Platformatic @platformatic

The platform bridging the gap between Node.js developers, operators, and wider business. So you can focus on building 🚀 platformaticHQ.com Joined April 2022

Tweets

1K
Followers

3K
Following

111
Likes

1K

Luca Maraschi @lucamaraschi

3 days ago

"Zero downtime deployment" means your server never went offline. It says nothing about the users already on your site when you shipped. They're running old JS bundles, old React state, against a server that's already moved on. Next API call hits an endpoint that no longer exists. Checkout fails. Form resets. Zero downtime. Broken session. Version skew isn't a bug. It's structural. It happens on every deploy. The fix isn't preventing it; it's building systems that let multiple versions run together. Cookies, ingress routing, replica set management, and a control plane that handles the rest. Building this with @matteocollina on Wednesday. 8am PT / 5pm CEST → streamyard.com/watch/RnVyfVTn…

0 2 2 157 0

View Details

Platformatic @platformatic

4 days ago

Cold start problem in agent infrastructure? The harness and agent are new processes on an already running pod. Pod capacity is managed by the scaler; it's already there. Startup time: instant. The restore time depends on the state size. @matteocollina 's honest about that...

0 1 3 357 0

View Details

Platformatic @platformatic

a week ago

@medusajs on K8s doesn't need four repos and four pipelines. One Watt monorepo: backend, storefront, admin, image optimization. One build. One deployment. Server-side calls stay inside the process. No network hop.

Luca Maraschi @lucamaraschi

a week ago

This is the default for most Medusa teams: backend, storefront, admin, and image optimizer split across separate repos. Four Docker builds. Four pipelines. Every internal API call burns time and resources crossing the network. 𝐁𝐮𝐭 𝐡𝐞𝐫𝐞’𝐬 𝐭𝐡𝐞 𝐫𝐞𝐚𝐥𝐢𝐭𝐲: 𝐧𝐨𝐧𝐞

0 2 4 299 0

0 0 3 181 0

View Details

Faris Aziz @farisaziz12

a week ago

We’ve had support from @CloudflareDev, @AWS, @stripe , @platformatic, @ImagekitIo, @storyblok, @vercel, GYFF, Smallpdf, @OnlyDust_com, @jetbrains , SwissDevJobs, Dynatrace, Digitec Galaxus, GetYourGuide, @sentry, Tiptap, Namespace, Stormkit, SAE Institute Zürich, and many more.

1 1 5 286 0

View Details

Platformatic @platformatic

a week ago

Old bundle in the browser. New server. One click away from a broken session. Version skew is already happening. Next Wed → Version skew vs API versioning → Faster deploys increase exposure → Skew protection lets you ship all day Wed 8am PT / 5pm CEST streamyard.com/watch/VG7m2CbS…

0 0 2 225 0

View Details

Yagiz Nizipli @yagiznizipli

2 weeks ago

I’ve deployed @platformatic Watt for our new loggedout page, excellent multithreading performance with observability. Have you tried multithreaded @nodejs yet?

3 5 58 7K 9

View Details

Platformatic @platformatic

2 weeks ago

gVisor: one policy per container. Platformatic: one policy per agent process. "Agent 1 has these policies. Agent X instance of type A has these." That's the difference between container-level and agent-process-level granularity. Episode: youtube.com/live/i0VuBKeLX…

0 0 2 269 1

View Details

Luca Maraschi @lucamaraschi

2 weeks ago

x.com/i/article/2064…

1 7 17 3K 10

View Details

Platformatic @platformatic

2 weeks ago

Median latency flatters every SSR framework. Only tail latency reveals the real story under production pressure. Back in March, we benchmarked @tan_stack, React Router, and Next.js across Node, PM2, and Watt at 1,000 req/s on AWS EKS. TanStack and React Router handle it cleanly.

Luca Maraschi @lucamaraschi

2 weeks ago

Most SSR benchmarks measure average latency. Average latency hides what matters in production. In March @platformatic ran @tan_stack, React Router, and Next.js across Node, PM2, and Watt at 1,000 req/s. Real findings: → TanStack + React Router: 100% success on all runtimes.

