OS signals & PID

[sadoyan]

@Taqman-probe

I know you are working on Windows fork of Aralez and that's great . Thank you for that.
In current dev branch I have implemented honoring of OS signals and writing PID file (hardcoded as /tmp/aralez.pid for now )
This is for graceful upgrades with zero session loss during it . So you can upgrade the binary on the fly and no one will even notice, that something happened.

For that I wrote some code, which may be very specific to Linux. Can you please review it and share your thoughts about implementing the same for Windows as well ?

Related changes are:

• start.rs lines 116 -> 135
• webserver.rs lines 78 -> 96

For that I have added sd-notify dependency, should be pure for Linux.

Please share your thought .
Thanks !

[Taqman-probe]

@sadoyan

Thank you for reaching out! I'm thrilled to see the new graceful upgrade features in the dev branch.
This is a huge milestone for Aralez on Linux!

Regarding the Windows support, here is my review and thoughts on how we can handle these Linux-specific changes:

OS signals

Windows does not support SIGINT, SIGTERM, or SIGQUIT; instead, it uses CTRL_C_EVENT, which is equivalent to SIGINT.
When you actually press CTRL+C, pingora outputs the following log on Windows:

INFO pingora_core::server: Ctrl+C received, gracefully exiting
INFO pingora_core::server: Broadcasting graceful shutdown

Even though CTRL+C was sent, pingora behaves as if it received SIGTERM. In other words, on Windows, at a minimum, we must specify the following:

grace_period_seconds: 0 // In Windows environments where a graceful shutdown is not supported, this should be set to 0.

This is the best we can do to produce consistent logs like the following:

INFO pingora_core::server: Graceful shutdown: grace period 0s starts
INFO aralez::web::webserver: Exiting...
INFO pingora_core::server: Graceful shutdown: grace period ends

PID file

You might want to set the output directory for PID files to std::env::temp_dir() using the #[cfg(windows)] directive.
While this isn't required on Windows, there are situations where it can be helpful.

Graceful upgrades & sd_notify

To be honest, no one expects "live graceful upgrades" on Windows, so please don't worry about it at all!
On Windows, the OS completely locks the running .exe binary, making it impossible to overwrite the file on the fly anyway.

Windows does not use systemd; instead, it uses Windows Services (Service Control Manager). Registration with Windows Services is handled by dedicated applications such as NSSM or Shawl (also a Rust project), so there is no need to write any additional code in individual applications.

So, the best and cleanest approach is to completely disable the live upgrade code on Windows using #[cfg(unix)] blocks, and let Windows handle a simple single-process lifecycle.

To be completely honest, I’m actually a total beginner when it comes to running production reverse proxies. I just happened to find Aralez while evaluating tools for our internal company applications, and I've been experimenting with it on Windows.

So I might not know all the best practices, but from my testing on Windows, here is my review and thoughts...

[Taqman-probe]

I've added Windows support to the latest commit by sadoyan/aralez.
https://github.com/Taqman-probe/aralez/tree/main