After updating the secret restart controller or delete the pod, *Kubernetes will automatically create a new one (assuming it's managed by a Deployment, StatefulSet, or similar controller), and the new pod will pick up the updated secret value.

Kubectl change secret with kubectl patch

A shell script to manually authenticate and fetch the manifest for a private container image from the GitHub Container Registry (GHCR).

This script is useful for debugging authentication issues with GHCR or for programmatically inspecting image metadata (like layers and digests) without needing to perform a full docker pull.

How It Works
It performs the standard two-step authentication flow required by Docker V2-compatible registries:

Request a Bearer Token: It first authenticates against the GHCR token endpoint (https://ghcr.io/token) using a GitHub Personal Access Token (PAT). It requests a temporary, scoped token with pull access to the specific repository.
Fetch the Manifest: It then uses the temporary bearer token to make an authenticated GET request to the registry's manifest API endpoint to retrieve the image manifest.

Fetch Private GHCR Image Manifest

Ubuntu VM ReverseProxy with SSL

Azure Service Bus (Producer-Consumer SDK)

Jupyter Notebook - List all installed packages

Two popular and reliable options are Google Public DNS and Cloudflare DNS.

Query DNS Server

PR Comment Boilerplate

# Demo: how rate-limits shape throughput and latency.

- MockElevenLabs class with two fake endpoints:
- clone_voice() (simulates 0.8-10 s work)
- tts() (simulates 0.2-0.4 s work)

Give each endpoint limits that look like the real ones
- Voice-clone: only 1 job may run at a time, 1 request per second.
- TTS: up to 4 jobs in parallel and 5 requests per second.

How those limits are enforced
- “Token bucket” → counts how many calls have happened in the last second (RPS).
- “Semaphore” → counts how many calls are running right now (concurrency).
- If either bucket is full, the co-routine just await asyncio.sleep(0.05) and keeps trying until a slot opens.

Fire a bunch of users at it
- simulate_batch() starts 2 or 5 concurrent co-routines that call either clone_voice or tts.
- run_all() does this for both endpoints, prints how long the whole group had to wait.

Why the times are what they are
- Voice-clone requests are forced to run one after another, so 5 users need roughly 5 × (0.8-10 s).
- TTS requests can overlap (max 4 at once) and start 5 per second, so they finish much faster.

TLDR: throttle layer + a quick load test that shows how rate-limits shape throughput and latency.

Throughput and latency controls by rate-limits

The Identity Provider (IdP): This entity is responsible for authenticating a user and then "issuing" a signed JWT. It holds the private key.
The Service Provider (SP): This is our simulated service. It needs to verify incoming JWTs to grant access. It holds the corresponding public key.

Prepare the project (Node.js):
```sh
mkdir jwt-simulation
cd jwt-simulation
npm init -y
npm install jsonwebtoken
```

Identity Provider (IdP) x  Service Provider (SP)

Complete table of all current and recent Amazon EKS Kubernetes versions, including their standard and extended support end dates, as listed in the official AWS documentation.
https://docs.aws.amazon.com/eks/latest/userguide/kubernetes-versions.html

AWS EKS Support Periods and Versions

 It gathers information on screen dimensions, browser viewport size, and the device's pixel ratio, which is crucial for understanding how things are rendered on high-DPI (or "Retina") displays.



Console Script to Gather Display Information

Script to create a sorted version of a CSV in case column date is on this format: DD/MM/YYYY

Sort CSV by DD/MM/YYYY column

 Immediately Invoked Function Expression (IIFE)
--
The script finds all images within elements having the class .masonry-brick, changes their URLs to point to what is presumed to be a much larger version of the same image, and then copies the complete list of these new URLs to the clipboard, with each URL on a new line. This is a typical automation script one might run in a browser's developer console to quickly grab high-resolution image links from a gallery-style webpage.

Scrap all images from a page (IIFE)

tRPC is designed to provide end-to-end type safety by leveraging TypeScript's type inference, allowing developers to avoid manually writing types for API contracts. This approach is a core part of its developer experience and a key reason for its popularity in the TypeScript community.
How tRPC Handles Types
Type Inference by Default: tRPC automatically infers input and output types from your procedures and schemas (often defined with libraries like Zod), so you rarely need to explicitly annotate types in most of your codebase.

No Extra Schemas: Unlike tools like OpenAPI or GraphQL, tRPC does not require separate schema definitions or code generation steps. Types flow directly from your backend procedures to your frontend consumers.

Colocated Types: Types are often defined alongside the functions that use them, and can be exported for use elsewhere if needed, but explicit type definitions are minimized unless you want to share types across boundaries (e.g., separate repos).

tRPC

To optimize AI interaction with your codebase, .windsurfrules files should employ XML-like tags (e.g., <tag_name>) to group related rules, ensuring clarity, precision, and scalability. This structured approach is vital as rule sets expand.

Key categories to encapsulate within these tags include <clarityAndSpecificity> for unambiguous rule definition, <projectContext> for details on tech stack, style, and structure, and <codingStandards> for referencing style guides and organization. Additionally, <frameworkAndLibraryUsage> should specify required tools, <safeEditingAndRestrictions> should list non-modifiable elements and restricted APIs, and <maintainability> should guide the ongoing updating of rules in line with project evolution and global Windsurf standards.

.windsurfrules

About

A comprehensive guide to evaluating multi-class classification performance for AI agents with hierarchical categorization systems.

Multi-Class Classification Evaluation for AI Agents

Learn how to use JSON Schema and a central registry to manage data contracts in a microservices architecture.

Schema-First: JSON Schemas for Microservice Data Contracts

Learn how to scale your n8n workflows from a single Docker container to a resilient, multi-replica Kubernetes setup using PostgreSQL and queue mode.

Scaling n8n: From Local Docker to Robust Kubernetes Deployments

Explore the AHA Stack (Astro, HTMX, Alpine.js) – a powerful trio for building fast, server-rendered web applications with minimal client-side JavaScript.

The AHA Stack: MPA / Web Development, Simplified

Explore typical career paths for Software Engineers, from foundational roles to technical leadership and management tracks. Includes a visual guide and key considerations for growth.

Software Engineer Career Paths

Learn how to automate the conversion of Figma designs into production-ready HTML, CSS, and React Native code using an AI assistant like Claude and a specialized Figma MCP server from Composio.

From Figma to Code: AI-Powered UI Generation with MCP

Quickly learn to automate web tasks like form filling using Claude and Playwright Model Context Protocol (MCP).

AI-Powered Automation for Mission-Critical Systems

What We Build

99%+ Critical Accuracy Automations

Legacy to AI Transformation

Precision AI Agentic Systems

Insights from our AI Lab

Multi-Class Classification Evaluation for AI Agents

Schema-First: JSON Schemas for Microservice Data Contracts

Scaling n8n: From Local Docker to Robust Kubernetes Deployments

Let's Build your Future!