mirror of
https://github.com/hanzoai/docdb.git
synced 2026-10-30 18:39:39 +00:00
zeekay
2aa796ca9e
Both files were dark. This repo has .hanzo/workflows, so the forge reads that directory and only that one, and neither of these has produced a run since cicd.yml landed. They could not run on GitHub either — both ask for hanzo-build-linux-amd64, which no GitHub-hosted runner answers, so there they queue rather than fail. packages.yml was upstream FerretDB's release train: eval, development and production image variants pushed to Docker Hub, GHCR and Quay behind DOCKER_HUB_TOKEN and QUAY_TOKEN, plus .deb and .rpm builds. Every run it ever got on the forge failed. Nothing here consumes any of those artifacts. cleanup.yml ran a weekly docker-cleanup and `go clean -cache -modcache` on the runner. The fleet's runners are ephemeral containers — there is no cache that survives a job to reclaim. hanzo.yml's `test:` block is unchanged and remains the gate. Its comment no longer points at packages.yml as the image lane, because there is no image lane here now — said plainly in the file rather than left implied. Capability dropped: multi-registry image publication and Linux packaging. Neither had worked.
39 lines
2.1 KiB
YAML
39 lines
2.1 KiB
YAML
# Canonical CI/CD config for hanzoai/docdb — read by the hanzoai/ci reusable
|
|
# (.hanzo/workflows/cicd.yml) and by platform.hanzo.ai.
|
|
#
|
|
# Test-only caller, and there is now no image lane in this repo at all.
|
|
# packages.yml used to claim that role — the repo's own Taskfile driving
|
|
# bin/task docker-local, QEMU, definedockertag and a push to Docker Hub, GHCR
|
|
# and Quay — but it never once succeeded: it was dark on the forge (this repo
|
|
# has .hanzo/workflows, which the forge reads instead), and every run it did
|
|
# get, it failed. It wanted DOCKER_HUB_TOKEN and QUAY_TOKEN, and it published
|
|
# eval/dev/prod variants plus .deb and .rpm packages, none of which anything
|
|
# here consumes.
|
|
#
|
|
# Declaring `images:` in its place would not restore it; it would invent a
|
|
# different artifact under the same name. What actually runs in the cluster is
|
|
# ghcr.io/hanzoai/docdb-sqlite, and it is not built from this tree. Whoever
|
|
# needs a docdb image from this repo should declare the one they want, once,
|
|
# right here.
|
|
#
|
|
# The gate is the repo's own `test-unit-short` (Taskfile.yml), minus the
|
|
# `tools` sub-task that needs bin/task on PATH:
|
|
# - the unshallow FIRST. build/version is generated by `git describe --dirty`,
|
|
# and this workflow checks out shallow with no tags, so generate panics with
|
|
# "No names found, cannot describe anything" and every version string lands
|
|
# as "unknown". packages.yml buys the same thing with `fetch-depth: 0` plus
|
|
# an explicit tag fetch (actions/checkout#290); this is that, from inside
|
|
# the gate, since the checkout above is not ours to configure.
|
|
# - `-tags=docdb_dev` is the repo's BUILD_TAGS. Without it the dev-only
|
|
# packages are excluded and their tests silently do not run.
|
|
# - `-short` skips the suites that need a live PostgreSQL/MongoDB (those are
|
|
# integration/, a separate module ./... does not reach anyway).
|
|
|
|
test:
|
|
- name: unit
|
|
run: |
|
|
set -euo pipefail
|
|
git fetch --tags --force --unshallow 2>/dev/null || git fetch --tags --force
|
|
go generate ./build/version
|
|
go test -short -tags=docdb_dev -count=1 ./...
|