←
Bartłomiej Płotka7,533 posts
Bartłomiej Płotka avatar

Bartłomiej Płotka

@bwplotka

Senior SWE TL @Google | ex Principal @RedHat | @ThanosMetrics @PrometheusIO | Bsky: bsky.app/profile/bwplotka.de… | Efficient Go 📖 @OReillyMedia

◷ Joined October 2015▧ 1,074 Photos and videos
Bartłomiej Płotka avatar
Bartłomiej Płotka@bwplotka·19 Nov 2022

My "Efficient Go" book with @OReillyMedia has been released!🎉 I wrote some words about the book, content and generally my experience writing it (with photos!): bwplotka.dev/book Good for a slow Saturday--maybe it will help you to decide to write your own book? 🤗

[NEW RELEASE] Efficient Go -- Author @bwplotka provides the tools and knowledge required to make your systems faster and less resource-hungry. oreil.ly/SyTuj #OReillyExperts

Varun Mohan avatar
⟳ Bartłomiej Płotka retweeted
Varun Mohan@_mohansolo·15h

Time to first token (TTFT) for Carbon is now 1.5x faster on Antigravity. Will follow soon with Argon too. Lots of work went into this so you all could have a faster experience. More to come 👀

Tomás Senart avatar
⟳ Bartłomiej Płotka retweeted
Tomás Senart@tsenart·Nov 15

CoreDNS merged its first Perfloop PR. CoreDNS is the default DNS server in Kubernetes. Its cache can remember a failed lookup for a set time, so a struggling upstream is not asked again straight away. The setting accepts fractions of a second, but the cache stored whole seconds. Set it to 500ms and the failure expired at once. Set it to 1.5s and you got one second. Perfloop found it, filed the issue and opened the fix with a test on a controlled clock. Thanks Yong Tang for the merge.

0:17
Bartłomiej Płotka avatar
Bartłomiej Płotka@bwplotka·Nov 15

Many asked about future of @PrometheusIO given the overlap and rise of @opentelemetry The beauty is that @PrometheusIO and @opentelemetry actually don't compete here, they supplement each other. My intention (and prediction) is clear: 1. @PrometheusIO as ecosystem will be the best, most efficient, pull based OSS metric full solution including collection, but also storage and querying, etc. Both monolith and distributed system (with @ThanosMetrics 2.0) 2. Otel-arrow will become the most efficient Otel collector for multiple signals and processors. More capable collection but the same efficiency as Prometheus (see why on 3). 3. New Prometheus(.rs) agent will literally import Otel arrow Rust code, but will simplify the pipeline and ensure Prometheus compatible config. Thoughts? (:

Tomás Senart avatar
⟳ Bartłomiej Płotka retweeted
Tomás Senart@tsenart·Nov 14

In September I made a chat thread my VP of Engineering. It built an org of twenty agent threads. In October it audited the bureaucracy it had created, handed off its work and fired itself. That took 47 minutes. Last week the org that replaced it merged 1,119 pull requests.

Post attachment
Bartłomiej Płotka avatar
Bartłomiej Płotka@bwplotka·Nov 12

Links & benchmarks: 📊 Full benchmark analysis: github.com/bwplotka/prombenc… 🦀 Prometheus scrape.rs crates: github.com/prometheus/scrape… 🏹 OpenTelemetry Arrow (otel-arrow): github.com/open-telemetry/ot…

Bartłomiej Płotka avatar
Bartłomiej Płotka@bwplotka·Nov 12

This is huge. Probably the biggest realization for open-source @CloudNativeFdn metrics this year. 🔥 Scraping 345,000 active @PrometheusIO series every 15s and sending via Remote Write 2 or @opentelemetry OTLP takes on modern @golang collectors: • ~1 GiB of RAM (WSS) with ups & downs (WAL compaction, cardinality bumps) 🚀 With @rustlang and @ApacheArrow via @opentelemetry otel-arrow using our new Prometheus scrape.rs crates: • ~30 MiB of RAM (WSS, 5 MiB live heap) • 2.6x–7.3x less CPU, flat for 23 hours 😱 I couldn't believe it. How is this even physically possible? We have an opportunity to save $2.5B–$4B/yr in collector compute globally, and easily $100M+/yr at single-hyperscaler scale. We can scale 30x per pod, get hardware back, and deprioritize complex scalability band-aids like DaemonSet VPA and sharding. Yes, I wrote the @OReillyMedia "Efficient Go" book, but for the last 2 weeks I've been becoming a Rust and columnar developer. There is no other way. This is the way. And yes, we are starting a Prometheus.rs (ecosystem-wide) SIG — see #prometheus-rs-sig on CNCF Slack! Huge thanks to the OpenTelemetry otel-arrow community for doing most of the work, Laurent Quérel for consulting, and the @ThanosMetrics & @PrometheusIO communities for columnar inspirations 💪 (Disclaimer: early numbers without full parity yet—with staleness, relabeling, and K8s discovery, compute needs will grow. Still, starting at 30 MiB WSS gives us tons of room!)