1 0 10 5K 2

1 1 13 4K 5

View Details

Platformatic @platformatic

2 weeks ago

A default set in 2015 silently cost Node.js up to 26% throughput. Fixed in Node.js 26.3.0. → Why the fix helps some workloads 26% → How @matteocollina traced it to the OS level → What it takes to prove a change like this is safe streamyard.com/watch/9y9Q4GVG…

0 1 5 661 1

View Details

Platformatic @platformatic

3 weeks ago

Kata Containers and gVisor draw the boundary at container level. We draw it at the process. Set isolation unit at agent process, not the whole machine. You get fine-grained control with Linux primitives K8s understands. cgroups. No extra tooling. youtube.com/live/i0VuBKeLX…

0 2 4 203 0

View Details

Platformatic @platformatic

3 weeks ago

Most ECS autoscaling is wrong for Node.js. Not the setup. The metric. → CPU-based: reacts to the symptom → ELU-based: reacts to the cause → Predictive ELU: acts before saturation 20ms vs 929ms. Under the same load. blog.platformatic.dev/aws-ecs-autosc…

0 0 2 126 0

View Details

Platformatic @platformatic

3 weeks ago

AWS ECS autoscaling watches CPU. Node.js chokes on event loop saturation, long before CPU spikes. Target Tracking: 929ms median, 74.76% success rate. ICC (ELU-based, predictive): 20ms median, 99.99%. Wrong metric. Wrong timescale. Wrong tool for Node.js. ⬇️ ⬇️ ⬇️ ⬇️ ⬇️

Luca Maraschi @lucamaraschi

3 weeks ago

The AWS ECS autoscaler doesn't know Node.js exists. It watches the CPU. Node.js doesn't choke on CPU; it chokes on event loop saturation. By the time CloudWatch fires, the event loop has been backed up for minutes. Target Tracking: 3 minutes before the alarm fires. Then

0 0 3 223 0

0 0 1 165 0

View Details

Platformatic @platformatic

3 weeks ago

Memory vs. performance. Usually gets detected too late. Next Wed @matteocollina @lucamaraschi : → Find the tradeoff at design time, not after → What flamegraphs show about Kafka that metrics miss → Debug memory. No guessing Wed 8 am PT / 5 pm CEST streamyard.com/watch/H56Fpjtq…

0 0 2 116 0

View Details

Platformatic @platformatic

4 weeks ago

AWS ECS autoscaling reacts after the problem starts. We benchmarked AWS ECS Target Tracking & Step Scaling against our ICC's predictive scaling. Results: Up to 97.8% lower p95 latency Up to 95.4% fewer scaling actions Better handling of traffic spikes blog.platformatic.dev/aws-ecs-autosc…

0 1 3 123 1

View Details

Platformatic @platformatic

4 weeks ago

Four problems. One missing primitive. → Can't require() a module from memory → Tenant sandboxing requires fragile path logic → AI-generated code must hit disk before import → SEA assets need 20–40MB of boilerplate @platformatic/vfs fixes all four. node:vfs

0 0 4 204 0

View Details

Platformatic @platformatic

4 weeks ago

platformatic/vfs hooks into both fs and the module resolver — in-memory files, tenant sandboxing, runtime code imports, SEA assets. All transparent to your dependencies. blog.platformatic.dev/why-nodejs-nee…

0 0 4 139 0

View Details

Platformatic @platformatic

a month ago

If your Node.js rate limiter lives in process memory, it breaks the moment you scale. @matteocollina & @lucamaraschi will dig down into Rate Limiting in Node.js! Next Wednesday, June 3rd, 8 am PT / 5 pm CEST → streamyard.com/watch/7VZi7cfP…

0 0 2 104 0

View Details

Platformatic @platformatic

a month ago

@matteocollina built a module that logs every AI agent mistake to a markdown file. New session starts → reads the top 30 past mistakes → doesn't repeat them. Not a vector store. A markdown file. He's running it daily. Full episode → youtube.com/live/OzemdgOno…

1 1 4 109 0

View Details

Platformatic @platformatic

a month ago

Node.js has 5 ways to call native C. Most teams know 2 → Native addon changes your build → Subprocess adds process boundary → WASM only if C compiles to it → Microservice adds network hop → node:ffi load .so, describe ABI, ~15ns. Nothing changes. blog.platformatic.dev/destino-doom-t…