This document describes the storage of the package registry and how it can be managed.
## Deduplication
The package registry has a build-in deduplication of uploaded blobs.
If two identical files are uploaded only one blob is saved on the filesystem.
This ensures no space is wasted for duplicated files.
If two packages are uploaded with identical files, both packages will display the same size but on the filesystem they require only half of the size.
Whenever a package gets deleted only the references to the underlaying blobs are removed.
The blobs get not removed at this moment, so they still require space on the filesystem.
When a new package gets uploaded the existing blobs may get referenced again.
These unreferenced blobs get deleted by a [clean up job]({{< relref "doc/advanced/config-cheat-sheet.en-us.md#cron---cleanup-expired-packages-croncleanup_packages" >}}).
The config setting `OLDER_THAN` configures how long unreferenced blobs are kept before they get deleted.
## Cleanup Rules
Package registries can become large over time without cleanup.
It's recommended to delete unnecessary packages and set up cleanup rules to automatically manage the package registry usage.
Every package owner (user or organization) manages the cleanup rules which are applied to their packages.
| Type | Every rule manages a specific package type. |
| Apply pattern to full package name | If enabled, the patterns below are applied to the full package name (`package/version`). Otherwise only the version (`version`) is used. |
| Keep the most recent | How many versions to _always_ keep for each package. |
| Keep versions matching | The regex pattern that determines which versions to keep. An empty pattern keeps no version while `.+` keeps all versions. The container registry will always keep the `latest` version even if not configured. |
| Remove versions older than | Remove only versions older than the selected days. |
| Remove versions matching | The regex pattern that determines which versions to remove. An empty pattern or `.+` leads to the removal of every package if no other setting tells otherwise. |
| `release.*` | Match versions that are either named or start with `release`. |
| `.+-temp-.+` | Match versions that contain `-temp-`. |
| `v.+\|release` | Match versions that either start with `v` or are named `release`. |
| `package/v.+\|other/release` | Match versions of the package `package` that start with `v` or the version `release` of the package `other`. This needs the setting _Apply pattern to full package name_ enabled. |
The cleanup rules are part of the [clean up job]({{< relref "doc/advanced/config-cheat-sheet.en-us.md#cron---cleanup-expired-packages-croncleanup_packages" >}}) and run periodically.
The cleanup rule:
1. Collects all packages of the package type for the owners registry.