Post attachment
Bartłomiej Płotka avatar
Bartłomiej Płotka@bwplotka·Nov 12

Google Jev! (:

Today we’re launching the Gemini Decisions API, powered by our new Gemini Decision 1 model in @GoogleAIStudio. Instead of generating text token by token, it’s built for closed-set choices scoring up to 1,000 questions across text, images, audio, video + PDFs in a single pass. A few details: - 3 decision types: yes/no, multiple choice (up to 255 options) + rubric scoring - $0.10 / 1M input tokens with $0 billed for output tokens (~5x cheaper than Flash-Lite) - leads on several benchmark evals with lowest calibration error Google teams are already building with it, below is a look at Stitch Rapid designing a full product flow in seconds (1x speed)

0:13
Ahmed avatar
⟳ Bartłomiej Płotka retweeted
Ahmed@ahyield·Nov 9

Please read this, your Claude code gonna thank me later

Post attachment
Mitchell Hashimoto avatar
⟳ Bartłomiej Płotka retweeted
Mitchell Hashimoto@mitchellh·Nov 6

libghostty 1.0 is here! I started Ghostty ~5yrs ago to work towards this. Today, libghostty is in hundreds of projects used by tens of millions of people. 1.0 marks ABI stability, quality, and an invite for distros to begin packaging. mitchellh.com/writing/libgho… Since starting the Ghostty project in 2022, the most common question I've been asked is "why?". Today, I'll answer this by showing the broad industry impact libghostty has already had. On top of JetBrains and Neovim already mentioned (that alone being over 10 million users): -KDE developers are experimenting with a Konsole replacement built on it. - Vercel's Turborepo uses it to render task output in its TUI. - devenv uses it to drive shells. - Antithesis used it to fuzz TUI programs and found real bugs in btop and others. -Ghostel brings it to Emacs. - ghostty-web is the WASM build with an Xterm.js-compatible API. - Commercial products like OrbStack embed it. - And a whole wave of new terminals and "agentic" coding tools, such as herdr and cmux, have been built on it rather than writing yet another emulator. There are community bindings for Rust, Go, JavaScript, Swift, Python, .NET, Elixir, Dart, Java, and more, and the awesome-libghostty list tracks well over a hundred projects. More people today use a terminal powered by libghostty than use Ghostty the application, by a very large multiple. And libghostty is having a much larger impact than being "just" a terminal emulator. I'm excited to see where people take it from here.

Bartłomiej Płotka avatar
Bartłomiej Płotka@bwplotka·Nov 5

Any @PrometheusIO and SRE expert would like to join us at @Google in London under an amazing (I can promise that!) manager? 💪 google.com/about/careers/app…

Bartłomiej Płotka avatar
Bartłomiej Płotka@bwplotka·Nov 4

Go (vtprotobuf + snappy) vs Rust (prost + snap) results: • Varints: Rust 43% faster (LLVM unrolls shifts) • RW2: 18% faster ns/op & 15% fewer allocs, but +31% B/op (Vec doubling) • Full binary: ~4x less RSS once Go GC kicks in github.com/bwplotka/benchfmt… #rustlang #golang

Bartłomiej Płotka avatar
Bartłomiej Płotka@bwplotka·Nov 3

I love when #opensource investments pan out! I’ve built many OSS projects over the years—only a few ever get real adoption. Seeing "mdox" still useful and growing a community 5y after its last release is an amazing feeling! Just cut v0.10.0! ❤️ github.com/bwplotka/mdox/rel…

Tomás Senart avatar
⟳ Bartłomiej Płotka retweeted
Tomás Senart@tsenart·Nov 2

Prometheus merged another Perfloop PR this morning. PromQL range functions like predict_linear and quantile_over_time can take scalar arguments. For every argument, on every series, at every step, the evaluator built a fresh one-sample vector. Now each slot is set up once and overwritten before each call. On the range query benchmark Prometheus already had, at 100 series and 1,000 steps, that's 52% fewer bytes and 89% fewer allocations. Thanks @bboreham, Julien Pivotto and @bwplotka for the review.

Post attachment