Summary

axios v1 has brought a few breaking changes, so now's the time to reconsider whether it still is the best solution we have.

Motivation(s)

• http2/http3 support
• streaming support
• reduce dependencies

Downsides

• Breaking change - axios is very tied to our responses, anyone who currently depends on them will have some work to do before they can upgrade

Alternatives to consider

• axios: it's working, stick to it, but update to a v1.x.x+ version (no http2 support yet)
• fetch: built into recent browsers and node 18+
• gaxios: close to the axios interface, supports http2, no browser support
• got: Seems to satisfy a lot constraints. No browser support and has a lot of dependencies
• ky: tiny, no dependencies, supports deno
• undici: from the core node devs, no browser support but much better performance
• ...

Criteria for success

• Compatible license (required)
• Node.js 18+ support (required)
• Brings in a minimal set of dependencies (required)
• Promise based (required)
• Streaming support: allows us to expose the streaming ListObjects
• http2/http3 support: reduces request overhead
• Deno support (Support for Deno #17)
• Cloudflare Worker and Vercel support (Support for Cloudflare Workers/Vercel Edge Runtime #72)
• Better support for proxies (https://github.com/orgs/openfga/discussions/310)
• Browser support: needed for the Playground
• Preferably not another dependency (bonus)
• Error handling and automated retries (bonus - lets us drop our own implementation)
• Credential flow support (bonus - lets us drop our own implementation)
• Built-in caching support (bonus)

State (as of June 2024)

We're currently favouring undici and planning on starting work on this soon