I lived on GitHub Pages for static hosting for a long time. For example, this blog been hosted as https://mprokopov.github.io for years, just with the custom domain https://prokopov.me

But last week I finally moved everything to CloudFlare Pages and quite happy about it!

The major driver for the change was a hassle of management github-pages. It’s either your have to keep everything in a separate branch, with the name like github-pages, or have to use completly different repository. I chose the latter, and for a long time kept using two repositories, one with Hugo sources, second with rendered HTML. It was a bit painful to manage two git repositories in the same working tree, hugo build renders the website to the public folder. In addition, the repository with HTML should be public.

We all love the fast software. Ghostty is crazy fast.

What I use with my macos keyboard comes from Emacs. Namely, Cmd+ instead of Control+ for navigation in shell. Jump to a beginning of line, +a, to the end of line +e.

Why?

1. Because my macOS keyboard doesn’t have TWO Ctrl keys on both sides.
2. Try to compare ctrl-a vs cmd-a yourself 😀

Here is an example of a config extracted from this discussion.

Example of the Python code to catch EC2 termination signal and ping Jenkins to run termination job. Useful to execute graceful shutdown if you’re using EC2 spot instaces.

You will need Jenkins Generic Webhook plugin.

import json
import boto3
import urllib3
import os

JENKINS_JOB_TOKEN =  os.environ.get('JENKINS_TOKEN')
JENKINS_URL =  os.environ.get('JENKINS_URL')

def invoke_jenkins(target):
    http = urllib3.PoolManager()
    return http.request('POST', f"{JENKINS_URL}/generic-webhook-trigger/invoke?instance={target}", headers={'Authorization': f"Bearer {JENKINS_JOB_TOKEN}"})

def find_target(instance_id):
    ec2 = boto3.client("ec2")
    tags = ec2.describe_tags(Filters=[{'Name': 'resource-id', 'Values': [instance_id]}])

    for tag in tags['Tags']:
     if tag['Key'] == "Name":
        return tag['Value']

def lambda_handler(event, context):
    instance_id = event["detail"]["instance-id"]

    target = find_target(instance_id)

    if not target:
        return {
            'statusCode': 422,
            'body': f"target for {instance_id} not found"
        }

    print(f"Jenkins termination job called for instance: {instance_id} target {target}")

    resp = invoke_jenkins(target)

    print(f"Jenkins response status {resp.status}")
    print(f"Jenkins decoded response")
    print(resp.data.decode('utf-8'))

    return {
        'statusCode': resp.status,
        'body': json.dumps(resp.data.decode('utf-8'))
    }

HashiCorp Nomad Job

It’s often the job increases retry counter, but there is no time to waste after the root cause of failing job being removed.

Here is the way to enforce re-run of the job. Unfortunately, I couldn’t find any means to achieve the same from the UI

nomad job eval -force-reschedule rundeck

The problem

The problem of interpreter and dependency management is quite common, but also challenging.

1. Appliations sometimes are written using interpreted languages. Examples: brew in Ruby, ansible in Python.
2. Software development. For instance, mainstream AI development requires Python.

In order to run they always need two things:

• Interpreter
• Dependencies

This boils down to the following problems. How to download and keep interpreters of different versions? How to download, keep and manage dependencies? How to include or not to include the interpreter and dependencies into a distribution?

It’s common task to prepare README.me, but how to check the preview?

If you use uv, the fastest way is to use grip like this

Use this oneliner in the folder with markdown and open browser with http://localhost:6419 address

uv run --with grip -m grip

which is the same as

uv tool run grip

or even shorter

uvx grip

Learn how to access and work with your private data using Large Language Models with Model Control Protocol.

Cloud Cost Management, or FinOps is tricky. Imagine Team A started consuming much more resources and the management just starts seeing hair-pulling consequences at the end of the month in AWS bill.

How to react fast and prevent unnecessary cost? Or become very restrictive by blocking spinning up a new resources w/o explicit management approval, this can slow down and progress and also can be catastrophic for the company.

• Bumped Backstage dependencies with migration to the new backend system.
• Okta authentication (with wildcard sign-in)
• Building a roadmap for an engineering team
• Troubleshooting Unifi WiFi connectivity issues for the network with 12 access points and 200+ roaming clients.

Hardware

• Bought Meta Quest 3 Games: Underdogs, Assassin Creed Nexus, Batman Arkham.

Playground

• Model Control Protocol Server for the own ledger system
• Developed several Python modules with UV package manager.
• Tested V0 generation.
• Tested NextJS docker container.
• Tested blog with NextJS and Strapi.
• Played with Svelte and SvelteKit.

Using sattelites for internet communications is not a genuinely new idea. Number of companies provided such services before Starlink.

But the key difference between previous generation solutions is the latency.

The problem is, 600ms for a signal to travel almost 35700km to a sattelite and back is a lot. No gaming or realtime communication is feasible with such a delay.

The only way to decrease the lattency is to position sattelites at lower altitude. This is why Starlink sattelites fly between 340km and 1200km above the surface. Giving us incredible 30ms latency even in remotest regions of Earth.