Skip to content

Latest commit

 

History

History
60 lines (40 loc) · 5.48 KB

customizing.md

File metadata and controls

60 lines (40 loc) · 5.48 KB
Raw

Customizing Kubo

You may want to customize Kubo if you want to reuse most of Kubo's machinery. This document discusses some approaches you may consider for customizing Kubo, and their tradeoffs.

Some common use cases for customizing Kubo include:

  • Using a custom datastore for storing blocks, pins, or other Kubo metadata
  • Adding a custom data transfer protocol into Kubo
  • Customizing Kubo internals, such as adding allowlist/blocklist functionality to Bitswap
  • Adding new commands, interfaces, functionality, etc. to Kubo while reusing the libp2p swarm
  • Building on top of Kubo's configuration and config migration functionality

Summary

This table summarizes the tradeoffs between the approaches below:

Boxo Kubo Plugin Bespoke Extension Point Go Plugin Fork
Supported? ✅ ✅ ✅ ❌ ❌
Future-proof? ✅ ❌ ✅ ❌ ❌
Fully customizable? ✅ ✅ ❌ ✅ ✅
Fast to implement? ❌ ✅ ✅ ✅ ✅
Dynamic at runtime? ❌ ❌ ✅ ✅ ❌
Add new commands? ❌ ✅ ❌ ✅ ✅

Boxo: build your own binary

The best way to reuse Kubo functionality is to pick the functionality you need directly from Boxo and compile your own binary.

Boxo's raison d'etre is to be an IPFS component toolbox to support building custom-made implementations and applications. If your use case is not easy to implement with Boxo, you may want to consider adding whatever functionality is needed to Boxo instead of customizing Kubo, so that the community can benefit. If you are interested in this option, please reach out to Boxo maintainers, who will be happy to help you scope & plan the work. See Boxo's FAQ for more info.

Kubo Plugins

Kubo plugins are a set of interfaces that may be implemented and injected into Kubo. Generally you should recompile the Kubo binary with your plugins added. A popular example of a Kubo plugin is go-ds-s3, which can be used to store blocks in Amazon S3.

Some plugins, such as the fx plugin, allow deep customization of Kubo internals. As a result, Kubo maintainers can't guarantee backwards compatibility with these, so you may need to adapt to breaking changes when upgrading to new Kubo versions.

For more information about the different types of Kubo plugins, see plugins.md.

Kubo plugins can also be injected at runtime using Go plugins (see below), but these are hard to use and not well supported by Go, so we don't recommend them.

Bespoke Extension Points

Certain Kubo functionality may have their own extension points. For example:

  • Kubo supports the Routing v1 API for delegating content routing to external processes
  • Kubo supports the Pinning Service API for delegating pinning to external processes
  • Kubo supports DNSLink for delegating name->CID mappings to DNS

(This list is not exhaustive.)

These can generally be developed and deployed as sidecars (or full external services) without modifying the Kubo binary.

Go Plugins

Go provides dynamic plugins which can be loaded at runtime into a Go binary.

Kubo currently works with Go plugins. But using Go plugins requires that you compile the plugin using the exact same version of the Go toolchain with the same configuration (build flags, environment variables, etc.). As a result, you likely need to build Kubo and the plugins together at the same time, and at that point you may as well just compile the functionality directly into Kubo and avoid Go plugins.

As a result, we don't recommend using Go plugins, and are likely to remove them in a future release of Kubo.

Fork Kubo

The "nuclear option" is to fork Kubo into your own repo, make your changes, and periodically sync your repo with the Kubo repo. This can be a good option if your changes are significant and you can commit to keeping your repo in sync with Kubo.

Kubo maintainers can't make any backwards compatibility guarantees about Kubo internals, so by choosing this option you're accepting the risk that you may need to spend more time adapting to breaking changes.