amedee.be - 🚀 Smarter CI with GitHub Actions Cache for Ansible (aka “Stop Downloading the Internet”)

Mood: Slightly annoyed at CI pipelines 🧨
CI runs shouldn’t feel like molasses. Here’s how I got Ansible to stop downloading the internet. You’re welcome.

Let’s get one thing straight: nobody likes waiting on CI.
Not you. Not me. Not even the coffee you brewed while waiting for Galaxy roles to install — again.

So I said “nope” and made it snappy. Enter: GitHub Actions Cache + Ansible + a generous helping of grit and retries.

🧙‍♂️ Why cache your Ansible Galaxy installs?

Because time is money, and your CI shouldn’t feel like it’s stuck in dial-up hell.
If you’ve ever screamed internally watching community.general get re-downloaded for the 73rd time this month — same, buddy, same.

The fix? Cache that madness. Save your roles and collections once, and reuse like a boss.

💾 The basics: caching 101

Here’s the money snippet:

path: .ansible/
key: ansible-deps-${{ hashFiles('requirements.yml') }}
restoreKeys: |
  ansible-deps-

🧠 Translation:

Store everything Ansible installs in .ansible/
Cache key changes when requirements.yml changes — nice and deterministic
If the exact match doesn’t exist, fall back to the latest vaguely-similar key

Result? Fast pipelines. Happy devs. Fewer rage-tweets.

🔁 Retry like you mean it

Let’s face it: ansible-galaxy has… moods.

Sometimes Galaxy API is down. Sometimes it’s just bored. So instead of throwing a tantrum, I taught it patience:

for i in {1..5}; do
  if ansible-galaxy install -vv -r requirements.yml; then
    break
  else
    echo "Galaxy is being dramatic. Retrying in $((i * 10)) seconds…" >&2
    sleep $((i * 10))
  fi
done

That’s five retries. With increasing delays.
💬 “You good now, Galaxy? You sure? Because I’ve got YAML to lint.”

⚠️ The catch (a.k.a. cache wars)

Here’s where things get spicy:

actions/cache only saves when a job finishes successfully.

So if two jobs try to save the exact same cache at the same time?
💥 Boom. Collision. One wins. The other walks away salty:

Unable to reserve cache with key ansible-deps-...,
another job may be creating this cache.

Rude.

🧊 Fix: preload the cache in a separate job

The solution is elegant:
Warm-up job. One that only does Galaxy installs and saves the cache. All your other jobs just consume it. Zero drama. Maximum speed. 💃

🪄 Tempted to symlink instead of copy?

Yeah, I thought about it too.
“But what if we symlink .ansible/ and skip the copy?”

Nah. Not worth the brainpower. Just cache the thing directly.
✅ It works. 🧼 It’s clean. 😌 You sleep better.

🧠 Pro tips

Use the hash of requirements.yml as your cache key. Trust me.
Add a fallback prefix like ansible-deps- so you’re never left cold.
Don’t overthink it. Let the cache work for you, not the other way around.

✨ TL;DR

✅ GitHub Actions cache = fast pipelines
✅ Smart keys based on requirements.yml = consistency
✅ Retry loops = less flakiness
✅ Preload job = no more cache collisions
❌ Re-downloading Galaxy junk every time = madness

🔥 Go forth and cache like a pro.

Got better tricks? Hit me up on Mastodon and show me your CI magic.
And remember: Friends don’t let friends wait on Galaxy.

💚 Peace, love, and fewer ansible-galaxy downloads.

🚀 Smarter CI with GitHub Actions Cache for Ansible (aka “Stop Downloading the Internet”)