gbp-archive

A Gentoo Build Publisher plugin to dump & restore builds.

Description

The gbp-archive package provides two subcommands for gbpcli: dump and restore. These subcommands are "server-only" meaning they are only available from the GBP server instance. The dump subcommand can dump all the builds on the instance, just a particular machine or machines, or a particular build or builds, or any combination of those.

Then restoring is as simple as running gbp restore:

"Gentoo" and the Gentoo logo are trademarks of the Gentoo Foundation. This project is not affiliated with or endorsed by the Gentoo Foundation.

Installation

This assumes you already have a working Gentoo Build Publisher installation. If not refer to the GBP Install Guide first.

Install the gbp-archive package onto the GBP instance.

cd /home/gbp
sudo -u gbp -H ./bin/pip install gbp-archive

There is no need to restart any services after installation.

Process

Dump

How the dump process works is a tar archive is created. Inside that tar archive are 3 items:

Metadata

The first item is a JSON file containing metadata for the dump. Currently the metadata are:

• The version number of the dump file
• A timestamp for when the dump was created
• The hostname of the GBP instance that created the dump
• The list of builds included in the dump

Records

The second item is a file which includes all the "records" for the dumped builds. This is the build metadata stored the GBP database. This information is serialized into JSON and stored in a single file, records.json.

Storage

The third item is another tar archive consisting of the "storage" for the dumped builds. By storage I mean all of the repos, binpkgs, and configuration for builds. All of them are stored in an inner tar archive called storage.tar. The reason for the single archive and not one archive per build is to preserve the multiple hard links which span across builds (for a given machine). They would not be preserved across multiple archives.

Restore

For the restore process, we open the outer tar archive and then the records.json file is deserialized and loaded into the instance's database. Then we extract the contents of storage.tar to the root of the instance's storage root. Currently the restore process is all-or-nothing. But in the future I will add the ability to filter out what builds get restored.