@thomas_ankcorn The best part was - when I asked how you tested your code (not if), fully expecting you to give me BS, you just said “we don’t” - instant trust.
George - Are you pushing messages into the Queue via Workers bindings or using the HTTP API? Are you sure that in both cases you are capturing failures properly? Keep in mind that with bindings errors are propagated via exceptions and you need to catch them.
We have zero tolerance for data loss and take these sort of issues seriously. Usually the issues are with failing to capture insertion errors. Can you DM me with your support ticket and account information?
It depends on what the symptoms of your bug are. Is it throwing an exception? Logic error? Something else?
What would you reach for if you had this exact same issue on any other platform?
Workers have production logs, tails, local dev with chrome dev tools attachment, remote dev with chrome dev tools attachment
How would you debug this in AWS Lambda or similar ?
ok, so. dynamic workers.
when LLMs are generating a ton of code, you need something better to run it in.
imagine eval, but better. way better. control what's exposed to the env. control outgoing fetches. zero cold start.
left: code in a string. right: output.
https://t.co/qS07TNyThb
OK folks, today is the day when I reveal what this tweet was about.
We made MCP work better by:
1. Convert the MCP tools into a TypeScript API
2. Tell the agent to write code against this API
We call it "code mode"
It uses a new feature in Workers: Dynamic Worker Loading
Super excited to see @Cloudflare launching v1 of their @vite_js plugin, with first class React Router v7 support.
Huge shout out to James Opstad for working with us to ensure support on top of our recent Vite Environment API work.
https://t.co/iI4Zpi1tH2
Has anyone else shipped a production-ready remote MCP yet or is Sentry the first?
Was pretty easy to do on CF so I gotta imagine we're not alone. Curious how others have approached the UX.
🍕 SQLite-on-the-server isn't just for hobby projects - it's actually better at hyperscale than microscale
With @CloudflareDev Durable Objects & @tursodatabase, SQLite-per-partition is the future of distributed databases at scale
Blog post below 📝
@ilias_yahia@CloudflareDev I'm sorry this is confusing! We use GB-seconds to make it 'easy' to compare to other platforms (such as AWS Lambda) that have been using these units for much longer.
A GB-second is the amount of memory you are using (128MB) multiplied by the amount of time your worker is active
🎈 PartyServer: Build real-time apps on Workers
I'm super happy to announce PartyServer (beta), a new way of building real-time apps on Workers. Based on everything I've learnt from building PartyKit, with a new and ambitious roadmap. Today's Day 1.
https://t.co/w3w7SxnwKS
🧵
🎈 PartyServer: Build real-time apps on Workers
I'm super happy to announce PartyServer (beta), a new way of building real-time apps on Workers. Based on everything I've learnt from building PartyKit, with a new and ambitious roadmap. Today's Day 1.
https://t.co/w3w7SxnwKS
🧵