norgopolis-server

Crates.ionorgopolis-server
lib.rsnorgopolis-server
version0.1.4
sourcesrc
created_at2023-07-08 18:59:45.485372
updated_at2024-02-21 16:28:13.73629
descriptionA backend server for Norg related applications.
homepage
repositoryhttps://github.com/nvim-neorg/norgopolis
max_upload_size
id911624
size63,529
(vhyrro)

documentation

README

Norgopolis

(Client Implementation / Module Implementation)

Norgopolis is a lightweight communication, launcher and utility services client for the Neorg rust-native modules ecosystem on Desktop. It is designed to connect, launch and manage instances of modules such as the Norgberg database system which should only exist once per desktop, but provide services to n frontends such as multiple neovim processes, without the need for intermediate process ownership changes or process re-launches.

Norgopolis also provides for the ability to download and connect different services together without the need for recompilation or complete re-launch, making for easy modifying and extension by the common user.

Why?

There are four problems that Norgopolis solves:

  • How do we provide singleton services like a database-over-notes while having more than one frontend at the same time?

  • How do we ensure that the exit of the first-launching frontend (which must initiate all the background services) does not cause loss of service, interruption of service, failure to handover to another thread, or similar issues?

    • How do we ensure that the processes exit when the last frontend has indeed also exited?
  • How do we allow people to extend their installation with new native modules without the need for recompilation or volatile shared object dynamic loading?

  • How do we get these modules to talk with each other and any frontend callees in a lightweight, performant manner?

Norgopolis addresses these.

How it works

Norgopolis comes on two parts, the client and the server backend. They communicate with each other using gRPC.

When a client is launched, it tries to find an existing server instance. If it does not find an existing instance, it will initiate a server boot as an independent process. If it finds an existing server, it will connect to that server and log on as a client session.

The Norgopolis server reference-counts registered clients. If clients die or time out with no communication for a certain duration, they are removed from the reference count. If the reference count hits zero, Norgopolis will gracefully exit itself and all its child processes.

The core of Norgopolis is a communications router between the frontend gRPC connection(s) and the set of currently loaded modules. "Modules" are executable binaries that conform to the Neorg native module standard. The standard defines a protobuffer gRPC interface over a stdin/stdout-based communication channel. Modules must be able to accept a method invocation call and on startup provide a message declaring their name, semantic version number and potential dependencies. Norgberg is an example of such a module, providing database services.

Related to the router is Norgopolis' loader capability. Norgopolis routes calls using a module name specified in the gRPC protocol buffer data. If no such module is currently registered on the router, Norgopolis will attempt to locate the binary in a given directory and spawn it as a child process. This allows for inherent lazy loading. Modules may declare transient module dependencies in their startup information structure, in which case the Norgobolis loader will verify availability and fail the load if dependencies are not availiable. Likewise, inabililty to locate an executable binary will fail gracefully.

Modules only have to be compiled binaries or otherwise executable programs. They run as independent child processes of Norgopolis. Note: that means if Norgopolis exists, so will these child processes. This is by design and should be bypassed sparingly. Norgopolis will broadcast shutdown gRPC messages when it exits.

This also describes the utilities Norgopolis offers: they are related to loading, managing and shutting down modules.

Client interface

The client exposes Rust functions and an interface aimed at Lua, and comes as a separate crate. This is used to bridge method calls over into gRPC and execute the stateful startup-registration logic for the server, as well as unpack and deliver method invocation return values.

Router interface

The router talks with any clients or other external callers which are not modules it maintains over gRPC, using the Norgopolis protobuffer spec. Note, this does mean you can try to call the server (if it is online) and through it any modules you may want, from any other application with the correct protocol buffer and mpack specification, such as web applet plugins or one-off scripts.

Modules talk to the router and through it with callers and other modules using stdin/stdout-piped protobuffers that follow the same specification. The server provides the asynchronous routing.

Norgopolis Secure Launch

As a general policy, Norgopolis is designed to serve n frontend clients at the same time. This has inherent security implications since Norgopolis will by default not refuse connections from new clients that connect. If more security is desired, Norgopolis can be launched in single-connection mode as the child of a single frontend. This isolates communication and callability from outside, but will cause any further client launches to fail.

[!NOTE] This behaviour is currently unimplemented.

Neat hacks and implications

Since Norgopolis launches executable files as child processes and communicates using protocol buffers over stdin/stdout, any executable that can talk according to our spec can be loaded as a module. This not only inoculates against Rust ABI changes between versions, but also allows modules compiled in other languages (or interpreted on the spot) to be loaded up into your system. Be however aware of the performance implications, and modules are generally understood to be persistent, not transient, until unloaded with a command to the server or the server exits, and thus also all module threads. (That said, you can exit as a module yourself. Again, just beware of the implications.)

Self-loading

The ability to load new modules is limited in Norgopolis by security design. However, if is possible to load modules on-demand by placing the module executive in the loading folder and then placing a method invocation call to that module on the gRPC connection as a client. This will initiate loading and execution. This can be used to execute larger batch processing-type activities from external scripts, using the external script as the initiator and controller and your custom module as the inner worker on the Norgopolis system.

Default lazy loading

Norgopolis default will only try to load a module if a call is received and it does not have an active thread of that module name to route to. This implicit lazy loading can be used to reduce startup times. Norgopolis has a loading capability to automate the startup of certain services in part due to its transient dependency checking, but this builds on the lazy loading behavior.

Commit count: 36

cargo fmt