Automation

You have a list of files. You need to process each one. The naive approach: 1 2 3 for file in $(cat files.txt); do process "$file" done This works until it doesn’t — filenames with spaces break it, and it’s sequential. Enter xargs. The Basics xargs reads input and converts it into arguments for a command: 1 2 3 4 5 # Delete files listed in a file cat files.txt | xargs rm # Same thing, more efficient xargs rm < files.txt Without xargs, you’d need a loop. With xargs, one line. ...

Code review catches problems. CI catches problems. But the fastest feedback loop? Catching problems before you even commit. Git hooks run scripts at key points in your workflow. Use them to lint, test, and validate — automatically. Where Hooks Live 1 2 3 4 5 6 ls .git/hooks/ # applypatch-msg.sample pre-commit.sample # commit-msg.sample pre-push.sample # post-update.sample pre-rebase.sample # pre-applypatch.sample prepare-commit-msg.sample # pre-merge-commit.sample update.sample Remove .sample to activate. Hooks must be executable (chmod +x). ...

Cron is deceptively simple. Five fields, a command, done. Until your job runs twice simultaneously, silently fails for a week, or fills your disk with output nobody reads. Here’s how to write cron jobs that actually work in production. The Basics Done Right 1 2 3 4 5 6 7 8 # Bad: No logging, no error handling 0 * * * * /opt/scripts/backup.sh # Better: Redirect output, capture errors 0 * * * * /opt/scripts/backup.sh >> /var/log/backup.log 2>&1 # Best: Timestamped logging with chronic 0 * * * * chronic /opt/scripts/backup.sh chronic (from moreutils) only outputs when the command fails. Perfect for cron — silent success, loud failure. ...

Every ops engineer has a story about a shell script that worked perfectly — until it didn’t. Usually at 3 AM. Usually in production. These patterns won’t make your scripts bulletproof, but they’ll stop the most common failures. Start Every Script Right 1 2 3 #!/usr/bin/env bash set -euo pipefail IFS=$'\n\t' This preamble should be muscle memory: set -e: Exit on any error (non-zero return code) set -u: Exit on undefined variables set -o pipefail: Catch errors in pipes (not just the last command) IFS=$'\n\t': Safer word splitting (no spaces) Without these, a typo like rm -rf $UNSET_VAR/ could wipe your root filesystem. ...

If you’re running services on Linux, you’re almost certainly using systemd. But there’s a gap between knowing systemctl start nginx and actually writing your own robust service units. Let’s close that gap. The Anatomy of a Service Unit A systemd service unit lives in /etc/systemd/system/ and has three main sections: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 [Unit] Description=My Application Service After=network.target Wants=network-online.target [Service] Type=simple User=appuser Group=appgroup WorkingDirectory=/opt/myapp ExecStart=/opt/myapp/bin/server Restart=on-failure RestartSec=5 [Install] WantedBy=multi-user.target Let’s break down what matters: ...

The AWS Console is fine for exploration. For real work—auditing, automation, bulk operations—the CLI is essential. Here’s how to use it effectively. Output Formats 1 2 3 4 5 6 7 8 9 10 11 # JSON (default, best for scripting) aws ec2 describe-instances --output json # Table (human readable) aws ec2 describe-instances --output table # Text (tab-separated, grep-friendly) aws ec2 describe-instances --output text # YAML aws ec2 describe-instances --output yaml Set default in ~/.aws/config: ...

GitHub Actions tutorials show you on: push with a simple build. Real projects need caching, matrix builds, environment protection, secrets management, and reusable workflows. Here’s what actually works. Workflow Structure 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 # .github/workflows/ci.yml name: CI on: push: branches: [main] pull_request: branches: [main] jobs: test: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - name: Run tests run: npm test Caching Dependencies Without caching, every run downloads the internet: ...

Ansible is easy to start and hard to master. A simple playbook works great for 5 servers. The same playbook becomes unmaintainable at 50. Here are the patterns that keep Ansible codebases sane as they grow. Project Structure Start with a structure that scales: a ├ ├ │ │ │ │ │ │ │ │ │ ├ │ │ │ ├ │ │ │ └ n ─ ─ ─ ─ ─ s ─ ─ ─ ─ ─ i b a i ├ │ │ │ │ └ p ├ ├ └ r ├ ├ └ g └ l n n ─ ─ l ─ ─ ─ o ─ ─ ─ r ─ e s v ─ ─ a ─ ─ ─ l ─ ─ ─ o ─ / i e y e u b n p ├ └ s ├ └ b s w d s c n p p a ├ └ l t r ─ ─ t ─ ─ o i e a / o g o _ l ─ ─ e o o ─ ─ a ─ ─ o t b t m i s v l ─ ─ . r d g k e s a m n t a / c y u h g ├ └ i h g └ s . e b o x g r v v f / c o r ─ ─ n o r ─ y r a n r s a a g t s o ─ ─ g s o ─ m v s e / r u i t u / t u l e e s s l o s p a w s p a r s / . t n . _ l e . _ l s . y . y v l b y v l . y m y m a . s m a . y m l m l r y e l r y m l l s m r s m l / l v / l e r s . y m l The key insight: separate inventory per environment. Never mix production and staging in the same inventory file. ...

The biggest challenge with LLMs in production isn’t getting good responses—it’s getting parseable responses. When you need JSON for your pipeline, “Here’s the data you requested:” followed by markdown-wrapped output breaks everything. Here’s how to reliably extract structured data. The Problem 1 2 3 4 5 6 7 8 response = client.chat.completions.create( model="gpt-4", messages=[{"role": "user", "content": "Extract the person's name and age from: 'John Smith is 34 years old'"}] ) print(response.choices[0].message.content) # "The person's name is John Smith and their age is 34." # ... not what we needed You wanted {"name": "John Smith", "age": 34}. You got prose. ...

If you work with APIs, logs, or configuration files, you work with JSON. And if you work with JSON from the command line, jq is indispensable. Here are the patterns I use daily. The Basics 1 2 3 4 5 6 7 8 9 10 # Pretty print echo '{"name":"alice","age":30}' | jq '.' # Extract a field echo '{"name":"alice","age":30}' | jq '.name' # "alice" # Raw output (no quotes) echo '{"name":"alice","age":30}' | jq -r '.name' # alice The -r flag is your friend. Use it whenever you want the actual value, not a JSON string. ...