Pete Vilter

Nice. Are those datalog-based?

React and TypeScript, with a REST API to a NodeJS backend where DuckDB is running.

My initial prompt to Claude was something like "I want to explore DuckDB data in a UI that looks like DataDog, with React and TypeScript" etc

and of course the classics, 'filter to this value' and 'sort asc/desc'.

I lose my mind when table UIs don't have 'filter to this value' 😂

you can also filter by clicking on the histograms, inspired by Datadog's log search UI

Used Claude Code to vibecode a UI for @duckdb.org in a few days. Here it is navigating foreign keys

are tarriffs going to send prices back up?

Datadog up next?

A context-aware LLM agent built directly into Grafana Cloud: Introducing Grafana Assistant | Grafana Labs With our new integrated AI agent you can write in natural language to ask observability questions, see certain data, build dashboards, and make changes—all without leaving the Grafana UI.

Grafana has one too grafana.com/blog/2025/05...

It's The End Of Observability As We Know It (And I Feel Fine) The history of observability tools over the past decade has been about a pretty simple concept, but LLMs bring the death of that paradigm.

Neat to see LLMs ingesting observability data and generating explanations www.honeycomb.io/blog/its-the...

@peet.io I see you added support for taking profiles for JS a while ago: github.com/microsoft/vs... very cool!

Think there's a chance of getting this upstreamed to the main debug adapter protocol so other languages can implement it?

whoops, that lock implementation had a bug — forgot to set holder back to null, messing up next acquire. now fixed gist.github.com/vilterp/bf02...

bank.ts GitHub Gist: instantly share code, notes, and snippets.

whoops, repo is still private. source here gist.github.com/vilterp/bf02...

bank source here: github.com/vilterp/cons...

try here: dist-sys-simulator.netlify.app

If we make the client grab a lock around reading & writing the new balance, it's impossible to get a double spend:

the first client to acquire the lock wins; the next sees that there's not enough in the account.

Distributed systems should feel like physical puzzles we can manipulate!

As an example, I modeled a bank in my simulator.

Without a lock, there's a race condition: if the two clients' gets and puts of the balance are interleaved, the same money can be withdrawn twice.

I think the closest experience to what I'm looking for is Apple's Instruments or even the Chrome/Safari devtools, but a standardized protocol could really help the ecosystem flourish.

Tools like py-spy and memray have built fancy UIs for profile inspection, but each language/tool shouldn't have to.

Could also be interesting to attach to running servers in the cloud or continuous profiling services.

And VSCode would be free to visualize the data however it sees fit, e.g. in the debugger UI or by annotating source lines with a heatmap.

As with LSP and DAP, each language runtime would be free to do it its own way.

Different profile types, like CPU, memory allocation, contention, and blocking could be supported.

Under the hood, this could look like VSCode saying to the process "profile yourself please" over the debugger protocol, and getting a stream of stack frames in response, each tagged with a thread ID.

Perhaps it could use the OTel profiling standard as inspiration: github.com/open-telemet...

I wish @vscode.dev had richer & more standardized support for profiling; maybe a "Profiler Adapter Protocol" along the lines of the Debug Adapter Protocol and LSP.

When running a multithreaded program under the debugger, I want to see live CPU usage for each thread, and click to get a flame graph.

Each 'actor' is defined as a TypeScript function, which runs in the browser with simulated network latency. Messages can be 'sent' between them with async/await syntax.

It's hard to wrap your head around all the failure modes of a distributed system.

What if you could play with the sequence diagram of a run to see how it responds, and see if you can get it into a failure state?

dist-sys-simulator.netlify.app

Thank you! Curious what your vision looks like

Little script to make a flamegraph of a codebase by lines of code github.com/vilterp/cloc...

actor-trace.netlify.app#?systemInsta...

Transparent means that the order hasn't made it to the server yet.

Am I becoming a fintech person? 😂

Added a simulation of a commodity market to my distributed system simulator.

Clients submit buy and sell orders (blue/right and orange/left, respectively); the server uses DB triggers to execute a trade when there's a match.