Day 54 of 100
The Idle Resource Hunt
🔥 Problem Statement
Finance flags that cloud spend has grown steadily and asks for a cost-reduction effort. A quick look across the environment turns up what looks like a promising target: hundreds of resources, and a fair number of them look idle — low or no recent usage. The obvious next move is deleting the ones that look unused. The obvious next move is also the one most likely to break something important if done carelessly.
🏗️ Environment
- Environment: hundreds of cloud resources across several teams
- Tagging: partial, inconsistent ownership tags
- No current process for identifying and cleaning up idle resources
- Goal: reduce cost by removing genuinely unused resources
🤔 Your Challenge
How would you identify resources that are safe to delete, rather than just resources that look idle?
- What's the difference between a resource with low current usage and a resource that's genuinely safe to remove?
- Could a resource be idle most of the time but critical during a specific, infrequent event — a monthly batch job, a disaster recovery path?
- Why does inconsistent tagging make this harder, and what does it put at risk if ignored?
- What approval or verification step should happen before anything is actually deleted?
Solution Hidden
Think through the problem yourself before looking at the answer.
💡Solution
Step 1 — Understand the symptoms
"Looks idle" based on recent usage is a starting signal, not a safe-to-delete determination on its own. A resource can have genuinely low day-to-day usage and still be essential — a disaster recovery standby, a resource that only activates during a monthly or quarterly batch process, a dependency that something else references indirectly in a way that isn't obvious from the resource's own metrics. Usage data alone can't distinguish "unused" from "used rarely but critically."
Step 2 — Identify the likely bottleneck
Inconsistent tagging compounds this risk significantly — without reliable ownership tags, there's often no clear way to ask "does anyone still need this" of the actual person who'd know, which pushes toward either overly cautious (nothing gets cleaned up, cost stays high) or overly aggressive (things get deleted based on usage data alone, and something breaks) approaches, neither of which is what a careful cost-optimization effort should look like.
Step 3 — Investigation
- Build a full resource inventory with whatever ownership and tagging information does exist, explicitly flagging resources with missing or unclear ownership as a separate category needing extra care.
- Check usage patterns over a longer window than just "recent" — a full quarter or year, where feasible, specifically to catch infrequent-but-critical usage patterns (monthly reports, annual processes) that a short window would miss entirely.
- Check for dependencies — does anything reference this resource that wouldn't be obvious from its own usage metrics? Infrastructure-as-code references, DNS records pointing at it, or other resources configured to depend on it are all worth checking directly rather than assumed absent.
- Classify by environment — genuinely stale development or test resources are a much lower-risk cleanup target than anything touching production, even if production usage looks low.
Step 4 — Recommended action
For resources with clear ownership and confirmed-safe-to-remove status, proceed with cleanup. For resources with unclear ownership or ambiguous usage, don't delete based on inference alone — reach out to identify a real owner, or if none can be found, stop the resource (rather than delete it) for an observation period before final removal, giving anything unexpectedly dependent on it a chance to surface loudly rather than silently breaking.
Require explicit approval before deletion, especially for anything touching production or anything without clear, confirmed ownership — a lightweight sign-off step is far cheaper than restoring something deleted in error, particularly if backups or snapshots weren't retained.
Beyond this one cleanup effort, this is a strong case for enforcing tagging discipline going forward — required ownership and environment tags at resource-creation time (via policy, not convention) would make the next cost-optimization pass dramatically faster and safer.
Step 5 — Engineering lesson
Cost optimization done by deleting whatever looks idle, based on usage data alone, trades a real risk (breaking something critical but infrequently used) for a speculative benefit (marginal savings on resources that might not even be the largest cost drivers). The safer, more durable approach — ownership, dependency checking, an observation period before final deletion — takes longer but avoids turning a cost-saving initiative into its own incident.
🧠 Today’s Takeaway
Delete based on evidence, ownership and dependency—not simply age.
🔔 Don’t Miss Tomorrow’s Challenge
A new real-world engineering challenge is released every day.
100 Days → 100 Challenges → 100 New Things Learned.
Get the useful stuff, not the noise.
Occasional notes on engineering, Platform Engineering, AI, cloud and things I’m learning along the way.

