Quick reference
Untested backups
- A copy that never restored is a hypothesis
- The dashboard is green, the restore is not
- One drill a month is the cheapest test
Unbounded logs
- Retention by size, not by hope
- One chatty service can fill a disk
- The archive is a second bill
Cold data in hot storage
- Standard bills every idle byte
- Requests per GB decides the class
- Move it, then delete the source
Orphaned volumes
- An unattached disk bills the same
- An idle IP bills by the hour
- A snapshot with no owner is forever
Note: Each leak is invisible until the invoice arrives.
Explained
Untested backups
A copy that never restored is a hypothesis. The job writes a file, the dashboard turns green, and the first real test is the incident. The bill is not the problem here; the recovery time is.
Unbounded logs
Retention by size, not by hope. One chatty service fills a disk, the archive becomes a second bill, and the useful line is buried in a terabyte. A policy with a number is the fix.
Cold data in hot storage
Standard bills every idle byte. A bucket with a few requests per GiB is paying the hot price for cold data, and the invoice never names the bucket. Move it, then delete the source copy.
Orphaned volumes
An unattached disk bills the same as a busy one. An idle address bills by the hour, a snapshot with no owner lives for ever, and nobody sees the line because it belongs to the account, not the team.
The honest caveat: a leak is only a leak if nobody owns it. Some cold data is hot for a week, and some orphans are disaster-recovery copies that must stay.
The lesson: data costs money after it stops being useful, not while it is.
In one line
Nothing leaks faster than storage. Four leaks that never show up as a line item with an owner.
Count the orphans
python3 orphan_resources.py inv.jsonMore notes: Cloud: AWS & Azure
Open in the notes app: search, filters, copy This note on GitHub