that P4510 just didn't get recognize by the boot loader. It seemed that I need to do some more work to boot ESXi from this NVMe SSD.

The simplest solution is using Clover Boot Loader as secondary boot loader. But the automatic OS scans provided by Clover does not support VMware ESXi, meanwhile creating a custom entry is quite bother job as writing XML manually by human is just another disaster.

After some simple research I reached out that unified extensible firmware interface itself dose support load driver before OS boot, so EFI Shell with a customized startup script may solve this problem for me. Firstly create a ESP partition on one disk with GUID partition table which is recognizable by the stock firmware, copy UEFI Shell binary as well as the NVMe driver binary to that partition, then create startup.nsh in that partition with following script:

load -nc fs0:\NvmExpressDxe-64.efi
connect -r
map -u
fs1:\EFI\BOOT\BOOTX64.EFI

The script is quite easy to understand. But remember to change the locations and names of these binaries to your own.

After that, use efibootmgr under Linux or some utilities under Windows PE to create a new EFI boot entry with EFI shell as the loader, and then set it to default entry. Note that the disk alias may be different on each machine, to find out your own aliases just boot into the efi shell, load the NVMe driver and use the map command to show partition aliases.

May be the simplest solution is just install ESXi to one HDD and place all virtual disks in the NVMe SSD?

References

1. [Guide] NVMe-boot without modding your UEFI/BIOS (Clover-EFI bootloader method)
2. UEFI Shell Specification
3. Error-handling in EFI startup shell scripts - Super User

其实这个叫 GitHub Actions 的持续集成已经上线很久了，在刚刚公开测试时我就尝试去使用，结果翻来翻去没找到 MongoDB 数据库的项目应该如何配置基础环境 不要问我当时为什么用了 MongoDB，我现在非常后悔（雾）。今天摸鱼的时候突然想到 Actions 支持 Docker，Docker 又能快速的启动一个 MongoDB 官方提供的容器，那么这就好办了，直接对着自己的博客下手，来试一试这个工具。

在仓库首页里点击 Actions 标签进入配置页面，并直接创建一个新的 Node CI 类型的 Workflow，这样就能获得一个适用于 node.js 项目的配置模版。和 Travis 一样，Actions 的模版配置也是使用 YAML 格式，写起来比 JSON 舒服多了。之后只需要在 steps 中添加以下内容即可启动一个 MongoDB 容器：

- name: Launch MongoDB
  run: |
    sudo docker container run --name mongo -v $PWD:/src -p 27017:27017 -d mongo
    sleep 4s
- name: Import initial database
  run: |
    sudo docker container ls --all
    sudo docker container exec mongo mongo localhost/newBlog /src/db.default.js

人挺多的，手办也挺多的，好看的手办也有不少，只可惜拙劣的拍照技巧让不少照片都非常失败：没对上焦、手抖、光圈太大……

所使用的器材全部是索尼 A73 + 索尼蔡司 FE55。

这里就随便丢几张自己觉得说的过去的吧。页面里的图片都是压缩到 1600 像素长边的，如果需要看原图（6000x4000），点击即可。

bilibili | 22 33 | ISO 100, 1/200s, f/3.5

假设某服务器有一个接口 eth0，并配置了三个 IP 地址 10.0.0.4, 10.0.0.5, 10.0.0.6。需求就是，如何知道这一台服务器每个 IP 的历史出网流量。

思路由 @a350063 提供，很简单，活用 iptables 的计数器就 OK 了。这种方法非常灵活，可以对任何能用 iptables 规则描述的流量数据进行统计，比如统计某个源地址指定端口的进出流量。首先我们使用 iptables 增加一些规则：

# iptables -I OUTPUT -s 10.0.0.4/32 -j ACCEPT
# iptables -I OUTPUT -s 10.0.0.5/32 -j ACCEPT
# iptables -I OUTPUT -s 10.0.0.6/32 -j ACCEPT

执行完成之后，就可以直接使用命令 iptables -nvxL OUTPUT 看到每个 IP 的出网流量了：

# iptables -nvxL OUTPUT
Chain OUTPUT (policy ACCEPT 3856 packets, 1914416 bytes)
    pkts      bytes target     prot opt in     out     source               destination         
   12880   11794276 ACCEPT     all  --  *      *       10.0.0.4             0.0.0.0/0           
    6850    5801012 ACCEPT     all  --  *      *       10.0.0.5             0.0.0.0/0           
     110       9315 ACCEPT     all  --  *      *       10.0.0.6             0.0.0.0/0