Release pipeline for OpenVox packages. Downloads artifacts from S3, signs them (RPM, DEB, MSI, DMG), builds Apt/Yum repo metadata, and deploys to S3. Repo metadata lives in git for audit and rollback.
- Reference: environment variables, platform filter format, tasks, container, architecture
- Running locally: local equivalents of every GitHub Actions workflow
- Secrets: generating secrets, GitHub Actions configuration, local secrets file
- GCS backup setup: bucket creation, Workload Identity Federation, cost estimation
- Bootstrap: initial migration from previous repo tooling (historical)
Example: release openvox-agent 8.26.0.
- Go to Actions > Reset Test Repos > Run workflow. This syncs the test repos from production so you're testing against a clean baseline.
- Go to Actions > Release > Run workflow.
- Project:
openvox-agent - Version:
8.26.0 - Leave
productionunchecked.
- Project:
- The workflow performs:
backup->release->deploy->verify. For the test run, no commit is created. - If desired, manually verify the test repos (install a package, check metadata).
- Re-run the Release workflow with the same project and version, this time checking
production. The workflow performs:backup->release->deploy-> pushstate/to main ->verify.
Options:
- Component: Defaults to
openvox8. Set toopenvox9(or other component) if releasing for a different repo component. - Platform filter: Enter a comma-separated list of platforms in vanagon format (e.g.
el-9-x86_64,debian-13-amd64) to release only specific platforms. Leave blank for all. - Force overwrite: Check this to overwrite packages that already exist in the repos (e.g. when re-signing). Do not do this unless you really have to, because people's package managers will complain about the packages having a different hash.
Note
When doing the very first release of packages into a new component that doesn't yet exist in the repo (e.g. openvox9, openvox8-nightlies), this automation will work to create the relevant paths in the repo, but you should not make a noarch package (e.g. openvox-server or openvoxdb) the first released package. Because we fan out noarch packages into each architecture-specific location, you should release a package like openvox-agent first to populate all of these relevant architecture paths.
Check Security release on the Release workflow and fill in advisories and severity. The release then writes an advisory into the yum repo metadata, so dnf update --security, yum update --security and zypper patch --category security pick the packages up and dnf updateinfo info shows the references. apt has no equivalent, so Debian and Ubuntu users see nothing special.
- advisories: CVE IDs and GitHub advisory links, separated by spaces or commas, and nothing else. CVE IDs become references that
dnf update --cvematches, a GitHub advisory link becomes a reference with its GHSA ID, and the advisory's description names them all. Leave it empty while nothing is assigned yet. - severity: the severity of the worst item: Critical, Important, Moderate or Low. GitHub's critical, high, medium and low map to those in that order, as do CVSS scores of 9.0 and up, 7.0 to 8.9, 4.0 to 6.9, and lower.
To mark a release after the fact, or to add a CVE ID that was assigned later, go to Actions > Mark Security Release > Run workflow with the project, the version, the full list of references and the severity. It replaces the advisory of that release in every yum repo that contains its packages and deploys only the changed metadata, which clients pick up at their next metadata refresh. No package changes. Leave production unchecked to try it against the test repos first, then re-run with production checked.
The small noarch openvox*-release RPM/DEB packages register the OpenVox Apt/Yum repos on end-user systems. They drop .repo files, .list files, GPG public keys, and Apt pin preferences into the correct locations. Platform definitions live in files/repo_packages/*.json, with a file for each repo component (e.g. openvox8).
Example: add Fedora 44 and Ubuntu 26.04 to the openvox8 repos.
- Go to Actions > Add New Repo Package Platform > Run workflow.
- Component:
openvox8 - Platform(s):
fedora-44,ubuntu26.04 - Optionally check Build and upload to test and/or Build and upload to production to build immediately.
- Component:
- The workflow updates the platform definition in
files/repo_packages/openvox8.jsonand pushes to main. If build options were checked, it builds and uploads the packages.
Architecture suffixes are ignored (these are noarch packages). Hyphens are optional (el9 and el-9 both work). The task infers the package kind (RPM or DEB) from the OS name, normalizes the format, sorts the platforms list, and commits the change.
Rebuild and upload repo packages for all or specific platforms. Generally only needed when something fundamental changes (e.g. a new GPG key or repository URL).
- Go to Actions > Build Repo Packages > Run workflow.
- Leave
productionunchecked to build packages pointing at the test repos and upload to the test buckets. - Optionally enter specific platforms (e.g.
fedora-44,sles-16). Leave blank for all.
- Leave
- Verify the test packages work as expected.
- Re-run with
productionchecked for production.
The workflow will refuse to overwrite packages that already exist in S3 unless Overwrite existing artifacts in S3 is checked.
Roll back the repo metadata to a previous release state. The workflow performs: rollback -> deploy -> push state/ to main (only if production is checked) -> cleanup. This creates a new commit rather than reverting or resetting, so history is preserved.
- Find the commit to roll back to:
git log --oneline state/ - Go to Actions > Rollback > Run workflow.
- Enter the commit SHA.
- Leave cleanup checked to remove orphaned packages.
- Leave
productionunchecked to test first.
- After verifying the test repos, re-run with
productionchecked.
If the S3 repos are corrupted or accidentally deleted and a rollback can't fix it, restore from the GCS backup. This is a destructive operation that overwrites the current S3 repos with the backup contents.
- Go to Actions > Disaster Recovery (Restore from GCS) > Run workflow.
- Choose a target (
all,apt,yum, ordownloads). Defaults toall. - Leave
productionblank to restore to the test buckets first. Optionally check Use the production GCS backup as source to test with production backup data. - If the current GCS backup is also bad, enter a timestamp between the good and bad backups (RFC 3339 format with timezone, e.g.
2026-04-15T13:30:00-07:00). This works within the 90-day soft-delete window. - Manually verify the test repos.
- Re-run, typing
I want to restore to productionin theproductionfield. - If the restored state is older than the current
state/directory, follow up with a rollback to the matching commit.